← Knowledge Base 🕐 116 min read

Knowledge Base

Workflow Redesign

· March 2026

The single most predictive variable separating AI high performers from the rest of the corpus. Tools deployed without...

The single most predictive variable separating AI high performers from the rest of the corpus. Tools deployed without workflow redesign produce Stage 1 outcomes (−12.6 pp growth vs. industry average per MIT CISR, n=721). Tools paired with redesign produce Stage 3+ outcomes (+11.3 to +17.1 pp growth).

IBM IBV “Finance Execution Unlocks AI Value at Scale” (n=1,025, May 2026) — The 8%-to-18% Gap

IBM IBV’s most recent finance-specific survey quantifies exactly how much workflow redesign is worth in dollar terms:

Experienced AI adopters in finance: median 8% reduction in total finance costs
Finance organizations with AI embedded end-to-end: 18% cost reduction — more than double
The gap is not explained by tool selection or model capability. IBM’s analysis attributes it entirely to execution maturity: redesigned decision points, standardized inputs, reduced handoffs, defined action triggers
69% of CFOs call AI integral to their strategy; IBM’s framing of “from experimentation to execution — why sequencing matters” implies the majority are still in the experimentation phase despite high conviction

This is the first study in the corpus to put a specific dollar-value range on the workflow-redesign premium in the finance function specifically. It corroborates the McKinsey (21% redesign rate), WEF/Accenture (~15%), and MIT CISR (Stage 2→Stage 3 performance jump) findings with a finance-function-specific sample.

Source: research/04-consulting-firms/ibm-ibv-finance-execution-ai-scale-2026.md — MEDIUM / TIER 1

IBM IBV “From AI Projects to Profits” (n=2,500, Jun 2025) — The Core-Function Penalty

The 31%→7% ROI decline as GenAI pilots scaled is not a story of AI disappointment — it is the predictable cost of deploying AI on core functions without first redesigning the workflow. Early high-ROI pilots ran on peripheral, low-coordination tasks (chatbots, email summarization, code autocomplete). Core-function deployment requires data interoperability, cross-functional governance, and redesigned decision rights that most organizations have not built.
Fewer than a quarter of organizations are reimagining workflows with AI at the center — consistent with the Accenture (21%), McKinsey (21%), and MIT CISR findings across independent samples.
The top decile (18% ROI) has completed this redesign. The average (7%) has not.
64% of AI budgets now target core functions: organizations are making the right strategic bet; most just haven’t done the workflow architecture work that makes it pay off.

Source: research/12-agent-workers/ibm-ibv-ai-projects-to-profits-2025.md — MEDIUM-HIGH / TIER 2

HCLTech AI Impact Imperatives 2026 (n=467 G2K leaders, May 2026) — Legacy Systems as the Workflow Ceiling

51% of enterprise applications at G2K organizations are legacy systems — and AI deployed on top of fragmented legacy architecture cannot deliver the workflow integration that generates measurable returns.
Organizations deploying AI without first addressing legacy application constraints encounter the same failure pattern documented in the corpus: 43% of major AI initiatives expected to fail, with execution gaps (not technology) as the primary cause.
The 49% workflow integration rate (AI in existing workflows) aligns with the corpus-wide finding that fewer than half of AI investments are past the point where workflow redesign has occurred.

Source: research/07-adoption-challenges/hcltech-ai-impact-imperatives-2026.md — MEDIUM / TIER 1

Microsoft New Future of Work 2025 — The Workshopping Mechanism

Accenture “From Early Impact to Enduring Advantage” (n=3,650 C-suite, Jan 2026) — The Process Codification Prerequisite

Only 21% of organizations report redesigning end-to-end processes with AI at the core — identical to the McKinsey finding (21% workflow redesign, n=1,491, Mar 2025), from an independent survey of a different population. The convergence across two large, independent institutional sources strengthens both numbers.
The binding constraint is not AI capability but process opacity: most enterprise workflows rely on linear sequencing, batch handoffs, and rigid checkpoints. Decision rules and exception paths live in email threads and tacit knowledge — not systems. AI agents cannot automate what they cannot understand.
Regional bank example: implementation timeline slipped months before automation could begin, not due to model failure but because decision rules had never been documented. Codification is a prerequisite for any agentic deployment, not a nice-to-have.
~70% of technology budgets still support legacy systems, structurally starving investment in the modular architectures and workflow redesign that AI requires to compound value.

Source: research/04-consulting-firms/accenture-impact-to-advantage-2026.md — MEDIUM / TIER 1

Microsoft New Future of Work 2025 — The Workshopping Mechanism

The Microsoft Research 5th annual synthesis (MSR-TR-2025-58) names and quantifies the primary mechanism by which individual AI productivity gains fail to compound into organizational value: “workshopping” — AI-generated content that appears useful but lacks substance, passed between employees without verification.

40% of employees received AI-workshopped content from a colleague in the past month; estimated at 15% of total AI-generated content in circulation (Niederhoffer et al., 2025, HBR, n=1,150)
Workshopping flows: 40% peer-to-peer, 18% upward in hierarchy, 16% downward — corrupting both the inputs managers receive and the instructions teams execute
Workflow redesign that builds in explicit verification steps (not just “human in the loop”) is the structural fix; training on AI limitations reduces workshopping at the individual level but doesn’t eliminate organizational propagation without process change
This directly corroborates the Workday finding that 40% of AI time savings are consumed by rework, and the Atlassian 6% executive ROI confirmation rate

Source: research/07-adoption-challenges/microsoft-new-future-of-work-2025.md

AlixPartners Enterprise Software Predictions 2026 — Coding Productivity Paradox

AlixPartners’ analysis of 300+ software company client engagements documents the same workflow-redesign gap from the software vendor angle: AI coding tools deliver 20-30% productivity gains, but those gains do not convert to reduced R&D spending or faster product cycles. Teams absorb the freed capacity into low-priority peripheral work (“zombie ideas,” minor refactoring) rather than directing it toward the high-leverage bookends of the SDLC — product strategy (top of funnel) and launch/feedback loops (bottom of funnel). These are the stages with the lowest AI productivity uplift (23-25%) but the highest value creation potential. The AlixPartners framing: “the engine is more powerful, but the driver can’t yet handle the speed.” This is the same mechanism Goldman Sachs identifies when noting that task-level productivity gains don’t automatically translate to firm-level value without intentional reallocation.

Source: research/07-adoption-challenges/alixpartners-enterprise-software-predictions-2026.md

Gensler Global Workplace Survey 2026 (n=16,400, 16 countries, March 2026)

AI power users (30% of office workforce) spend 37% of their workweek on solo work vs. 42% for late adopters — a 5pp difference that maps directly onto how much workflow redesign has already occurred in their roles.
The time reallocation shows the mechanism: AI absorbs routine cognitive load, and employees who have truly integrated AI redirect the freed time to learning (+50% vs. late adopters) and collaboration, not more solo screen time.
Organizations where workflow redesign hasn’t happened show up in this data as the 36% late-adopter cohort — same tools available, no behavioral change.
AI power users are 3x more likely to perceive their organization as highly innovative — which suggests that workflow integration, not tool access, is what drives organizational perception of AI value.

Source: research/07-adoption-challenges/gensler-global-workplace-survey-2026.md

Ju & Aral (MIT/Johns Hopkins) — When Workflow Redesign Is Absent, AI Substitutes Rather Than Augments (February 2026)

The largest human-AI teamwork RCT (n=2,234) shows what happens when workers use AI without deliberate workflow design: they delegate 58% of tasks to the AI (vs. 50% in human-human teams), make 62% fewer direct edits, and produce output that converges toward a homogenized mean. Volume increases 50%, but creative variance collapses and real-world outcomes (ad performance, CTR, CPC) show no statistically significant advantage.

The finding directly illustrates the “substitution vs. augmentation” distinction: without a designed workflow specifying what AI handles and what humans own, workers default to wholesale delegation.
Organizations that define AI as “handles the production task” rather than “augments my judgment” will get more output with less differentiation — the worst outcome for brand voice, client communications, or thought leadership.
The behavioral signature to watch: if workers are making significantly fewer direct edits after AI deployment, workflow redesign is needed. AI is writing; humans are approving. That is a different cognitive posture than AI assists and human edits.

Source: research/01-ai-native-landscape/ju-aral-collaborating-ai-agents-rct-2026.md

Why It Matters

McKinsey State of AI 2025 (n=1,993): 55% of high performers fundamentally redesigned workflows when deploying AI vs. 18% of others.
BCG AI at Work 2025 (n=10,600): 70% of total AI value resides in people, organization, and process design — not technology.
Stanford Enterprise AI Playbook 2026 (51 deployments): Organizations handing AI 80%+ of workflow (humans handle exceptions only) achieve 71% median productivity gain vs. 30% for human-primary deployments — same tool, different workflow architecture.
Faros AI (10,000+ developers): 98% more PRs after Copilot rollout, zero delivery throughput improvement. Speed evaporated because the bottleneck moved from coding to review and nobody redesigned the review step.

Accenture UK AI Study 2026 (Feb–Mar 2026, n=1,891 employees + 510 leaders, YouGov)

Only 1 in 10 organizations has successfully scaled AI into core operations; 9 in 10 are stuck below enterprise-wide impact.
Only 23% of employees report their team restructured a major process around AI — the rest got tools without redesign.
46% of executives say AI has delivered little P&L impact despite rising employee usage; individual efficiency ≠ organizational productivity.
Revenue uplift potential from AI is more than double the impact of cost savings — but requires workflow integration, not just tool deployment.

Source: research/07-adoption-challenges/accenture-uk-ai-productivity-scaling-2026.md

Stanford HAI AI Index 2026 (Apr 13, 2026)

88% organizational adoption vs. 5% capturing substantial financial gains (aggregated from BCG/McKinsey) — the adoption-return gap is the workflow-redesign gap.
Capability moved faster in 2025 than any prior year: SWE-bench 60%→~100%, OSWorld agents 12%→66%. The frontier outran workflow readiness.
Stanford’s framing: “a widening gap between what AI can do and how prepared we are to manage it” — an explicit capability-vs-readiness gap that maps to workflow redesign as the binding constraint.

Source: research/01-ai-native-landscape/stanford-ai-index-2026.md

Google Cloud AI Agent Trends 2026 — Digital Assembly Line Architecture

Research file: research/01-ai-native-landscape/google-cloud-ai-agent-trends-2026.md

88% of agentic early adopters report positive ROI from multi-agent workflow deployments — the highest adoption-to-outcome ratio in the 2026 vendor survey corpus. Source: Google Cloud/ROI of AI 2025, n=3,466; vendor-funded, apply caveat.
Danfoss (Danish industrial): 80% of routine business decisions automated; data analysis time reduced from 42 hours to real-time via multi-agent orchestration. This is the “digital assembly line” pattern — multiple specialist agents, human-in-the-loop for exceptions, shared context object between steps.
A2A + MCP protocol convergence removes the integration friction that blocked multi-agent workflow deployments before 2026. A2A (Google, Apache 2.0) handles agent-to-agent communication and capability discovery; MCP (Anthropic) handles agent-to-live-data-source connections. Together, they reduce the custom integration cost of multi-agent workflow redesign from six-figure engineering work to platform configuration.
The report’s “digital assembly line” framing aligns with BCG’s 10/20/70 rule and McKinsey State of Organizations “structure to flow” findings: the value is in the end-to-end workflow architecture, not the individual agent. Named failure mode: deploying agents as point tools on unchanged workflows captures the 10% algorithm value; orchestrated workflow redesign captures the 70%.

Source: research/01-ai-native-landscape/google-cloud-ai-agent-trends-2026.md

BetterUp / Stanford “Workslop” Study — The Quality Cost of Volume-Optimized Deployment (Sep 2025)

Research file: research/07-adoption-challenges/betterup-stanford-workslop-ai-quality-cost-2025.md

40% of U.S. desk workers received workslop (AI-generated content that appears polished but lacks substance) from a colleague in a given month. Stanford Social Media Lab / BetterUp, n=1,150, September 2025.
$186 per employee per month in salary-weighted cleanup time — $9 million annually for a 10,000-person organization. Scales to ~$900K for a 1,000-person company.
Workday/Hanover Research (n=3,200, Jan 2026) independently finds 40% of AI time savings consumed by rework — corroborating the same phenomenon from the recipient side.
The mechanism: deployment optimized for adoption volume (more output, more prompts, more documents) without quality standards redistributes judgment work from the sender to the recipient. Cleanup is invisible on adoption dashboards.
Organizational consequence: 53% of workslop recipients were annoyed, ~50% rated the sender as “less creative and reliable.” Workflow redesign that specifies output standards — not just use cases — prevents the trust erosion that stalls second-wave adoption.

Source: research/07-adoption-challenges/betterup-stanford-workslop-ai-quality-cost-2025.md

BCG “How Leaders Build an AI-First Cost Advantage” (Berthion/Brunelli/Catchlove/Goydan, Mar 26, 2026) — the 10/20/70 rule

BCG’s March 2026 CFO-cost-transformation piece names the split explicitly: in a typical AI implementation, only 10% of value comes from the algorithms, 20% from technology and data, and 70% from managing process change — mainly redesigning workstreams and processes end-to-end. This is the numeric spine of the workflow-redesign thesis.
Incremental overlay of AI on existing processes captures the 10% algorithm gains and part of the 20% technology gains. End-to-end workflow reinvention captures the 70%. BCG’s published multiplier: reinventing processes generates 3–4x the impact of traditional incremental improvements.
The “why-it-fails” diagnostic in the piece names workflow-redesign underinvestment as one of five recurring failure modes behind the 60% of companies reporting minimal or no AI value. Companies without solid data foundations compensate by keeping “many people in the loop to check outcomes, eroding value” — the operational expression of the workflow-redesign gap.
BCG’s CFO sequencing prescription: start with proven-deployment workflows (procurement 5–25% savings in 3–6 months, spec-review 5–10%, inventory 5–15%), bank those savings inside a transformation budget, and use them to fund the Year Two end-to-end redesign. This is workflow redesign as a financing strategy, not a process-improvement project.
Named leader differential: AI leaders deliver 3x greater cost reduction, 1.6x higher EBIT margins, and 2.7x return on invested capital vs. peers. BCG-defined cohort — apply vendor caveat — but consistent with McKinsey Manifesto’s 20% EBITDA uplift finding on a different leading-company sample.

Source: research/04-consulting-firms/bcg-ai-first-cost-advantage-2026.md

McKinsey State of Organizations 2026 — “From Structure to Flow” (Maor/Krivkovich/Srinivasan et al., Feb 19, 2026, n>10,000)

The largest executive sample in the corpus (>10,000 senior executives across 15 countries and 16 industries) lands the same answer BCG’s 10/20/70 rule lands from the CFO side: the 2026 productivity unlock is workflow redesign, not automation.
38% of leaders say redefining process flows is the biggest productivity unlock over the next 1–2 years — more than any other lever. Two-thirds describe their organizations as “overly complex and inefficient.”
Traditional productivity approaches — restructuring, delayering, downsizing, cost reductions — show diminishing returns. Automating a bad workflow locks in the complexity.
Four specific flow-redesign levers McKinsey names: (1) simplify workflows and decision routines first; (2) reduce handoffs and duplication; (3) eliminate unnecessary meetings; (4) clarify decision rights and streamline approval chains. All four are delegable to a process-excellence function without waiting on AI deployment decisions.
Shared services as the first full test: 84% of organizations plan shared-services expansion within 1–2 years; only 6% currently realize full value from advanced technology in those centers. The GBS rebuild is where most mid-market companies will run the end-to-end human-AI orchestration pattern first.
McKinsey vendor caveat applied; triangulates with BCG AI-First Cost Advantage (same 70% workflow-value finding) and McKinsey March 2025 State of AI (workflow redesign = #1 EBIT predictor out of 25 attributes).

Source: research/04-consulting-firms/mckinsey-state-of-organizations-2026.md

BCG AI Radar 2025–2026 (n=1,803 / n=2,360)

Full research file: research/04-consulting-firms/bcg-ai-radar-2025-2026.md

94% of CEOs will continue investing regardless of near-term payoff (BCG AI Radar 2026, n=2,360). Conditional AI budgeting has ended.
0.8%→1.7% of revenue — AI spending as a share of company revenue doubled in a single year across all nine surveyed industries. Investment is accelerating while workflow readiness lags.
>30% of 2026 AI budgets committed to agentic AI at the CEO level; Trailblazers (top ~15%) direct ~60%. Agentic deployment requires workflow redesign by definition — agents cannot operate autonomously in workflows that have not been redesigned for machine execution.
60% of organizations track no financial AI KPIs. The workflow-redesign deficit and the KPI deficit travel together — organizations that have not redesigned workflows have nothing specific to measure.
2.1x ROI from focused vs. broad deployment: leaders prioritize 3.5 use cases (vs. 6.1 for laggards) and embed them end-to-end. Breadth without redesign produces laggard returns.
72% of CEOs are now the main AI decision maker (2x YoY, n=640) — the shift from CIO-led to CEO-led is the structural precondition for end-to-end workflow redesign rather than siloed pilots.
Only ~15% of CEOs (“Trailblazers”) apply AI end-to-end. 58% of Trailblazers agree end-to-end transformation is their biggest opportunity for agentic AI in the next 12 months vs. 30% of Followers — a ~2x gap.

Source: research/04-consulting-firms/bcg-ai-radar-2025-2026.md — TIER 1, MEDIUM-HIGH, n=1,803 (2025) / n=2,360 (2026)

McKinsey “Rewiring” — Workflow Redesign as the Dominant EBIT Predictor (March 2025, n=1,491)

The dedicated research file at research/04-consulting-firms/mckinsey-rewiring-2026.md covers this finding in full. Core data for this page:

Workflow redesign ranks #1 out of 25 organizational attributes tested in McKinsey’s relative-weights regression analysis (R²=0.20) for predicting enterprise-level EBIT impact from gen AI — outranking technology investment, talent strategy, and governance structure.
Only 21% of gen AI users have fundamentally redesigned at least some workflows. The remaining 79% are deploying AI onto unchanged processes.
>80% report no tangible enterprise-wide EBIT impact. Only 17% attribute ≥5% of EBIT to gen AI use (July 2024, n≈1,200).
The November 2025 companion survey (n=1,993) confirmed via a separate 31-variable analysis that workflow redesign is “one of the strongest contributions to achieving meaningful business impact of all factors tested” — high performers are nearly 3x more likely to have fundamentally redesigned workflows.
The technology-first anti-pattern: Organizations delegating AI to IT and pursuing a use-case-by-use-case approach accumulate tooling without structural change. The EBIT evidence supports the practitioner critique directly.

Source: research/04-consulting-firms/mckinsey-rewiring-2026.md — TIER 1–2, n=1,491, 101 nations, July 2024 data

McKinsey State of AI — March 2025 (n=1,491, Jul 2024 survey)

Out of 25 organizational attributes tested, workflow redesign has the biggest effect on EBIT impact from gen AI — the single strongest predictor across all company sizes.
Only 21% of organizations using gen AI have fundamentally redesigned any workflows. The remaining 79% are layering gen AI onto existing processes.
CEO oversight of AI governance is the attribute most correlated with EBIT impact at large companies ($500M+ revenue). Only 28% have CEO-level governance ownership.
Only 1% of developed-market executives describe their gen AI rollout as “mature.” Fewer than 1 in 5 track KPIs for gen AI — the adoption practice with the highest EBIT correlation.
Temporal note: This is the March 2025 edition (Tier 2). The November 2025 edition (n=1,993) updates adoption to 88% and narrows high performers to 6%. The workflow-redesign granularity in this edition is not replicated in the later survey.

Source: research/01-ai-native-landscape/mckinsey-state-of-ai-march-2025.md

Workday “Beyond Productivity” (Hanover Research, Jan 2026, n=3,200)

The largest primary survey on the rework cost of underpowered AI deployment. The finding: 40% of AI time savings are lost to rework — correcting errors, rewriting drafts, verifying outputs. Against the 85% of employees who report saving 1–7 hours/week, the net is approximately 0.6–4.2 hours — and only 14% of employees consistently achieve clear positive net outcomes.

The role-redesign gap is the mechanism: 89% of organizations updated fewer than half their roles to reflect AI-augmented work. Without role redesign, employees use AI as a drafting assistant for content a human expert must verify with the same judgment they would have applied to creating it. The rework is not accidental — it is the predictable outcome of deploying a tool into an unchanged workflow.

The training differential makes the mechanism precise: among employees with consistently positive AI outcomes, 79% received increased skills training. Among heavy AI users generally, only 37% received that training. The 40% rework tax is a training and role-redesign deficit, not a technology limitation.

Corroborates METR (19% slower experienced developers), McKinsey 6% high performers, and Writer 48% disappointment — all from different angles on the same underlying gap.

Source: research/07-adoption-challenges/workday-beyond-productivity-ai-rework-2026.md

ActivTrak State of the Workplace 2026 — Work Density as the Mechanism (Mar 2026, n=163,638)

The most operationally significant finding in the report is a before/after comparison (376 companies, 10,584 users, 180-day window): after AI adoption, every measured work category increased — email +104%, chat +145%, business management tools +94%. No category declined. Focus session length dropped 9% (14m 23s → 13m 7s) while collaboration surged 34%.

This is the workflow-redesign gap made visible at behavioral scale. AI generates output quickly, but that output requires review, correction, and coordination. Without workflow redesign that absorbs AI outputs into a structured review process, organizations get more communication load and less focused work — not freed capacity.

57% of users spend less than 1% of work hours in AI tools — technically adopted, not integrated
Only 3% reach the 7–10% usage window where 95% productivity rates are observed
50% of organizations using AI do not measure workforce productivity impact

The mechanism corroborates Workday/Hanover Research (40% of AI time savings consumed by rework, n=3,200) and METR (19% slower experienced developers on complex tasks). ActivTrak’s behavioral data makes the mechanism visible: AI adoption without workflow redesign increases throughput and communication load while degrading focus time — the opposite of the expected outcome.

Source: research/07-adoption-challenges/activtrak-state-of-workplace-2026.md

Workflow Selection Methodology and ROI Definition Framework (Multi-source synthesis, Apr 2026)

The corpus proves workflow redesign matters but historically provided no methodology for which workflows to select. This synthesis fills that gap.

Where to look first (BCG, n=1,250, Sep 2025): 70% of total AI value concentrates in core business functions — R&D/innovation (15%), digital marketing/customer journey (17%), manufacturing (9%), sales (7%), maintenance (6%), supply chain (6%). Support functions (HR, finance, legal, procurement) account for only 30%. Start where the value is.

How to score any workflow — five diagnostic criteria:

Process standardization — does it run the same way every time?
Handoff complexity — 1–2 clean handoffs vs. 4+ with information loss?
Data coherence — single system of record vs. Excel supplementing the ERP?
Decision-point density — clear criteria vs. judgment calls requiring “just ask Sarah”?
Workflow debt load — designed intentionally vs. accumulated workarounds since 2019?

Score 0–5: overlay candidate. Score 6–10: rebuild before deploying AI. Score 11–15: rebuild required — overlay will automate dysfunction.

ROI definition before deployment is equally decisive: Pertama Partners (2,400+ initiatives): 54% success rate with pre-defined financial metrics vs. 12% without. Three required parameters: (1) empirical baseline — 30–60 days of measured current cost per cycle; (2) counterfactual definition — control unit or explicit adjustment for volume/seasonal changes; (3) time-gated kill criteria — adoption >25% at 90 days, unit cost declining at 6 months, documentable P&L impact at 12 months.

2026 measurement shift (Futurum, n=830, Feb 2026): Financial ROI (revenue + profitability) nearly doubled YoY to 21.7% as primary metric; productivity fell 5.8 points. Business cases built on “hours saved” will not survive a 2026 CFO review.

Source: research/07-adoption-challenges/ai-workflow-selection-roi-definition-framework.md

IT Team AI Enablement: The Automation-First Workflow Sequence (2026)

The mid-market IT team must redesign its own workflows before it can enable AI workflows for others. This is the counterintuitive sequencing insight most CIOs miss — the same workflow redesign logic that applies to revenue functions applies internally.

60–80% of IT budget consumed by operational maintenance (Gartner benchmark) leaves no capacity for AI enablement without first automating the operational floor.
54% of I&O leaders are already adopting AI to cut operational costs (Gartner, n=253, May–Jul 2025) — creating the capacity that funds the IT team’s own transformation.
The AI redeployment map: Helpdesk → AI tool administrator; Systems admin → data readiness lead; Network engineer → AIOps operator; IT PM → AI program coordinator. Each role shifts from technology management toward workflow enablement.
Why outsourcing fails here: AI governance requires institutional context about the company’s risk tolerance, client relationships, and competitive dynamics that an MSP cannot hold. Outsource the operational floor; own the AI enablement function.

Source: research/07-adoption-challenges/it-team-evolution-ai-mandate-shift.md

Supporting Research

File	Angle
research/07-adoption-challenges/ai-workflow-selection-roi-definition-framework.md	Workflow selection criteria + pre-deployment ROI definition
research/07-adoption-challenges/ai-workflow-redesign-overlay-vs-rebuild.md	Overlay vs. rebuild decision framework
research/07-adoption-challenges/workflow-redesign-before-deployment.md	Sequencing: redesign first, deploy second
research/07-adoption-challenges/workflow-redesign-facilitation-skillset.md	Skill profile of the facilitator
research/01-ai-native-landscape/unified-ai-maturity-framework-synthesis.md	Cross-framework synthesis: why redesign drives stage progression
research/07-adoption-challenges/stanford-enterprise-ai-playbook-2026.md	51-deployment evidence on automation intensity
research/07-adoption-challenges/bcg-widening-ai-value-gap-2025.md	Companies that redesign capture the value
research/07-adoption-challenges/ai-job-redesign-methodology.md	Job-level redesign mechanics
research/07-adoption-challenges/ai-reaccumulation-problem-why-subtracted-processes-grow-back.md	Why redesigned processes regress
research/09-ai-adoption-cycle/mid-market-ai-strategy-document.md	79% claim strategy but only 37% well-formulated; 42% abandoned majority of AI initiatives in 2025
research/07-adoption-challenges/ceo-ai-not-delivering-diagnostic.md	CEO diagnostic: workflow bypass as #1 root cause of mid-market AI disappointment; 21% redesign rate vs. 6% EBIT capture
research/07-adoption-challenges/performative-compliance-diagnostic.md	Four-layer diagnostic distinguishing genuine workflow adoption from performative compliance; 56% of CEOs report zero ROI despite high usage dashboards
research/07-adoption-challenges/how-to-pick-your-first-ai-tool.md	Tool selection process: define the problem before naming a tool; 54% success with pre-defined metrics vs. 12% without
research/01-ai-native-landscape/hbs-cybernetic-teammate-rct-2025.md	Functional silos erased: generalists with AI match specialist-quality output, weakening the case for rigid functional boundaries (n=776 P&G professionals, pre-registered RCT)
research/01-ai-native-landscape/academic-ai-productivity-papers.md	Eight RCTs synthesized: productivity gains collapse to near-zero on complex brownfield tasks; “tools without redesign” is the most common failure mode across all six studies
research/01-ai-native-landscape/statcan-ai-productivity-capability-maturity-2026.md	Statistics Canada SDTIU: raw 16.8% productivity premium collapses to 5.1% (statistically insignificant) once complementary capabilities (data analytics + robotics) are controlled — capability stack precedes AI, not the reverse
research/07-adoption-challenges/epoch-ai-ipsos-ai-workplace-usage-2026.md	Epoch AI / Ipsos probability-based national survey (n=2,021, April 2026): task displacement outpacing task creation (27% vs. 21%); workflow redesign required to convert time savings into organizational capacity

Stanford SALT Lab WORKBank: H3 Partnership as the Design Target for Workflow Redesign (February 2026)

The WORKBank database (1,500 workers, 52 AI experts, 844 tasks, 104 occupations) provides the clearest empirical grounding yet for the claim that AI workflow design must start with worker-desired agency, not capability maximization.

45.2% of occupations have H3 (equal partnership) as the dominant worker-desired collaboration level — more than any other level. The workflow redesign implication: most knowledge-worker workflows should be designed for human-AI co-contribution, not AI-primary execution with human review. The default assumption in many enterprise deployments is H1 or H2 (AI does the work, human approves). WORKBank data suggests that is inverted for the majority of occupational contexts.
Workers prefer more human involvement than AI experts deem necessary in 47.5% of tasks. This gap persists even after workers are explicitly asked to consider job loss risk and task enjoyment. The implication for workflow redesign: building in H2 or H1 automation without surfacing this preference gap produces the workaround behavior (users reverting to manual steps, shadow workflows, selective non-use) that registers as “adoption failure” in dashboards.
The mismatch with current AI investment patterns: 41% of Y Combinator company-task mappings fall in the Low Priority and “Red Light” zones (high capability but low worker desire). The same pattern appears inside organizations — AI programs targeting impressive technical capability in low-desire areas face the steepest adoption friction. Workflow redesign that starts from the desire-capability landscape rather than the capability-only landscape has a structurally higher adoption success rate.
The Green Light zone is under-addressed: Tasks where both desire and capability are high — primarily administrative burden, data entry, scheduling, routine reporting — represent the fastest path to accepted workflow automation. These are not venture-capital-attractive, but they are where worker buy-in comes without the friction cost.

Source: research/01-ai-native-landscape/stanford-salt-lab-future-work-ai-agents-2026.md · Stanford SALT Lab / Brynjolfsson et al., arXiv:2506.06576v3, February 2026 · HIGH · TIER 2

NBER Shifting Work Patterns (Dillon et al., May 2025, n=7,137)

The sharpest evidence that tools alone do not redesign work: 6 months of Copilot access in a randomized cross-industry field experiment moved zero of the coordination-dependent metrics (meeting time, recurring meetings, meeting types, documents completed).
The only metrics that moved were unilateral: email time (-31% for regular users), concentration blocks (+4 hours/week), out-of-hours email time.
Authors frame coordination change as “broad institutional efforts, not just local team coordination” — i.e. redesign is a leadership act, not an emergent team behavior.
Firm fixed effects explain far more adoption variance than peer density, industry, or individual pre-experiment behavior — implying managerial practices and training are the effective intervention points.

Source: research/07-adoption-challenges/ai-team-coordination-evidence.md

McKinsey “Seizing the Agentic AI Advantage” (Mar 2025)

Quantifies the redesign premium directly: 5–10% time savings when gen AI assists humans in existing workflows, 30–50% when agents automate specific steps, 60–90% when the workflow itself is redesigned around agent capabilities (call-center resolution example).
~78% of firms use gen AI; ~80% report no significant P&L impact — the paradox is the redesign gap.
~90% of vertical (function-specific) agent use cases remain stuck in pilot; fewer than 10% reach production. Failure mode is organizational (scattered efforts, AI/IT silos, data quality, change resistance), not technical.
Four-pillar framework: people, governance, technology, data. Data is the pillar that fails first in most mid-market rollouts because no one owns making it AI-ready before the agent is pointed at it.

Source: research/01-ai-native-landscape/mckinsey-seizing-agentic-ai-advantage-2026.md

MIT SMR — Compound Benefits (Kiron & Schrage, Apr 2026)

Names the mechanism that compounds AI returns inside a redesigned workflow: a three-step cycle of verification (did the output meet the standard), evaluation (what did it reveal), and learning capture (how is the insight made reusable).
Organizations with systematic human-AI feedback loops are 6x more likely to report substantial financial benefits; those investing in learning with AI are 73% more likely to report significant financial impact.
Only 15% of AI-adopting companies use the technology for organizational learning — consistent with the 5%/6% “high performer” share in BCG and McKinsey 2025/2026 data.
Capture-layer examples: Anthropic’s CLAUDE.md files live inside the workflow, not after it; shared prompt repositories function as “version control for organizational judgment.”
Executive dashboard implication: measure cycle metrics (interactions verified, evaluated, captured, speed of practice change), not consumption metrics (licenses, hours saved).

Source: research/07-adoption-challenges/mit-smr-compound-benefits-generative-ai-2026.md

Deloitte + Docusign “Capitalizing on AI: How Automated Agreement Workflows Drive ROI” (Apr 16, 2026)

First primary-survey ROI benchmark in the corpus for contract-lifecycle workflow redesign. n=1,100+ senior leaders, six countries.
~30% higher ROI for organizations running agentic workflows inside end-to-end agreement platforms vs. peers using fragmented AI point tools — the premium attaches to platform consolidation + agentic orchestration, not to AI in general.
Adoption gap is precise: 65% of organizations run four or more separate agreement tools; 61% still extract post-signature insights manually. Fragmentation — not lack of AI — is what separates the 30% ROI cohort from the rest.
Cross-industry average bundle: 36% efficiency gains, 36% cost avoidance from risk mitigation, 29% labor-cost savings, 72% agreement-accuracy improvement. The four outcomes move together — isolated gains on one dimension are a signal the redesign did not happen.
Department-specific redesign gains that map onto McKinsey’s 60–90% workflow-redesigned band: legal 37% time reclaimed (one team scaled from 100–200 contracts/year to 1,000); sales 43% time savings, 29% fewer deal delays, 1–2% revenue uplift ($4.8M on 300 renewals × $670k baseline); procurement 33% vendor spend reduction; HR 45% time savings; customer experience 39% more completed agreements.
Vendor caveat applies heavily: Deloitte + Docusign are co-publishers; “agentic workflows within end-to-end platforms” closely mirrors Docusign’s own product positioning. Self-reported survey data, not RCT. Direction is credible; magnitudes on the department figures should be tested against a narrow instrumented pilot before extrapolation.

Source: research/04-consulting-firms/deloitte-docusign-agreement-management-roi-2026.md

Gartner: Assistive AI Abandonment Prediction (Apr 2, 2026)

Gartner predicts >50% of enterprises will stop paying for assistive AI (copilots, smart advisors) by 2028, favoring platforms that commit to workflow results via delegated execution.
By 2030, software companies layering bolt-on AI over legacy apps face up to 80% margin compression — the strongest analyst signal that the copilot layer is a dead end.
The distinction: assistive AI keeps humans in the execution loop; outcome-focused AI removes them from execution while keeping them in supervision (“Agent Steward” role).
First disruption targets approval-heavy, timing-sensitive workflows where AI collapses decision latency and reallocates authority to policy-bound agents.
Vendor evaluation reframed: not “does this tool have AI?” but “does this platform control identity, permissions, policy, and system-of-record access to act autonomously within guardrails?”

Source: research/05-analyst-firms/gartner-assistive-ai-abandonment-2026.md

Gartner AI Maturity Longevity Study (Jun 2025, n=432) — Operational Survival as the Workflow Redesign Signal

High-maturity organizations keep AI projects in production for 3+ years at 2.25x the rate of low-maturity peers (45% vs. 20%). Operational longevity — not launch count — is the differentiating metric; it reflects whether workflow redesign happened before or after deployment.
Organizations that invest 4x more (as a share of revenue) in data quality, governance, AI-ready talent, and change management achieve up to 65% better business outcomes. The investment differential is the proxy for whether workflow redesign was treated as Phase 1 work or deferred.
The business-unit readiness gap quantifies the workflow-redesign shortfall directly: 57% of high-maturity organizations report business units are ready and trust AI; only 14% of low-maturity organizations report the same. Workflow redesign is what moves organizations from 14% to 57% business-unit readiness.

Source: research/05-analyst-firms/gartner-ai-maturity-enterprise-2025.md — MEDIUM-HIGH / TIER 1 (n=432, Jun 2025)

BCG “AI at Work 2025” — The Adoption-to-Value Gap (Jun 2025, n=10,635)

72% of workers use AI regularly, but only 5% of organizations achieve substantial financial gains — the largest documented gap between adoption and value in any major enterprise AI dataset.
The redesign split is structural: organizations that “Reshape” workflows around AI capabilities outperform those that deploy AI into existing workflows. Half of companies have moved to Reshape; half have not.
Frontline adoption stalled at 51% and has not grown since 2023 — a 27pp gap vs. manager adoption (78%) that training and workflow redesign (not tool access) closes.
Shadow AI is a workflow signal, not just a governance risk: 54% of employees use unauthorized AI tools regardless of policy, revealing that approved workflows are not meeting the actual work demand.
Only 13% of organizations have AI agents integrated into workflows, yet 75% recognize them as critical for future operations — the workflow redesign required for agent deployment has not happened yet for the majority.

Source: research/01-ai-native-landscape/bcg-ai-at-work-2025.md

Gartner CSO Survey 2026 — The Sales Reinvestment Gap

The starkest single-function proof that workflow redesign, not tool deployment, determines AI ROI. Gartner surveyed 210 CSOs and senior sales leaders (Jan–Feb 2026):

AI saves the average seller 4.8 hours/week — but 72% of sales organizations fail to redirect those hours into high-value selling activities.
Organizations that actively redesign workflows to capture those hours are 2.2x more likely to exceed customer growth goals and 3.1x more likely to exceed lead-to-opportunity conversion targets.
The ROI distribution is bimodal: 25% report 50%+ positive return; 20% report 50%+ negative return. The dividing line is whether management defined a reinvestment target before deployment.
The mechanism is universal: time savings do not automatically compound into commercial performance. They require explicit workflow direction. This extends the Humlum & Vestergaard finding (85% of workers reallocate AI time savings to more of the same work) into the sales function.

Source: research/05-analyst-firms/gartner-cso-sales-ai-reinvestment-gap-2026.md

MHI + Deloitte 2026 Annual Industry Report (n=500, Supply Chain, Apr 2026)

41% of supply chain organizations currently use AI (up from 30% in 2025, +11pp YoY) — the fastest adoption jump in 13 years of the MHI survey.
Only 22% of organizations in adjacent research have completed major workflow redesigns (MIT CISR Digital Colleagues, Apr 2026), underscoring that the 41% using AI are predominantly at the tool-deployment stage, not the workflow-redesign stage.
Top barrier to scaling: “unclear use cases and difficulty constructing business cases” — the canonical workflow-redesign framing problem: organizations don’t know what workflow to redesign, so they stall after initial deployment.
Agentic AI cited as the next evolutionary stage in supply chain: agents that proactively address disruptions, eliminate repetitive tasks, and update forecasts continuously — all of which require end-to-end workflow redesign as a prerequisite to agent deployment.

Source: research/04-consulting-firms/mhi-deloitte-supply-chain-ai-2026.md

Rewired (Lamarre et al., 2024) — Adoption/Scaling Chapter as Supporting Evidence

Rewired places workflow redesign within two chapters that together form its strongest contribution to the adoption evidence:

Ch. 5 — “Reimagining Workflows with Agentic AI” (pp.79–97): Four levels of agentic automation, each with different redesign requirements: individual augmentation (tool overlay — minimal redesign), task automation (step removal — moderate redesign), agentic workflows (end-to-end restructuring — significant redesign), agentic systems (cross-process orchestration — full redesign). The key principle: “choosing the right tool for the job is not a sign of technological conservatism; it’s a mark of maturity” (p.92). Workflow selection criteria — dynamic, unstructured, edge-case-heavy work benefits from agents; deterministic, structured, high-volume work benefits from rules-based systems — directly parallels the corpus’s five-criterion workflow scoring framework.

Ch. 30 — “Make Adoption Stick” (p.457): The book’s explicit position is that adoption is engineered, not wished for. Treating agents “like new employees” — define the job, onboard carefully, evaluate continuously — is the paraphrasable principle. This aligns with what the MIT CISR FinCo case shows in its failure mode: governance without an adoption engineering mindset does not stick.

The sequencing tension: Rewired places adoption/scaling as Capability 5, after technology and data backbone. The independent corpus places workflow redesign as a Phase 1 constraint that must run parallel with — not after — data architecture investment. This is not a contradiction: Rewired acknowledges the parallel in Ch. 5’s placement (it appears in Section 1 — Strategy — not Section 5 — Adoption). Practitioners should read the chapter, not just the capability numbering.

Rewired’s quantified adoption anchor: E.ON Next (Ch. 5, pp.79–80) — +6pp CSAT, −8% average handle time, +14% transaction success rate, ~50% cost-per-call reduction. The workflow was redesigned from a human-primary call-handling model to an agent-primary model with human exception handling. This is the Stanford Enterprise AI Playbook’s 71% vs. 30% productivity split in a named production deployment: same tool stack, different workflow architecture.

Source: research/04-consulting-firms/mckinsey-rewired-2nd-edition-synthesis.md

Cross-Concept Links

Firm Size and AI Outcomes — small firms redesign faster but have less to redesign
Industry AI Outcomes — industries with the largest redesign cost see the longest payback

Workflow Redesign Practitioner Tools

30-Minute AI Workflow Readiness Assessment — Section 2 (Decision Architecture, 8 points) directly operationalizes workflow redesign criteria: decision clarity, volume threshold, process documentation, and whether the workflow has been redesigned (not just studied) before deployment. Source: research/09-ai-adoption-cycle/ai-workflow-readiness-30-minute-assessment.md
AI Deployment Red-Flag Checklist — Flag 1 (no activity-elimination mandate), Flag 2 (IT-only scope), and Flag 13 (AI before workflow redesign) are the three workflow-specific pre-deployment stop signals drawn from this evidence base. Source: research/09-ai-adoption-cycle/ai-deployment-failure-mode-red-flag-checklist.md
Rewired Transformation Roadmap Template — 12-month quarterly roadmap keyed to Rewired’s six-capability sequencing. Milestone 2.1 is the structured workflow redesign milestone: domain owner leads a task-level redesign defining what AI handles, what humans handle, and what is eliminated — before the pilot deploys. Grounded in McKinsey (n=1,993), MIT CISR (n=721), Stanford Enterprise AI Playbook (n=51). Source: research/09-ai-adoption-cycle/rewired-transformation-roadmap-template.md

Practitioner voices (pillar 13 — VentureBeat “Beyond the Pilot”)

Andrew Ng (DeepLearning.AI). Workflow redesign, not model selection, is the binding constraint on moving from single-model copilots to orchestrated agent systems. Frames agentic AI as “the next enterprise S-curve” whose value is gated on process re-engineering. Source: research/13-multimodal-sources/beyond-the-pilot/2026-04-13-from-models-to-agentic-systems-the-next-enterprise-ai-s-curv.md
JPMorgan Chase — Derek Waldron. “Businesses run on long processes that cross multiple different types of teams. If we want to be able to really move the needle on those processes, there has to be a strategic element to actually rethink what the process itself will need to look like in a world of AI and AI agents.” Process re-engineering is JPMorgan’s explicit top-down strategy, not task automation. Source: research/13-multimodal-sources/beyond-the-pilot/2026-04-13-what-30k-jpmorgan-ai-agents-taught-me.md
Enterprise agents are “slop” (Beyond the Pilot 2026-04-14). Practitioner panel names the mode of failure: agents deployed without workflow redesign generate volume without signal — the “slop” label. Same diagnosis McKinsey’s 6%-high-performer cut and BCG’s 5%-substantial-gains cut converge on from survey data. Source: research/13-multimodal-sources/beyond-the-pilot/2026-04-14-most-enterprise-ai-agents-are-slop-heres-why-they-fail.md
Intuit — Mariana Tessel (EVP/GM). 3 million customers using Intuit Intelligence agents for bookkeeping and payroll automation; 85% re-engagement rate, 30% reduction in manual work, 62% report bookkeeping is easier. The gains required redesigning the financial management workflow around agent execution, not bolting AI onto manual entry. Source: research/13-multimodal-sources/beyond-the-pilot/2026-04-01-100m-agents-scaling-the-new-execution-stack-with-intuit.md
Bain Technology Report 2025. AI coding tools deliver 10–15% productivity when applied to code generation alone; 25–30% when applied across the full development lifecycle. Writing and testing code is only 25–35% of time from idea to launch — the other 65–75% is the redesign opportunity. Three out of four companies cite “getting people to change how they work” as the hardest part. Source: research/04-consulting-firms/bain-ai-research-2026.md
Bain Commercial Excellence Survey (n=1,263, Jan 2025). Winners deploy 4.5 use cases vs. laggards’ 3.3 and capture nearly 2x cost efficiency per use case — driven by best-practice adoption (process redesign, data quality, governance), not technology selection. Source: research/04-consulting-firms/bain-ai-research-2026.md
Outshift by Cisco — Orit Goren. “Chatbots are a premature enterprise AI abstraction” — adding chat to existing systems, rather than embedding agents into the workflow, defers the redesign problem rather than solving it. Source: research/13-multimodal-sources/beyond-the-pilot/2026-04-13-why-chatbots-are-a-premature-enterprise-ai-abstraction.md
Dr. Sam Zolfagharian (Jaytech President) — AI For The C-Suite EP 54 (Apr 14, 2026). Top-down AI mandates without employee co-design produce low adoption — the workflow redesign fails at the human layer, not the technology layer. Specific failure mode: employees told tools are being deployed to replace them refuse to engage; “You’re bringing these AI tools to replace me. So I’m not going to do anything.” The redesign must include the answer to “What’s in it for employees?” or adoption stalls at the mandate level. HR exclusion is endemic: CEOs frequently haven’t consulted HR on AI rollouts, missing the incentive and L&D levers that drive sustained behavioral change. Source: research/13-multimodal-sources/ai-for-the-c-suite/2026-04-14-dr-sam-zolfagharian-how-leaders-should-actually-approach-ai-.md
Accenture Front-Runners Guide (n=2,000, May 2025). 8% of companies are “front-runners” scaling AI enterprise-wide; they invest 51% of tech budgets in cloud+AI and have 4x higher talent maturity than experimenters. Companies scaling single strategic bets are ~3x more likely to exceed ROI expectations — concentration on fewer, deeper bets outperforms breadth. Source: research/04-consulting-firms/accenture-ai-research-2026.md
BCG “AI Will Reshape More Jobs Than It Replaces” (2026). 50–55% of US jobs will be reshaped; only 10–15% eliminated. The six-segment model (amplified/rebalanced/divergent/substituted/enabled/limited-exposure) shows that reshaping outcome depends on two variables: whether AI augments or substitutes, and whether demand for the output expands. Cutting workforce beyond AI’s ability to replace it causes productivity to drop and institutional knowledge to disappear. Task turnover within a role proposed as a KPI for how fast roles evolve toward higher-value work. Source: research/01-ai-native-landscape/bcg-ai-reshaping-jobs-2026.md
PwC 29th CEO Survey (n=4,454, Jan 2026) + 2026 AI Predictions. 56% of CEOs see zero AI return — the clearest large-n validation that tool deployment without workflow redesign fails. PwC’s 20/80 rule: technology delivers ~20% of initiative value; 80% comes from redesigning work. The 12% vanguard achieving both revenue + cost gains deploy AI more extensively across business areas and have built foundations first. PwC recommends centralized “AI studios” over siloed experiments. Source: research/04-consulting-firms/pwc-ai-research-2026.md
PwC “2026 AI Performance Study” (n=1,217, 25 sectors, Apr 2026). 74% of AI’s economic value is captured by just 20% of organizations — the top 20% generate ~7.2x more AI-driven revenue and efficiency gains than the average organization. The workflow redesign behavior gap: AI leaders are 2x as likely to redesign workflows rather than layer AI on existing processes. The same leaders are 1.7x more likely to have a Responsible AI framework and 1.5x more likely to have a cross-functional AI governance board — governance and redesign correlate in the high-performing cohort, not trade off against each other. Source: research/04-consulting-firms/pwc-ai-performance-study-2026.md

Deloitte 2026 Global Human Capital Trends (Mar 4, 2026, n=9,000+)

Largest 2026 workforce survey (89 countries, Oxford Economics fieldwork): 59% of organizations take tech-first AI approaches. Tech-first cohort is 1.6x more likely to miss AI ROI expectations than the human-centric cohort that redesigns work before tools.
Names the arithmetic shift: human + machine (parallel deployment) → human × machine (intentional redesign). The “+” is the 59% default; the “×” is where the return difference lives.
Cost-efficiency framing identified as a trap: companies treating AI as headcount-reduction cap their upside; the human-centric cohort redirects freed capacity to growth, not elimination.
Static-plans → dynamic-orchestration: annual workforce plans are obsolete under AI-compressed cycles. 70% of leaders prioritize “fast and nimble” over three years as a direct response to S-curve compression.
Triangulates with BCG 5% substantial-gains and McKinsey 6% high-performer shares — the minority cohort in all three datasets made the same operating-design choice (redesign before deploy), not the same technology choice.

Source: research/07-adoption-challenges/deloitte-global-human-capital-trends-2026.md

ISG Index Q4 2025 / ISG Predictions 2026. 30% of well-funded AI use cases now in production (up from 15% a year ago). 77% of companies plan to increase AI spending in 2026. HFS Research’s GBS survey (n=510) shows enterprises shifting from KPIs to KPOs (key performance outcomes) — from cost-cutting to value creation, which requires workflow-level redesign, not just tool deployment. Source: research/04-consulting-firms/services-analysts-ai-2026.md
Orica (ServiceNow Knowledge 2025). Deflection rate jumped from 18% to 94% in 8 weeks — but only after rewriting knowledge base content and restructuring escalation paths to work with the AI. The tool (Now Assist) was the same; the workflow redesign produced the outcome. Source: research/13-multimodal-sources/servicenow-knowledge/servicenow-knowledge-2025-enterprise-ai-sessions.md
ServiceNow Enterprise AI Maturity Index 2025 (n=4,500, Oxford Economics). Maturity scores dropped 9 points YoY despite 55% of enterprises deploying 100+ AI use cases. Only 30% deploy across multiple functions — siloed AI proliferation without cross-functional workflow redesign. Source: research/13-multimodal-sources/servicenow-knowledge/servicenow-knowledge-2025-enterprise-ai-sessions.md
EY (ServiceNow Knowledge 2025). 12,000 AI actions/day across 400,000 employees; 70% resolution notes accepted without edits. The $2.3M annual value required 12-year platform consolidation — the workflow infrastructure preceded the AI deployment. Source: research/13-multimodal-sources/servicenow-knowledge/servicenow-knowledge-2025-enterprise-ai-sessions.md
Gartner AI Research 2026. 72% of CIOs report breaking even or losing money on AI (n=506, May 2025). GenAI in Trough of Disillusionment; average $1.9M per GenAI initiative with <30% CEO satisfaction. Gartner frames 2026 as the year AI is sold by incumbent software providers, not bought as moonshot projects — procurement replaces experimentation. Customer service canary: 91% face pressure to implement AI, but half of companies cutting staff will rehire by 2027 (bottleneck migration). Source: research/04-consulting-firms/gartner-ai-research-2026.md

AWS re:Invent 2024–2025 Customer Sessions

United Airlines rebuilt its delay messaging workflow end-to-end on Amazon Bedrock + Lambda + ElastiCache: processing time dropped from 5–15 minutes to 28 seconds, and the system now handles 100,000+ messages. The gain came from the workflow rebuild, not from the model.
United Airlines (second deployment): mainframe access time dropped from 30 minutes to seconds after a natural-language interface replaced manual queries; the tool scaled from 3,000 to 15,000 frontline employees — adoption followed the redesign.
MUFG rebuilt its corporate sales pitch process on Bedrock: sales preparation that previously took hours to days now takes 1–2 minutes. Result: 10x deal generation at a maintained 30% conversion rate across 2,000 employees and 1M corporate clients.
Capital One deployed GenAI workflows to 10,000+ customer service agents in production, with explicit observability guardrails and a gradual risk slope-up. The observability layer is load-bearing — it is what made executive sign-off on a 10,000-agent rollout possible.
Pattern across all three: workflow redesign, not tool bolting, is where the measurable value was created. None of the metrics came from inserting AI into an unchanged process.

Roland Berger “Profitless Prosperity in AI” — Rewire vs. Wrap (n=203, Mar 2026)

~90% of firms report AI returns lagging spending despite near-universal executive involvement. Roland Berger names two root causes of the value gap, both of which reduce to workflow/integration failures: (1) flying blind on measurement; (2) shallow integration.
The shallow integration mechanism: “Many organizations have bought the Ferrari of modern AI infrastructure, but are running it on the go-kart engine of shallow integration. Wrappers accelerate early progress, but collapse under operational load.” This is the first named strategy-consulting articulation of the wrapper-collapse pattern that Faros AI (98% more PRs, zero throughput gain) and Atlan (median +159.8% ROI only after workflow redesign) document empirically.
What Industrializers (the ~10% who capture value) do differently: They “rewire rather than wrap” — embedding AI directly into data pipelines, workflows, and governance. They do not spend more or move faster. Governance is built into platforms rather than enforced by committees. Innovation is federated through shared standards, not siloed projects.
Deployment is the start, not the end: Industrializers treat go-live as the beginning of the operational investment cycle, not the end of a project. This is the clearest consulting-firm statement of the “deployment ≠ operationalization” gap that PwC (74% of value to top 20%), McKinsey (55% of high performers redesigned workflows), and BCG (70% of value in people/process) all quantify.

Source: research/04-consulting-firms/roland-berger-profitless-prosperity-ai-2026.md · Mar 2026 · MEDIUM-HIGH · TIER 1

Source: research/13-multimodal-sources/aws-reinvent/aws-reinvent-2024-2025-enterprise-ai-sessions.md

Microsoft Ignite / Build 2024–2025

C.H. Robinson compressed quote response time from hours to 32 seconds on 2,000+ daily pricing requests — a workflow-specific deployment that targeted a single bounded process.
ABB Group reported 40% cost savings in operations/maintenance and 80% fewer service calls after AI-driven workflow redesign in manufacturing.
Eaton documented 9,000+ SOPs with 83% time savings per procedure — again, a targeted workflow, not a general productivity overlay.
Campari Group early adopters save 16–30 min/day and 86% report improved work quality, but this is self-reported and lacks before/after workflow measurement.
Pattern consistent with AWS re:Invent evidence: companies targeting specific workflows report 40–96% time compression; companies reporting general productivity cite 15–30 minutes/day with unclear revenue impact.

Source: research/13-multimodal-sources/microsoft-ignite-build/microsoft-ignite-build-2024-2025-enterprise-ai-sessions.md

Google Cloud Next 2025–2026

RTL Deutschland rebuilt its entire video asset production pipeline around Vertex AI + Veo, generating 150,000–200,000 campaign-ready assets daily — a workflow replacement, not a tool addition. Cloud elasticity allows scaling to 6M assets/week when demand spikes.
Publicis Sapient stated on-stage: “fixing the glass layer without touching the core systems gives you immediate results but not long-term longevity.” This echoes the McKinsey finding (n=1,993) that 55% of high performers fundamentally redesigned workflows vs. 18% of others.
KPMG achieved 75% workforce adoption of Gemini Enterprise within 48 hours — preceded by a deep architecture review and security tiger team, not a mandated rollout. The pull-through adoption model maps onto Colgate and Citi patterns.
Pattern holds: quantified metrics came from companies that replaced workflows (RTL Deutschland, Quickplay); qualitative claims came from companies that added tools to existing processes (Berenberg Bank, Fisher & Paykel).

Source: research/13-multimodal-sources/google-cloud-next/google-cloud-next-2025-2026-enterprise-ai-sessions.md

Salesforce Dreamforce 2025

PepsiCo reported 25–30% efficiency gains after redesigning store-level field service delivery around Agentforce across 1.5M stores (targeting 5M by end 2026). Athina Kanioura, EVP/Chief Strategy Officer, presented on-stage — this is a workflow replacement at massive retail scale, not a chatbot overlay.
Engine rebuilt customer support around autonomous resolution: 50% of cases handled by agents, 15% handle time reduction, $2M savings. Deployed in 12 days — speed came from constraining to a single workflow.
Grupo Falabella scaled WhatsApp support 5x (40K→216K conversations/month) with 60% autonomous resolution. Before redesign, customers routinely abandoned WhatsApp for the call center; the workflow change eliminated the handoff.
Safari365 (35 employees) exceeded its 15% efficiency target, achieving 30%+ — CEO built agents personally in 6 weeks. Smallest company in the vendor-conference corpus with quantified workflow redesign outcomes.
Pattern consistent across all four vendor conferences: companies that redesigned workflows around the agent (PepsiCo, Engine, Grupo Falabella) produced hard metrics; companies that added agents to existing processes (Pandora, reMarkable, Endress+Hauser) produced qualitative testimonials only.

Source: research/13-multimodal-sources/salesforce-dreamforce/salesforce-dreamforce-2025-enterprise-ai-sessions.md

Databricks Data+AI Summit 2025

Adidas redesigned agentic workflow for 2M product reviews/year: 91.67% LLM cost savings, 60% latency reduction, 20% productivity increase across 500+ decision-makers in 150+ countries. The gains came from optimizing the data pipeline underneath agents, not the models.
Walmart cut time-to-value by 90% and saved $5.6M/year with AI/BI Genie — workflow redesign around self-service analytics rather than adding AI to existing reporting chains.
United Airlines reduced operational costs 50% by redesigning real-time data analysis from batch to streaming (20M+ rows in <10 min).
Standard Chartered achieved 80% faster incident detection, 92% faster threat investigation, 60% better accuracy by replacing traditional SIEM with AI-driven security — a full workflow replacement, not augmentation.
Pattern consistent across six vendor conferences now: data platform modernization + workflow redesign = hard metrics; bolt-on AI = qualitative testimonials.

Source: research/13-multimodal-sources/databricks-data-ai-summit/databricks-data-ai-summit-2025-enterprise-ai-sessions.md

McKinsey AI Transformation Manifesto (Apr 2026)

Adoption fails when adjacent upstream and downstream processes stay unchanged — McKinsey’s example: AI predicts equipment failures days in advance but maintenance still follows calendar-based scheduling, so nothing happens. Process redesign is the missing link, not the AI model.
Across 20 AI-leading companies, concentrating on 1–3 business domains with full workflow reinvention delivered 20% EBITDA uplift and $3 incremental EBITDA per $1 invested. Broad use-case portfolios without workflow redesign produced incremental wins.
The “economic leverage points” framing (Theme 2) is functionally equivalent to the workflow-redesign thesis: find the 1–3 processes where AI changes the decision architecture, not just the task speed.

Source: research/04-consulting-firms/mckinsey-ai-transformation-manifesto-2026.md

MIT CISR Scaling AI Maturity — 4S Framework (Mar 2026)

The Stage 2→3 transition (pilots → scaled value) is where workflow redesign becomes non-optional. MIT CISR’s “4S” framework names four organizational capabilities gating the transition: Strategy, Systems, Synchronization, Stewardship — none of which are technology purchases.
Updated financial data (n=152, Aug 2025): Stage 1 organizations post −26.5 pp growth vs. industry; Stage 3 posts +4.7 pp; Stage 4 posts +13.9 pp. Total spread: 40.4 pp of growth between bottom and top.
Guardian Life compressed RFP/quoting from 5–7 days to 24 hours — but only after consolidating enterprise data and moving to microservices/API architecture. The workflow redesign preceded the AI deployment.
Italgas invested in platform architecture since 2017 (IoT, data platform, 23 deployed models) before WorkOnSite delivered 40% faster project completion. Infrastructure-first, AI-second.

Sources: research/01-ai-native-landscape/mit-cisr-scaling-ai-maturity-bottom-line-2026.md, research/09-ai-adoption-cycle/year-2-ai-roadmap-scaling-beyond-pilot.md

BCG AI Workforce Transformation — 10-20-70 Operating Model Redesign (Feb 2026)

The fourth component of BCG’s 70% “people” value share is explicitly “develop an AI-enhanced operating model” — redesigned workflows, roles, governance, and org design combining human and AI capabilities. Technology deployment without operating model redesign captures at most 30% of available value.
Future-built companies are 5x more likely to do strategic workforce planning than laggards, reshaping job architectures and organizational structures with AI at the core. Workforce planning is the mechanism that connects AI deployment to workflow redesign.
BCG frames AI adoption failure as spreading efforts across “dozens or hundreds of use cases” rather than concentrating on 3-4 central priorities — same domain-concentration finding as McKinsey’s Manifesto (1-3 domains).

Source: research/04-consulting-firms/bcg-ai-workforce-transformation-2026.md

Deloitte State of AI 2026

Deloitte’s sixth annual survey (n=3,235, 24 countries, Aug–Sep 2025) quantifies the workflow-redesign gap: 66% report productivity gains, but 37% describe AI use as surface-level with minimal process change. Only 34% are deeply transforming products, services, or business models.
84% of organizations have not redesigned jobs around AI capabilities — the most direct explanation for why productivity gains are not translating into business transformation.
The 34%/30%/37% breakdown (deep transformation / process redesign / surface adoption) aligns with BCG’s 5% and McKinsey’s 6% “high performer” thresholds: roughly one-third are doing the hard work, two-thirds are adding tools on top of existing processes.

Source: research/07-adoption-challenges/deloitte-state-of-ai-enterprise-2026.md — see also the fuller treatment of this survey further below, including preparedness scores and talent strategy breakdown.

Professional Services Pricing as Workflow Redesign

Professional services faces the sharpest workflow redesign pressure: AI collapses input-time (billable hours), forcing a shift to outcome-based pricing.
Accounting firms redirecting AI-compressed compliance into advisory report 30%+ higher recurring revenue (AICPA CAS 2025). The redesign is “do different work,” not “do same work faster.”
66% of professional services firms turn down work due to resourcing; 87% plan AI agents as workforce (Kantata 2025, n=200).

Source: research/09-ai-adoption-cycle/ai-augmented-professional-services-delivery.md

See also: research/06-industry-verticals/cpacom-ai-in-accounting-2025.md — CPA.com/AICPA 2025 practitioner synthesis: accounting firms reaching 80%+ tax-return automation rebuilt intake, review, and liability workflows before deploying AI on client deliverables; audit/advisory remain at pilot stage due to liability constraints, not capability limits (MEDIUM, TIER 2)

AI Native Adoption Cycle — Stage Progression as Workflow Redesign Proxy

The 6-stage adoption cycle (AI-Unaware → AI-Native) maps directly to workflow redesign maturity: Stage 2→3 is the critical transition where pilots fail because workflow redesign hasn’t occurred.
McKinsey (Nov 2025, n=1,993): two-thirds of enterprises cannot escape pilot purgatory — functionally a workflow-redesign failure, not a technology failure.
The spectrum analysis (Table Stakes → Radical Frontier) tracks the same dynamic: tools at the “emerging standard” tier (AI code review, test generation, multi-file edits) require workflow changes to capture value; tools at the table-stakes tier do not.

Sources: research/09-ai-adoption-cycle/ai-native-adoption-cycle.md, research/08-radical-vs-tablestakes/spectrum-analysis.md

AI Steady-State Operating Model — Workflow Redesign as Ongoing Discipline

The steady-state operating model positions workflow redesign not as a one-time deployment activity but as a continuous operating capability: quarterly workflow audits, AI governance committee reviews, and change management cadence that persists post-deployment.
Prosci (Jan 2026, n=1,107): organizations that sustain change management beyond initial deployment are 6x more likely to meet objectives — directly applicable to workflow redesign durability.
ModelOp (March 2026, n=100): 78% of enterprises with formal AI governance report measurable business impact vs. 23% without — governance sustains the redesigned workflows.

Source: research/09-ai-adoption-cycle/ai-steady-state-operating-model.md

AI Failure Pattern Library — Workflow Bypass Archetype (Mar 2026)

Pattern 3 (“Workflow Bypass”) accounts for a dominant share of the 42% AI project abandonment rate (S&P Global, n=1,006, Mar 2025). Organizations deploy AI into existing workflows without redesigning how work flows — individual productivity improves, organizational throughput does not.
ActivTrak (n=163,638 workers, 443M hours, 2025): after AI deployment with no workflow redesign, email volume +104%, chat +145%, deep focus sessions −9%. The tool added speed without subtracting work.
Deloitte (n=3,235, 2025): 37% of organizations use AI at a surface level with no process changes — the Workflow Bypass pattern in action.
BCG high performers follow 10/20/70: 10% algorithms, 20% technology/data, 70% people and process change. The workflow redesign is the AI project.

Source: research/07-adoption-challenges/ai-failure-pattern-library.md

Change Management Evidence — Why 70% Fail

Structured change management produces 88% project success rates vs. 13% without (Prosci benchmark). The gap is not marginal.
McKinsey’s high performers (6% of n=~2,000) share one trait: they redesign workflows around AI rather than layering AI onto existing processes — 3.6x more likely to pursue enterprise-level change.
BCG (n=10,600, Jun 2025): positive AI sentiment jumps from 15% to 55% with strong leadership support, but only 25% of frontline employees report receiving it. The trust gap (+1.09 exec vs. +0.33 frontline) is the hidden killer.
Training hours predict adoption: 79% of employees with 5+ hours become regular users vs. 67% with less (BCG, 2025).

Source: research/07-adoption-challenges/ai-change-management-best-practices.md

Gartner I&O AI Use Case Survey (Apr 2026, n=782)

33% of I&O leaders who achieved AI success attribute it to embedding AI into existing workflows and systems — not running it as a side project.
53% of I&O AI wins occur in ITSM, where processes are mature and well-defined — bounded workflow integration beats ambitious greenfield deployments.
Failures concentrate in auto-remediation, self-healing infrastructure, and agent-led cross-system workflows — use cases that skip workflow integration and attempt autonomous operation.
Only 28% of I&O AI use cases fully succeed and meet ROI; 57% of those who failed said they expected too much, too fast — the same overambition-without-redesign pattern seen in McKinsey and BCG data.

Source: research/05-analyst-firms/gartner-ai-io-stalling-2026.md

IBM IBV “The Enterprise in 2030” (Jan 2026, n=2,000)

68% of executives worry AI initiatives will fail due to lack of integration with core business activities — naming the same integration gap that McKinsey, BCG, and Gartner identify as the primary bottleneck.
AI-first organizations (those integrating AI into products/services and designing AI-first tasks) anticipate 70% greater productivity improvement, 74% greater cycle-time reduction, and 67% greater project delivery improvement vs. peers — all contingent on workflow redesign, not tool deployment.
68% say current organizational structures are impediments to realizing AI’s full value — the structural barrier to workflow redesign, not a technology gap.

Source: research/04-consulting-firms/ibm-ibv-enterprise-2030.md

IBM IBV “ERP meets AI” (May 2025, n=1,500)

Organizations embedding AI into core ERP systems report 27% higher ROI and 9% stronger operating margins vs. cautious adopters — ERP integration is a proxy for workflow-level AI embedding.
Investment levels between bullish (76%) and bearish (58%) groups are closer than expected; the real differentiator is execution capabilities: training (81% vs. 45%), governance (69% vs. 39%), and organizational permission to experiment (57% vs. 29%).
82% of AI-bullish expect enterprise functional processes to change from AI in the next year vs. 31% of bearish — the bullish cohort plans workflow redesign, the bearish cohort plans tool deployment.
~33% admit current architecture impedes AI progress — the same structural barrier identified across McKinsey, BCG, and Gartner data.

Source: research/04-consulting-firms/ibm-ibv-erp-meets-ai-2026.md

Deloitte Insights “Bridging the AI value gap” (Feb 27, 2026, n=1,394)

Disaggregates BCG’s 10-20-70 “people and process” pillar into three testable structural variables: team size, cognitive diversity, and cross-functional connectedness. Teams of 10+ report 2x the innovation, problem-solving, and efficiency gains of teams with ≤4 members.
91% of high-performing AI teams deliberately hire for varied skills (vs. 68%); 86% prioritize diverse experience (vs. 51%); 54% incorporate diverse viewpoints in decisions (vs. 27%) — the team-composition layer inside workflow redesign.
Cross-functional teams are 30% more likely to report significant efficiency and innovation gains — the connectedness layer that determines whether redesigned workflows actually move end-to-end throughput instead of just shifting the bottleneck.
Companion Deloitte Tech Trends research: 93% of tech funding flows to technology, 7% to people enablement — the same resource misallocation BCG’s 10-20-70 framework flags from the value-distribution angle. Workflow redesign without team redesign is a half-built intervention.
Vendor caveat: Deloitte Consulting benefits commercially from workforce-redesign engagement recommendations. Cross-sectional self-report survey — associations, not causation.

Source: research/07-adoption-challenges/deloitte-ai-value-gap-team-dynamics-2026.md

McKinsey Global Tech Agenda 2026 (Reil-Jerenz, Romanelli, Jogani, Catlin, Halawa, Himatsingka — Feb 2026, n=632 C-level)

Names “rewire the business around AI” as the fourth of four 2026 CIO imperatives, directly quoting the language of workflow redesign. More than half of top performers have transformed the IT function using AI in the past two years, compared with 38% of others. The divergence starts inside the IT function itself.
The structural counterpart: product-and-platform operating models. ~10% of top performers have fully adopted product-and-platform models across all teams, more than 4x the rate at others (~2%). Nearly half of top performers have at least half their teams operating this way. With the shift, decisions happen within days instead of months, handoffs drop, and information flow increases — the operating-model prerequisite for end-to-end workflow redesign.
The Aviva case study is the clearest example of the BCG 10/20/70 rule in practice: 80 AI models deployed across the end-to-end claims journey alongside a full operating model and cultural transformation. Outcomes: liability-assessment time −23 days, routing accuracy +30%, complaints −65%, CSAT 7x. The tech-only version of the same deployment would have surfaced as the 60% who report minimal value.
The change-management inversion signal: nearly a quarter of top performers — compared with just 15% of others — cite change management as a core challenge to scaling agentic AI. The most ambitious workflow-redesign work surfaces change management as the binding constraint. If change management is not yet in the top three reported blockers, the workflow-redesign scope is probably too narrow to generate returns.
Practical diagnostic for a mid-market CIO: if agentic AI is being deployed without a corresponding product-and-platform operating-model shift, the deployment is tool-layer work, not workflow redesign. The McKinsey 2026 benchmark is that both shifts happen together at the top-performer cohort.

Source: research/04-consulting-firms/mckinsey-global-tech-agenda-2026.md

Deloitte “State of AI in the Enterprise 2026” (“From Ambition to Activation”, n=3,235, Aug–Sep 2025)

The 34%/66% split is the most direct large-n quantification of the workflow-redesign gap: 34% of organizations are genuinely reimagining their business (Stage 3 workflow and business-model redesign); 66% are capturing efficiency gains without structural redesign. The 37% surface-level cohort — minimal process change — represents the organizations where tool deployment has not been accompanied by any workflow work.
84% of organizations have not redesigned jobs around AI. The talent adjustment emphasis (53% education; 48% upskilling) without corresponding role and workflow redesign is the most common form of the 10/20/70 misallocation: spending on the 10% and 20% while leaving the 70% untouched.
The revenue gap is the business consequence: 74% expect AI to drive revenue growth; 20% report it today. Revenue impact requires the end-to-end workflow redesign that BCG finds delivers 3–4x the returns of incremental process improvements. Efficiency gains (66% report them) do not accumulate to the revenue line without the structural redesign.
25% of leaders now report transformative impact — up from approximately 12% a year prior. The doubling is the leading indicator that the workflow-redesign cohort is growing. But the pace shows the distance remaining: the majority of the global enterprise population is still in efficiency-gain territory.
Preparedness scores expose the execution gap directly: Only 4% of companies rate themselves highly prepared on technology infrastructure; only 1% on data management — the two foundations every AI workflow deployment runs on. Strategy scores 42% highly prepared, risk/governance 30%, talent 20%. The inversion (high strategic confidence, near-zero operational readiness) is the structural condition that generates pilot-to-production failures.
Job redesign is lagging automation expectations: 36% expect at least 10% of jobs to be fully automated within one year; 82% expect the same within three years. With 84% not having redesigned roles yet, the gap between automation pace and workforce redesign pace is widening — not closing.
Talent strategy is education-heavy, redesign-light: 53% are educating, 48% upskilling, but only 33% redesigning career paths and 16% actually moving to non-hierarchical structures despite 53% having considered them.

Source: research/07-adoption-challenges/deloitte-state-of-ai-enterprise-2026.md

MIT CISR “Leveraging Digital Colleagues for Enterprise Value” (Weill & Woerner, Apr 16, 2026, n=132)

The “digital colleague” framing gives workflow redesign a structural test: AI assistants (prompt-response, session-bounded) cannot produce the 25% revenue-per-employee gains that 75% of respondents project, because they don’t require or enable the workflow redesign the MIT CISR regression identifies as the strongest value predictor. Digital colleagues — multi-step autonomous execution, auditability, conditional human handoff — only work when the workflow they operate in has been redesigned around them.
Statistically significant predictors of value capture (p<.05 across eight outcome measures): (1) major workflow redesign — 22% of organizations; (2) redefined roles and performance metrics — 9%; (3) high utilization. The 22% figure aligns precisely with McKinsey’s finding that only 21% have redesigned workflows (the #1 EBIT predictor). Two independent studies, different samples, same structural constraint.
Mallesons case study: 96% active adoption among 1,300+ legal staff required 300 documented use cases, monthly AI-strategy reviews, and a formal certification program — the operational evidence that high utilization is a lagging indicator of redesign, not a precondition.
The 75%/22% tension is the sharpest statement of the workflow-redesign gap in the April 2026 corpus: 75% project material revenue impact, 22% have done the work that MIT CISR’s data says is required to reach it.

Source: research/01-ai-native-landscape/mit-cisr-digital-colleagues-enterprise-value-2026.md

Palantir AIPCon 8 & 9 — Ontology as Workflow-Redesign Substrate (2025–2026)

Nebraska Medicine (AIPCon 8, September 2025): built a new medical-necessity validation workflow in 10 hours — only possible because a 6-month foundational Ontology (unified data model across patient flow, nurse allocation, clinical supplies, revenue cycle) had already been established. Speed for use case #2 through N is a downstream product of workflow redesign on use case #1.
The audited financial signal reinforces the pattern: Palantir’s Net Dollar Retention reached 139% in Q4 2025 — existing customers expanding spend 39% above renewal baseline. Companies that genuinely redesigned workflows are not canceling; they are expanding.
Palantir’s AIP Bootcamp model (5-day sprint, live data, working use case or no deal) is the most concentrated available proof that workflow-first deployment beats slideware-first evaluation. ~75% of bootcamp participants convert to contract; the historic enterprise AI sales cycle (12 months) compresses to days when the prospect builds on their own live data.
The failure mode this illuminates: treating AI pilots as standalone experiments rather than as the first layer of a compounding infrastructure. Every unnamed Fortune 500 case in the Palantir case library (CPG digital twin in 5 days, bank 60%/90% monitoring improvement, CAZ Investments 100x lead capacity) ran on top of existing data work, not instead of it.

Source: research/01-ai-native-landscape/palantir-aipcon-enterprise-agentic-deployment-2026.md

Failure-Mode Synthesis — Pre-Conditions That Predict Workflow Deployment Failure

The reverse of the success pattern above. From a corpus synthesis of failure-mode evidence across McKinsey (n=1,993), Writer (n=2,400), Grant Thornton (n=950), OutSystems (n=~1,900), MIT CISR FinCo, METR RCT, and Atlan 200-deployment analysis:

The #1 pre-deployment red flag: no explicit mandate to eliminate an activity (not just accelerate it). If the workflow design does not answer “what disappears?”, the bottleneck moves downstream, not away. ActivTrak (n=163,638 workers) measured the result: post-AI deployment, email +104%, chat +145%, deep focus -9%.
The workflow condition that produces negative productivity: experienced developers in mature codebases doing complex multi-system work. METR RCT (n=16, pre-registered) found these developers were 19% slower with AI tools. The Faros data shows +98% PRs with zero delivery throughput improvement — the bottleneck moved from coding to review and nobody redesigned the review step.
The 55/18 rule: 55% of McKinsey high performers redesigned workflows; 18% of others did. Among the 18%, ROI was minimal regardless of tool quality.
The production-path trap: most pilots are designed to answer “does AI work?” without a production path attached. The pilot budget clears; the 3-5x production budget does not. S&P Global (n=1,006) finds 46% of AI POCs scrapped before production as a result.

Source: research/07-adoption-challenges/ai-deployment-failure-modes-synthesis.md

The Functional Manager’s Workflow Selection Decision (April 2026)

The workflow-redesign research describes organizational-level patterns. The operational reality is that in a 200-2,000 person company, the workflow selection decision is made by a Director or VP, not a CIO or COO. The functional manager’s 30-minute workflow audit:

Map the three highest-volume repetitive decisions your team makes each week. Not the most complex — the most frequent.
Score each against: (1) structured, consistent input data? (2) definable correct output? (3) volume high enough to compound? (4) bounded error consequence? (5) auditable after the fact? (6) clear human handoff trigger?
Any workflow scoring “yes” on all six is AI-ready now, regardless of company AI maturity. The workflow-readiness question is independent of the organizational-maturity question.

The failure mode this catches: managers waiting for a transformation program before deploying anything, while one workflow in their team is already AI-ready and the bottleneck is permission, not capability.

Source: research/07-adoption-challenges/middle-manager-ai-deployment-playbook.md

Workflow-Level Readiness: The Six-Criterion Scorecard (April 2026)

Organizational AI maturity and workflow AI-readiness are independent variables. A company can be organizationally immature and still have one workflow that is AI-ready today. The six-criterion scorecard below tests a specific process in 30 minutes — no transformation program required.

The six criteria, each scored 1–5:

Data Structure Readiness — Input data is structured, consistent, and accessible without manual extraction. Score 5: single system of record, documented schema, 12+ months complete, quality audited. Failure signal: data lives in 3+ systems, requires manual export/cleaning.
Decision Clarity — The correct output for a given input is definable in writing and verifiable by a reviewer. Score 5: rule-based criteria, verification under 2 minutes, error patterns enumerated. Failure signal: correct output depends on context or judgment not present in the input.
Volume Threshold — The workflow executes often enough that automation compounds materially. Score 5: 5,000+ transactions/month, stable volume, 12+ months of history. Failure signal: under 50 transactions/month or highly variable.
Error Cost — The consequence of an incorrect AI output is priced and acceptable within the current oversight design. Score 5: minimal consequences, fully reversible, error rate tracked. Failure signal: errors create irreversible regulatory, legal, or safety consequences with no detection mechanism.
Auditability — AI output can be traced and reviewed after the fact by a non-participant. Score 5: complete audit trail (input, model version, output, reviewer, timestamp), quality monitored automatically. Failure signal: no audit trail, outputs not logged.
Human Handoff Clarity — It is unambiguous — to the AI system, the reviewer, and the escalation recipient — when the workflow should involve a human. Score 5: escalation criteria are system-enforced, automatic routing on anomalies, SLA violations trigger alerts. Failure signal: no written escalation criteria, reviewer discretion is the only mechanism.

Decision rule: A workflow scoring 4 or 5 on all six criteria is AI-ready now, regardless of organizational maturity. Any criterion scoring 1 is a structural problem that tool deployment will amplify — stop and remediate before investing in tooling.

Corpus evidence base: SlickDeals deal scoring (AWS re:Invent 2025) — 4-5 on all six, result: 360x latency improvement, 7% revenue gain. Klarna aggressive automation phase — scored 2 on error cost, auditability, and handoff clarity; result: quality collapse, forced reversal (multiple independent sources, 2025). Atlan 200-deployment analysis (n=200, France, 2022-2025): median +159.8% ROI with 8-month breakeven in assessed deployments; 4.2x fewer critical incidents with human-in-the-loop governance.

Source: research/07-adoption-challenges/workflow-level-ai-readiness-checklist.md

Bain: Agentic AI Deployment Sequence — Governance Before Orchestration (April 2026)

Bain’s framework for agentic AI deployment establishes a non-negotiable sequencing rule: governance and observability infrastructure must be in place before orchestration and scale. Organizations that attempt multi-agent deployment without this foundation accumulate failures they will eventually have to unwind.

Three-phase model: Foundation (governance, data quality, observability, identity) → Orchestration (multi-agent coordination, tool access management) → Cross-Domain Scale. The sequence is not a maturity theory — it reflects what separates organizations shipping production-grade agentic AI from those stuck in the pilot-to-production gap.
Mid-market advantage: Governance frameworks are simpler to build at lower organizational complexity. The sequencing risk is the same as at large enterprises; the implementation complexity is smaller.
Architecture shift: Traditional role-based access controls fail for AI agents (which need contextual, least-privilege permissions); legacy observability tools cannot trace reasoning paths through multi-step workflows. Both must be rebuilt before multi-agent orchestration begins.

Source: research/04-consulting-firms/bain-agentic-ai-production-2026.md

Stanford Digital Economy Lab — Enterprise AI Playbook: Invisible Costs Finding (Apr 2026, n=51)

The Playbook is the most empirically grounded cross-industry source in the corpus for the invisible-costs thesis — that workflow redesign, change management, and organizational factors are the binding constraints, not technology.

77% of the hardest AI implementation challenges were organizational, not technical:

Change management (adoption resistance, culture, trust-building)
Data quality (though LLMs are substantially lowering this bar — 91% of cases successfully processed unstructured data)
Process redesign — the workflow must change, not just be augmented
Building organizational trust in AI outputs

Technical challenges (model selection, API integration, latency, cost optimization) were consistently described as the easiest part of the project across all 51 cases.

Timeline variance is organizational: The same use cases — invoice processing, customer support, content generation, recruiting — took weeks at one organization and years at another. The technology was identical. The differentiators were entirely organizational:

Accelerators: Executive Sponsorship (43% cited as primary), Existing Foundation (32%), End-User Willingness (25%)
Delays: Learning Curve/Iteration (25%), Data Quality (21%), Regulatory/Compliance (21%), Process Documentation Gaps (21%)
Technology did not appear in the delay list

The workflow architecture determines the outcome:

Oversight Design	AI Workload Share	Median Productivity Gain
Escalation (humans handle exceptions)	80%+	71%
Collaboration (parallel human–AI)	Variable	~54%
Approval (humans sign off on all outputs)	Variable	30%

The 71% vs. 30% gap is workflow architecture, not model selection. Organizations that redesigned their process so AI handles 80%+ of the work and humans handle only exceptions achieved more than double the productivity gain of organizations that kept humans in the sequential approval path.

The “project definition” differentiator: Organizations that treated AI deployment as a software installation project failed. Organizations that treated it as a workflow redesign project that happens to involve software succeeded. That framing change determines which people are in the room (process owners, not just IT), what counts as success (workflow outcomes, not deployment milestones), and how long the project takes (shorter, because the right people have authority to change the workflow).

Process documentation as the hidden prerequisite: 27% of prior failures in the sample were attributed to “critical knowledge never captured” — process knowledge that existed in employees’ heads but was never documented. AI cannot reliably execute an undocumented process. Process documentation is not an IT task; it is the prerequisite that determines whether AI can take over the workflow or whether it requires constant human guidance.

Source: research/07-adoption-challenges/stanford-enterprise-ai-playbook-2026.md

PwC Digital Trends in Operations 2026: The 4% Threshold

PwC’s May 2026 survey of 767 U.S. operations and supply chain leaders quantifies what “complete” workflow redesign actually requires at the enterprise level. Only 4% of respondents meet all four criteria simultaneously:

AI fully embedded enterprise-wide
No significant barriers to scaling autonomous agents
Collaborative, horizontal operating structure
Technology investments fully delivering expected results

The structural gap is the differentiating factor: 94% of companies with siloed or partially integrated operating structures expect to shift toward horizontal, networked models — but only 41% currently operate that way. The expectation exceeds the reality by more than 2x. Companies are planning to reorganize for AI rather than reorganizing.

This pairs directly with PwC’s separate AI Performance Study (n=1,217): the top 20% of AI value capturers are twice as likely to have redesigned workflows. The Digital Trends in Operations data makes the organizational structure requirement explicit — it is not sufficient to redesign individual workflows. The operating model itself must become horizontal before agents can operate end-to-end without functional silos blocking them.

Source: research/04-consulting-firms/pwc-digital-trends-operations-2026.md

DORA 2026.01: The J-Curve as a Workflow Redesign Failure Mode

DORA’s ROI of AI framework (April 2026, n=~5,000 tech professionals) identifies a specific workflow redesign failure mode: organizations that deploy AI coding tools without redesigning their review and deployment processes produce a J-Curve — productivity falls before it rises.

Three mechanisms create the dip: learning curve during tool adoption, the “verification tax” of reviewing AI-generated code, and downstream bottlenecks (testing, change advisory boards, deployment gates) that were sized for the old code throughput. Organizations that misread the dip as failure and defund the program before the curve turns never capture the return.

The DORA finding matches the broader corpus pattern: the tool is not the constraint. The constraint is the surrounding organizational system. Instability is the measurable consequence — AI adoption correlates with rising change failure rates in DORA’s data, consistent with the Faros AI study’s finding of +91% review time and zero DORA metrics improvement at the team level.

Key implication: ROI from AI coding tools requires investment in automated testing, CI/CD hardening, and small-batch deployment practices alongside the tool deployment — not as a future phase.

Source: research/02-corporate-tools/dora-roi-ai-assisted-software-development-2026.md

Census Bureau BTOS AI Supplement — Function Deployment Pattern (2026)

The first nationally representative disaggregation of AI by business function finds Sales and Marketing as the leading deployment target — ahead of IT — with implications for how organizations should structure cross-functional AI governance.

Sales and Marketing is the #1 AI function at 52% of adopters — not IT (41%); Strategy and Business Development ranks second at 45%
This means AI deployment decisions are already being made outside IT’s purview in most adopting organizations; shadow AI risk and data governance gaps accumulate in revenue-generating functions first
57% of adopters use AI in three or fewer functions — most enterprise AI is still point-deployment, not enterprise-wide workflow redesign
The gap between function-level adoption (firms using AI in Sales/Marketing) and enterprise-wide integration is the workflow-redesign gap: tools are deployed on existing workflows without redesigning the surrounding process
Employment-weighted adoption in Information, Professional Services, and Finance large firms reaches 60–70% — the peer baseline in those sectors has already shifted; the competitive question is no longer whether to adopt but how deep the workflow redesign goes

Source: research/01-ai-native-landscape/census-btos-ai-diffusion-microstructure-2026.md

HBR Analytic Services — Workflow Integration as the Value Gate (April 2026)

Two companion HBR Analytic Services studies (n=325 December 2025, n=385 March 2026) quantify the specific gap between deploying AI and integrating it into workflows — and show the value difference between the two.

Only 18% of organizations have AI primarily integrated within workflows — 34% use standalone tools alongside processes; 34% use a mixed approach
71% of organizations embedding AI in processes achieve substantial or moderate value — vs. 16% overall reporting high measurable value; the gap is the integration delta
94% say connected data, processes, and apps are critical for AI success; only 27% have that connectivity — the gap between aspiration and architecture is the workflow redesign backlog
Only 39% have unstructured data prepared for AI vs. 65% for structured data — the data most critical to knowledge-worker workflows (contracts, emails, case notes) is least ready
69% agree legacy systems limit enterprise-wide AI scaling — legacy architecture is the structural blocker to workflow integration, not tool availability
92% agree AI agents need guardrails; only 48% have defined them — agentic deployment is outrunning workflow governance

MIT CISR “Leveraging Digital Colleagues for Enterprise Value” (Weill & Woerner, April 16, 2026, n=132)

Research file: research/11-education-approaches/mit-cisr-digital-colleagues-enterprise-value-2026.md

The freshest independent academic anchor on the workflow-redesign gap. Regression analysis (p<.05) confirms three capabilities predict value capture from agentic AI (“digital colleagues”):

Only 22% of organizations have completed major workflow redesign for digital colleagues — the statistically significant predictor of value capture.
Only 9% have formally redefined roles and metrics to integrate digital colleagues into workforce strategy — the most telling readiness gap in the 2026 corpus.
34% have high-level active use — and another third are still testing viability.
75% of executives project a 25% revenue-per-employee gain over three years (median: 15%) — creating a measurable gap between expected returns and the readiness work required to achieve them.
Mallesons case study: 96% staff active use, 50% using >4 days/week, 20% cycle-time reduction — achieved by framing Harvey AI as a team member rather than a tool and managing 40+ workflow updates monthly.

The 9%/22% figures directly contradict vendor claims that agentic AI is “ready to scale.” Scale requires redesign; redesign has happened at fewer than one in four organizations.

Source: research/11-education-approaches/mit-cisr-digital-colleagues-enterprise-value-2026.md

Source: research/07-adoption-challenges/hbr-analytic-services-ai-readiness-workflow-2026.md

WEF/Accenture MINDS “Proof over Promise” (Jan 2026)

32 named global case studies of AI organizations that achieved measurable production-scale results. Core finding: gains are large (double-digit productivity/revenue) but require operational redesign, not technology installation.

Five patterns of high-performer organizations: (1) AI into decision-making as structural capability; (2) workflow redesign for human-AI collaboration, not replacement; (3) deliberate data strategy (historical + real-time + synthetic); (4) unified platform architecture; (5) trust-by-design governance
90% of enterprises expect hybrid human-AI teams to become standard within 3 years — but only 23% have a formal operating model for those teams today. The redesign deadline is now.
Data quality named as “most persistent scaling barrier” — organizations that built differentiated data advantages scaled faster than those that applied AI to existing data
~75% of leading organizations reinvest AI returns into new deployment domains — the compounding loop that separates durable high performers from one-time efficiency wins

Source: research/07-adoption-challenges/wef-accenture-proof-over-promise-2026.md

Adjacent Voice: Ethan Mollick — “Jagged Frontier” Framing

Wharton professor Ethan Mollick’s “jagged frontier” framework offers a useful vocabulary for explaining why workflow redesign matters: AI is simultaneously superhuman at some tasks and near-useless at adjacent ones within the same role. The frontier is not a line; it is jagged. This explains why identical tool deployments produce radically different outcomes depending on where in the workflow the AI is applied. Workflow redesign is the practice of mapping the frontier before deploying.

Profile: research/15-adjacent-voices/ethan-mollick.md — MEDIUM credibility (independent academic; no vendor interest; practitioner framing, not primary survey data)

Oliver Wyman Forum CEO Agenda 2026 (April 2026)

n=415 CEOs, co-published with NYSE, ~10% global market cap. CEO-level finding on workflow redesign as the primary differentiator.

Deployment leaders redesign workflows at 49% vs. 38% for all CEOs — an 11-point gap at the CEO level confirms the pattern holds at the top of the org chart, not just among technology leaders
Deployment leaders achieve 49% ROI satisfaction vs. 17% for laggards — the same cohort that redesigns workflows generates 3x the ROI satisfaction rate
CEOs who redesign workflows treat entry-level talent as amplifiable (24% of deployment leaders plan to increase junior hiring), while non-redesigners are cutting it (43% of all CEOs deprioritizing junior hiring)
Consistent with: PwC AI Performance Study 2026 (2x workflow redesign likelihood among top 20%), Deloitte State of AI 2026 (34% deep transformation vs. 66% efficiency-only), BCG AI Radar 2026 (5% value-capturing cohort)

Source: research/04-consulting-firms/oliver-wyman-ceo-agenda-2026.md

Goldman Sachs Economic Research — Workflow Redesign as the Mechanism Behind 30% Gains (Feb–May 2026)

Goldman Sachs finds no economy-wide AI productivity gain but 30% productivity improvement in specific use cases — the difference is workflow integration depth.

Economy-wide: “basically zero” contribution to GDP growth despite 70% of S&P 500 management teams discussing AI on earnings calls — consistent with BCG/McKinsey finding that most deployment is task-level overlay, not workflow redesign
30% productivity gain in customer service and software development — both are characterized by deep AI integration into the primary work loop, which is the operational definition of workflow redesign
Goldman’s finding implies the 30%-gain use cases share a structural property: AI is embedded in the workflow rather than layered on top of it — the same condition PwC, BCG, and McKinsey identify as the separator between high performers and the median
“Productivity gains require sustained investment in business process re-engineering” — Goldman names redesign explicitly as the precondition; capital expenditure on AI without process investment does not appear in GDP growth statistics

Source: research/01-ai-native-landscape/goldman-sachs-ai-economic-research-2026.md · Feb–May 2026 · HIGH · TIER 1

HBR “The Psychological Costs of Adopting AI” — When Redesign Fails Because of Psychological Debt (May/Aug 2026)

Two HBR studies identify the human-side mechanism that causes workflow redesign to fail even when the technical integration is correct.

“Psychological debt” accumulates during sustained AI use — six costs (cognitive offloading, reduced autonomy, diminished competence, weakened social connection, credibility loss, identity threat) predict avoidance even among workers who acknowledge AI’s value (Champniss, n=1,200, May 2026)
The competence penalty removes social incentive to use AI in redesigned workflows — workers whose AI use is visible to peers receive 9% lower competence ratings; female engineers face −13%. In collaborative workflows where redesign requires visible AI use, this penalty actively undermines adoption (Acar et al., pre-registered n=1,026 + observational n=28,698)
Technical redesign that makes AI use visible without addressing the social penalty produces adoption regression — workers revert to non-AI methods to protect professional credibility, even after the workflow has been technically redesigned
The six psychological debt categories map onto the change management phase of any workflow redesign: cognitive offloading and reduced autonomy peak during transition; social connection and identity threat persist into steady state

Source: research/07-adoption-challenges/hbr-psychological-costs-ai-adoption-2026.md · May/Aug 2026 · MEDIUM-HIGH/HIGH · TIER 1/TIER 2

Capgemini Research Institute “AI Perspectives 2026” (Jan 2026)

n=1,505 executives, >$1B companies, 15 countries, November 2025 survey period. Companion CXO study: n=500 CXOs at >$10B companies.

63% pruning low-value AI projects — organizations that moved to the accountability phase report dropping broad shallow pilot coverage in favor of fewer, deeper deployments with redesigned workflows
Top operationalization enablers: executive sponsorship (67%), workforce upskilling (60%), governance (53%) — workflow redesign requires all three; technology is not named in the top 3
Organizations with redesigned workflows vs. AI overlay: only the former appear in the 5–8% high-performer pool (BCG/McKinsey benchmark); Capgemini’s 38% who “operationalized” includes many overlay deployments that have not yet hit financial performance thresholds
5-year investment horizon majority — organizations that treat AI as infrastructure (workflow-level investment) rather than productivity software are planning on the correct time horizon

Source: research/04-consulting-firms/capgemini-ai-perspectives-2026.md

Vanguard — Portfolio Architecture as Workflow Redesign ($500M, Oct 2025)

Named case study (Davenport/Bean, MIT SMR). TIER 2 (Oct 2025). Single firm; no control group.

Vanguard’s $500M AI value came from embedding AI at decision points, not task handoffs. Each of the five use cases (contact center, adviser intelligence, digital advisor, investment analytics, developer productivity) was chosen because a human decision was the bottleneck — not because the task was the easiest to automate.
CIO Nitin Tandon: “AI is embedded where it makes the most difference in decisions that improve investor outcomes and streamline how our teams deliver value.” This is the workflow-redesign frame: identify the decision, then design the AI role around it.
The dividend prediction LLM (22,000 earnings call transcripts, 5x cut-prediction accuracy) is the purest example: it does not replace an analyst workflow — it creates a new analytical input that did not exist before. AI that creates new capability beats AI that replaces existing labor for workflow ROI longevity.
Contrast with Klarna reversal (2024→2025): Klarna optimized for deflection rate (efficiency overlay on existing workflow), not customer decision quality (workflow redesign). Vanguard optimized for decision quality. The architectural difference explains the difference in outcomes.

Source: research/01-ai-native-landscape/vanguard-ai-roi-500m-case-study-2025.md · Oct 2025 · MEDIUM-HIGH · TIER 2

Coastal / Oxford Economics — “2026 AI Operations Report” (n=800, May 2026)

The problem-first sequencing failure and its operational consequences:

Only 26% of organizations begin AI initiatives with a clearly defined business problem. The majority start with a vendor or technology selection. This inverts the sequence that every major ROI study identifies as the primary driver of returns.
Organizations with a clear data and AI roadmap are 2.7x more likely to achieve positive ROI — but 74% of organizations lack one.
The consequence of problem-last sequencing: AI programs cannot define success before deployment, cannot measure ROI during operation, and cannot diagnose decay when returns decline. This is the operational mechanism behind Writer’s 75% “strategy theater” finding.
Only 1 in 6 organizations has a dedicated AI or transformation team — AI initiatives are handed back to business units post-launch with no function responsible for production performance. This is why Davenport finds “static deployment” costs −15% value at month six: no one is watching.

Source: research/07-adoption-challenges/coastal-oxford-economics-ai-operations-report-2026.md · May 2026 · MEDIUM · TIER 1

ServiceNow / Oxford Economics Enterprise AI Maturity Index 2026 — Pacesetter Workflow Gap (n=4,473, Apr 2026)

The second annual maturity index quantifies the workflow redesign gap between AI Pacesetters and the rest at population scale.

54% of Pacesetters are inventing new workflows leveraging human-AI collaboration vs. 12% of non-Pacesetters. This 42-point gap is the largest workflow redesign differential in the corpus — wider than the MIT CISR 22% finding (enterprises that completed major redesigns vs. those that haven’t).
60% of Pacesetters are connecting data and removing operational silos vs. 41% of others. Workflow redesign requires data readiness as a prerequisite; the organizations doing both are the same segment.
Platform approach: 70% of manufacturing Pacesetters take a platform-based AI deployment approach vs. 50% of others — consistent with Gartner’s April 2026 finding that enterprises are abandoning bolt-on AI for integrated workflow platforms.
The maturity decline narrative reinforces this: Organizations fell an average 9 points despite rising investment. The ones declining least are redesigning workflows. The ones declining most are layering AI onto unchanged processes.

Source: research/05-analyst-firms/servicenow-enterprise-ai-maturity-index-2026.md · Apr 2026 · MEDIUM · TIER 1

Stanford DEL — Enterprise AI Playbook: 51 Deployments (n=51, April 2026)

The most empirically grounded case study corpus on workflow redesign to date. Production-only criterion (3+ months sustained adoption, quantified outcomes), 41 organizations, 9 industries.

77% of the hardest implementation challenges were non-technical: change management, data quality, and process redesign dominated when practitioners were asked “what was the hardest thing to fix?” Technology was consistently described as the easiest part.
61% had a failed first attempt before successful deployment. First attempts failed because teams treated AI as a technology project rather than a process redesign project — applied to broken workflows, led by technical teams without business ownership. The workflow had to be fixed before AI could work on it, not after.
Escalation-based workflow models (AI autonomous 80%+, humans review only exceptions) delivered 71% median productivity gain vs. 30% for approval-based models (human reviews every output). The gap is not about oversight level per se — it’s about which workflow design captures the human judgment advantage at the right points rather than distributing human attention uniformly.
Agentic implementations — 20% of cases — showed 71% median productivity gains vs. 40% for high-automation non-agentic. The agentic cases succeeded in functions with high volume, clear success criteria, and recoverable errors — characteristics that let the workflow be designed around AI autonomy without catastrophic error exposure.
“Don’t just apply AI to your existing processes. That’s a mistake. We’re redesigning our workflow and that’s what makes us successful.” — Head of Operations, Technology Company (Stanford Playbook)

Source: research/07-adoption-challenges/stanford-enterprise-ai-playbook-2026.md · April 2026 · HIGH (Stanford DEL, Brynjolfsson, production-only) · TIER 1

Atlassian “State of Teams 2026” — The 6% ROI Problem (Jan–Feb 2026, n=12,035)

Research file: research/07-adoption-challenges/atlassian-state-of-teams-2026.md

85% of knowledge workers use AI; only 29% have embedded it in actual work flows. The gap between usage and integration is the workflow-redesign gap, measured with the largest knowledge-worker sample in the 2026 corpus.
Only 6% of executives can confirm clear, organization-wide AI ROI despite 89% saying AI increases speed. This is the adoption-to-value gap at the executive level — consistent with BCG’s 5% substantial-gains finding and McKinsey’s 6% high-performer cohort.
$161 billion annual fragmentation tax (Fortune 500) — coordination overhead, unclear ownership, and disconnected work that AI tools have not resolved because they were deployed on top of unchanged structural problems.
55% of executives report AI is widening performance gaps between teams — not narrowing them. Teams that did workflow redesign captured speed gains; teams that only adopted tools increased output into unchanged bottlenecks.
The 14% who cracked the ROI code share three workflow-design characteristics: AI used in planning and prioritization (5.6x more likely), AI increases collaboration quality (9.4x more likely), high worker trust in AI for information surfacing (2.3x more likely). The common thread: AI deployed upstream in the workflow (where decisions are made), not only downstream (where tasks are executed).

Source: research/07-adoption-challenges/atlassian-state-of-teams-2026.md · Jan–Feb 2026 · MEDIUM-HIGH (Atlassian vendor; double-blind; n=12,035; TIER 1)

Gartner Supply Chain — The Foundation-First Deployment Sequence (Apr–May 2026)

Source: research/05-analyst-firms/gartner-supply-chain-ai-adoption-barriers-2026.md

Supply chain domain evidence from two companion Gartner surveys (n=140 + n=509, Oct 2025 fieldwork) documents the same workflow-redesign gap in a specific function:

Only 17% of supply chain organizations pursue transformational operating model redesigns before activating AI; 83% apply AI incrementally to existing workflows
AI leaders in supply chain follow a consistent deployment sequence: data infrastructure → workflow redesign → role redesign → AI assistants → AI agents. Skipping to agents-first produces local optimisations that compound existing inefficiencies.
Gartner defines an “AI-native supply chain” as one designed from the ground up for AI decision-making — not existing workflows with AI functionality added. The distinction maps directly to the corpus-wide distinction between tool deployment and workflow redesign.
Corroborates BCG (5%), McKinsey (6%), Stanford DEL (71% gain only with redesigned workflows) from the supply chain angle

Source: research/05-analyst-firms/gartner-supply-chain-ai-adoption-barriers-2026.md · Oct–Nov 2025 fieldwork · MEDIUM-HIGH (Gartner independent; TIER 1)

Thomson Reuters CTO — The 50% AI Code Threshold (May 2026)

Source: research/13-multimodal-sources/enterprise-ai-innovators/2026-05-18-joel-ron-thomson-reuters-cto-on-ai-for-legal-and-tax-profes.md

Joel Ron (CTO, Thomson Reuters, 5,000 engineers, 100+ products) describes a specific workflow-redesign threshold:

When 50%+ of code is written by AI, the engineer’s role shifts from contributor to controller/governor. The mindset change is not incremental adoption; it is a categorical role redefinition. The engineer now builds systems to steer, correct, and guide AI rather than doing the work directly.
2025 OKR: 90%+ daily AI tool adoption. 2026 OKR: >50% of PRs shipped by AI, not just assisted. This is the sequencing the corpus tracks: tool access → high-frequency use → workflow integration → AI-primary execution.
Tax return agents as a fully redesigned workflow: Thomson Reuters agents ingest W-2s/1099s, interpret tax law, compute deductions, and deliver a draft return. The tax professional’s role becomes client advisory, not document processing. This matches the workflow-redesign pattern across the corpus — AI replaces volume tasks, humans own judgment and relationship.
The two trust blockers for enterprise agentic workflow redesign: (1) access control — what data and systems can the agent touch? (2) process change management — building workflows where agents act autonomously, not just advise.

METR 2026 Self-Report Survey — The RCT-to-Survey Transition (n=349, May 2026)

Source: research/01-ai-native-landscape/metr-self-reported-productivity-survey-2026.md · TIER 1 · HIGH (org) / MEDIUM (methodology)

METR’s May 2026 survey of 349 technical workers updates the RCT-era picture and shows how workflow adaptation changes the productivity signal over time.

Median perceived value: 2x — up from a retrospective 1.3x estimate for March 2025. Self-report surveys inflate reality by ~40 pp (METR’s own calibration), so the actual gain is likely 1.3–1.6x. The direction is upward; the magnitude is uncertain.
METR abandoned RCT design for late-2025 studies because 30–50% of developers refused to work without AI at all — making controlled experiments increasingly impossible. This is itself a workflow-redesign signal: developers have integrated AI deeply enough that exclusion affects baseline performance.
The gap between RCT (19% slower) and survey (2x faster) is not a contradiction — it is evidence that workflow redesign happened. The RCT (July 2025) measured developers using AI the way most people use it; the survey (May 2026) measures developers who have adapted their workflows over 10+ months. Workflow adaptation is the variable that closed the gap.
Implication for executives: the 19% slower result is what happens without workflow redesign; the upward trajectory in self-reported gains is what happens with it. Neither number alone is the right benchmark — the trajectory from RCT to survey tells the story.

Workday — The Copy/Paste Economy: System Integration as Workflow-Redesign Failure (May 2026)

Source: research/07-adoption-challenges/workday-copy-paste-economy-ai-fragmentation-2026.md · n=6,100 Finance/HR/IT/Ops globally · MEDIUM (Workday vendor; Harris Poll; TIER 1)

Workday’s May 2026 global study (n=6,100, Harris Poll) documents the specific mechanism by which workflow-non-redesign generates new coordination overhead rather than net time savings:

82% of employees spend significant time manually moving data between disconnected AI tools — the “Copy/Paste Economy” overhead that emerges when AI is deployed at the task level without redesigning end-to-end workflows
20% of employees lose 7+ hours weekly to this coordination overhead — a full additional workday consumed not by the original tasks but by managing the seams between AI tools
Only 27% of employees have AI connected directly to core systems of record; the remaining 73% are operating in the disconnected mode where this overhead accumulates
2.5x productivity differential: organizations with AI embedded in core workflows: 60% report meaningful gains; disconnected deployment: 24% — the same 2.5x gap the corpus documents through other lenses (BCG 5% vs. 95%, McKinsey 6% vs. 94%, Atlassian 14% vs. 86%)
40% rework tax on AI time savings, consistent with Workday’s January 2026 study (n=3,200, Hanover Research) and directionally with METR’s RCT finding (19% slower overall for experienced developers)
The compound failure mode: task-level AI reduces individual task time → 40% of savings lost to rework → remaining savings create coordination overhead from switching between disconnected tools → net system-level impact is substantially lower than task-level surveys capture
Corroborates Atlassian State of Teams 2026: 85% use AI, only 29% have embedded it in actual workflows; only 6% of executives can confirm clear org-wide ROI

IBM IBV “Blueprint for Agentic Operations” — Silos as the Bottleneck (May 2026, n=2,000+)

Source: research/12-agent-workers/ibm-ibv-blueprint-agentic-operations-2026.md · IBM IBV May 2026 · MEDIUM-HIGH (IBM commercial interest in watsonx/consulting; large-n C-suite primary survey) · TIER 1

IBM IBV’s May 2026 study (n=2,000+ C-suite, Oxford Economics sample design, 33 geographies) directly tests what blocks agentic workflow redesign at scale — and the answer is organizational, not technical.

82% of C-suite executives name functional silos as the primary barrier to agentic AI value — not model capability, not budget, not talent. The org chart is the bottleneck.
55% are already developing or deploying an agentic AI operating model; 23% have one in place today. The transition from assistive-AI-in-silos to agentic-AI-across-workflows is happening faster than most boards track.
Six capabilities predict whether agentic workflow adoption succeeds or stalls. When all six are present together, adoption is 5.4x more likely. Change management and AI governance are the two highest-leverage levers — exactly the factors most often treated as afterthoughts.
“Autonomous operators” — organizations that redesigned operating architecture for agentic execution, not just added agents to existing workflows — expect 64% ROI improvement vs. 56% for all others. The performance gap is structural.
The workflow-redesign implication: organizations that deploy agents within existing departmental workflows without redesigning cross-functional processes will capture automation efficiency gains within silos but miss the cross-silo value that agentic systems are designed to unlock.

MIT FutureTech — Rising Tide Task Automation: Workflow Implications (April 2026)

Source: research/01-ai-native-landscape/mit-futuretek-crashing-waves-rising-tides-ai-automation-2026.md · MIT FutureTech (Thompson et al.) · 17,000+ worker evaluations, 40+ models · April 2026 · HIGH · TIER 1

The MIT FutureTech task automation study provides the foundational evidence for why workflow redesign must be gradual and deliberate — AI capability is expanding as a broad “rising tide” across all task types, not targeting specific workflows for disruption.

~60% mean success rate on real professional tasks at minimally sufficient quality. The other 40% requires iteration, human review, or is beyond current AI reach — meaning effective workflow redesign must map precisely which sub-tasks are automatable vs. which require human judgment.
Task duration threshold expanding rapidly: 50% success rate on tasks taking up to one week of human time (as of Q3 2025), up from 3–4 hours eighteen months earlier. Workflow redesign today must account for the capability at 12-month horizon, not just current state.
The rising-tide pattern demands modular workflow design. Because AI improves broadly and continuously (not in targeted surges), workflow architects should decompose work into task-level modules that can be handed to AI at different transition points as capability crosses the success threshold for each module. A workflow redesigned for today’s AI success rate will need re-evaluation in 12–18 months.
Implication for pilot sequencing: workflows with 60%+ AI success on constituent tasks are ready for supervised human-on-the-loop deployment. Workflows with 30–50% success rates need augmentation-plus-review design. Below 30%, AI is best used for drafting and research support, not autonomous execution.

OpenAI B2B Signals — Depth vs. Volume: The Frontier Workflow Gap (May 2026)

Source: research/01-ai-native-landscape/openai-b2b-signals-enterprise-ai-depth-2026.md · OpenAI May 2026 · MEDIUM (vendor, OpenAI customer base only, no causal outcome data) · TIER 1

OpenAI’s B2B Signals (published May 6, 2026) provides the first usage-telemetry evidence for why the workflow-redesign gap creates a compounding performance differential — measured from the AI interaction side rather than the survey side.

Frontier firms (95th percentile) use 3.5x as much AI intelligence per worker as typical firms (50th percentile) — up from 2x one year ago. The gap is widening, not closing, as adoption spreads.
Volume explains only 36% of the frontier advantage. The remaining 64% is depth: task complexity, context richness, and output ambition. Firms that measure AI ROI by seat count or message volume are tracking 36% of the signal.
Agentic tools show the largest frontier gaps: frontier firms send 16x more Codex messages per worker. The frontier is built on delegation (agents completing multi-step work), not query-answering.
Education and learning use shows the largest task-level frontier gap — leading firms use AI to build capability, not just complete tasks. This is the workflow-redesign implication: the firms ahead are embedding AI into how work gets understood, not just how tasks get executed.
Consistent with BCG 5%, McKinsey 6%, Atlassian 6% ROI-confirmed cluster — all measuring the same concentration from different angles. B2B Signals now adds the usage-intensity dimension: the performance gap is visible in how deeply AI is integrated into workflow design, not in adoption rates.

AISI UK Government RCT — Task-Type Specificity as Workflow Redesign Constraint (February 2026)

Source: research/01-ai-native-landscape/aisi-uk-ai-productivity-rct-2026.md · UK AI Security Institute (AISI), DSIT · RCT n=500 · February 2, 2026 · HIGH / TIER 1

Government RCT using the O*NET Generalised Work Activities taxonomy to test AI productivity gains across four task types. The results provide the clearest task-level workflow-redesign map in the corpus, derived from a controlled experiment rather than company disclosures.

Monitoring and technical drafting (Tasks 1 & 2): +22–23% quality gain with no time change. These are bounded analytical tasks with clear right answers and verifiable outputs. Workflow redesign implication: integrate AI as a drafting and quality-check layer; human review remains but for exception handling rather than full review.
Interpreting information for others (Task 4): −42% time, +102% output per minute, no quality change. Pure throughput leverage. The workflow redesign implication is concrete: knowledge workers currently spending 20 hours/week synthesizing data for decision-makers can recover 8–10 hours. Redesign those roles around the recaptured capacity.
Open-ended strategic planning (Task 3): No statistically significant effect on quality, time, or throughput. Workflow redesign implication: do not embed AI as an autonomous actor in planning workflows — embed it as a structured-input synthesizer and leave the judgment step human.
The O*NET taxonomy provides a transferable audit framework. Organizations can classify their current workflows by O*NET task type and predict — from this and the MIT FutureTech dataset — which workflows are redesign-ready vs. which will show null returns on AI investment. The redesign priority sequence follows the gain profile: Task 4 first (throughput recovery), Tasks 1–2 (quality improvement), Task 3 last (augmentation only, not replacement).

Yale CELI Agentic Deployment Sequencing — Customer Proximity as Workflow Architecture Constraint (Apr–May 2026)

Source: research/01-ai-native-landscape/yale-celi-agentic-ai-enterprise-series-2026.md · Yale CELI (Sonnenfeld, Henriques, Griessel, Alam-Nist, Yu) · 12 sectors + public sector · Apr–May 2026 · HIGH (institutional) / MEDIUM (case data) · TIER 1

The Yale CELI six-month cross-sector study provides the most actionable workflow sequencing framework in the corpus: the decision of where to start agentic deployment is not about use-case ROI — it is about customer proximity, which determines governance architecture requirements before rollout.

Background operations (supply chain optimization, internal document processing, financial reconciliation): deploy now. These workflows have the most consistent ROI and the lowest governance overhead. C.H. Robinson: 29% volume increase with 30% fewer employees via agentic freight operations. EY.ai: 150 agents supporting 80,000 tax professionals processing 3M+ deliverables annually.
Mediated workflows (AI-assisted customer service with human escalation, AI-drafted communications with human sign-off): deploy deliberately. Governance architecture — reversibility mechanisms, audit trails, escalation protocols — must be built in before scaling, not retrofitted.
Direct-customer-facing deployments (autonomous customer service, AI-driven lending decisions, automated benefits determinations): proceed only with pre-built governance. CFPB complaints doubled post-ChatGPT. Gartner projects >40% of agentic projects will be canceled by 2027 — direct-customer deployments with inadequate governance are the primary driver.
The scaling constraint is data, not models. 80% of enterprises cite data limitations as the primary agentic obstacle; only 7% describe their data as “completely ready.” Workflow redesign for agentic deployment is blocked — not by workflow design skill — but by data infrastructure that cannot reliably supply agents with clean, structured context.
The sequencing implication: organizations that start with background operations bank early ROI, build governance competence on lower-stakes workflows, and use the resulting data infrastructure improvements to unlock mediated and eventually direct-customer deployments. Organizations that start with customer-facing AI before building governance and data foundations face the highest cancellation and compliance risk.

Bain Technology Report 2025 — The Micro-Productivity Trap and Five Transformation Actions

Bain’s sixth annual Technology Report provides two directly actionable workflow-redesign findings: a quantified productivity gap between tool-only and full-lifecycle redesign in software development, and a parallel finding in sales that names the structural mechanism behind the “minor productivity gains” most organizations report.

Software development: the SDLC composition problem. AI coding assistants address only 25–35% of the software development lifecycle — writing and testing code. The remaining 65–75% (requirements, design, architecture, review, deployment, maintenance) is unchanged by code completion tools. Bain’s finding: 10–15% productivity gain from code completion tools deployed on an unchanged SDLC; 25–30% from redesigning the full SDLC around AI capabilities. Goldman Sachs is cited as the production example: gen AI integrated into the internal development platform, fine-tuned on the internal codebase — full-lifecycle integration, not code completion overlay.

Sales: the fragmented day does not redesign itself. Sellers spend approximately 25% of their time actually selling. AI could raise that to 45%. Early successes show 30%+ improvement in win rates. Truly successful results remain rare. Bain’s diagnosis: a seller’s day is fragmented across dozens of tasks — CRM updates, internal coordination, meeting prep, follow-up. AI addresses individual tasks in isolation. The result is micro-productivity: individual tasks run faster, but the day is still fragmented. Bain’s prescription is explicit: top-down redesign of the full sales workflow, not bottom-up task-level automation. The pattern is structurally identical to the SDLC finding — tools applied to unchanged workflows capture the fraction of value that lives in individual tasks; redesigned workflows capture the fraction that lives in how tasks are sequenced and eliminated.

The five transformation actions (Bain, from client work):

Set ambitious goals based on top-down diagnostics, not pilots.
Charge general managers — not CIO or CTO — with meeting targets.
Redesign entire workflows, not siloed use cases.
Curate and clean data as needed for specific workflows, not holistically.
Make, buy, or partner at the workflow level, not enterprise-wide.

Action 2 deserves particular emphasis. The Bain finding that GM accountability — not technology-function accountability — drives transformation outcomes is corroborated by McKinsey’s State of AI 2025 finding that CEO-level governance ownership is the single attribute most correlated with EBIT impact at large companies. The workflow owner, not the technology owner, must carry the accountability target.

Source: research/04-consulting-firms/bain-technology-report-2025.md · Bain & Company Technology Report 2025 (6th Annual Edition) · MEDIUM-HIGH · TIER 2

EY-Parthenon Growth Strategy Survey 2026 — The Efficiency Trap as a Competitive Risk

EY-Parthenon’s April 2026 survey (n=271 US growth leaders, $500M+ companies) provides the clearest single-survey articulation of why efficiency-only AI deployment is a strategic trap, not a viable posture.

63% using AI for efficiency/productivity only. When the majority of the market is doing the same thing, efficiency gains commoditize — they become table stakes that competitors match, not advantages that compound.
14% using AI to stay ahead of competitors. This is the cohort building workflow redesign into competitive differentiation — new products faster, better pricing intelligence, superior customer understanding.
41% fear AI enables new competitive market entrants. The organizations using AI defensively are facing offensive pressure from AI-enabled entrants who don’t carry the same legacy constraints. Efficiency gains don’t address this threat.
The strategic trust gap: Only 34% trust AI for pricing, 28% for new product development, 27% for M&A evaluation. Workflow redesign requires deploying AI in high-stakes decisions — which requires building trust through validated deployment, not avoiding the decisions that matter.

The 63/14/7 distribution is the EY-Parthenon complement to the BCG 5%/McKinsey 6%/PwC 20% high-performer data from a different angle: it shows the strategic choice driving concentration, not just the outcome distribution.

Source: research/04-consulting-firms/ey-parthenon-growth-strategy-ai-2026.md · EY-Parthenon, n=271, April 2026 · MEDIUM-HIGH · TIER 1

KPMG Global AI Pulse Q1 2026 (n=2,110) — Cross-Functional Workflow Automation at Scale

The first quarterly global tracking survey of C-suite AI posture documents cross-functional workflow automation as the dominant agent deployment pattern:

73% of organizations are using AI agents to automate workflows that span multiple functions — not single-department tools. Cross-functional automation is the modal use case, not an advanced deployment pattern.
53% are using agents to route information or decisions between teams. When agents route decisions between teams, the workflow architecture has moved from parallel assistance to integrated coordination — which requires workflow redesign, not tool adoption.
51% are providing shared knowledge bases or unified dashboards via agents — creating a new information substrate that functions as enterprise memory.
The scaling gap paradox: organizations deploying agents into cross-functional workflows simultaneously report the highest difficulty scaling (65%, up from 33% one year earlier). Cross-functional deployment amplifies coordination complexity faster than organizations are building governance to manage it.
Workforce redesign is following: 55% are redesigning job roles as a result of agent deployment (alongside 87% upskilling and 68% recruiting for new roles). Role redesign is the structural workflow change that most organizations treat as downstream of deployment — but organizations that do it in parallel are the ones reporting business value.

Source: research/05-analyst-firms/kpmg-global-ai-pulse-q1-2026.md · KPMG, n=2,110, Feb–Mar 2026 · MEDIUM-HIGH · TIER 1

Forrester “Three Years Into GenAI” — AIQ Gap as Workflow Redesign Failure Mechanism (April 2026, n=1,500)

Source: research/05-analyst-firms/forrester-genai-enterprise-value-2026.md · Forrester, n=1,500 AI decision-makers, April 2026 · MEDIUM-HIGH / TIER 1

Three years into GenAI adoption, Forrester’s primary survey finds only 15% of enterprises report an EBITDA lift — and fewer than one-third can connect AI activity to any P&L change at all. The mechanism: AI is being layered onto existing processes (assistive deployment) rather than used to restructure them. Productivity micro-gains dissolve before reaching the income statement.

The AIQ gap is the workflow redesign problem named differently. Low AI fluency — Forrester’s “AIQ” — prevents organizations from identifying where AI can restructure value creation rather than accelerate task completion. Without that identification, workflow redesign never happens, and AI spend registers on IT budgets but not on financial statements.
48% of firms have cut headcount due to AI, but only 15% show EBITDA lift. Workforce reduction is running at 3× the financial return rate — the most concrete evidence in the 2026 corpus that companies are capturing cost savings without the workflow redesign required to translate those savings into measurable margin.
High adopters use CEO-driven strategy, customer-facing use cases, and structured talent development — not model sophistication or IT infrastructure. The differentiator is organizational design, not technology selection.
Cross-reference: BCG AI Radar 2026 (5% substantial gains), McKinsey State of AI Nov 2025 (6% EBIT impact), Deloitte 34% deep transformation — Forrester’s 15% EBITDA figure is the strictest financial threshold yet applied to the same adoption cohort.

Harness / Sapio Research 2026 — Measurement Redesign as Workflow Redesign

Source: research/01-ai-native-landscape/harness-engineering-excellence-2026.md · Harness/Sapio Research, n=700 engineering practitioners and managers, April 2026 · MEDIUM / TIER 1

Workflow redesign in software engineering is not only about how code gets written — it’s about how the engineering workflow gets measured. AI has changed the composition of developer work without triggering a corresponding update to measurement frameworks. The result is a feedback loop between incorrect metrics and incorrect investment decisions.

Only 6% of engineering leaders believe current frameworks can be fixed without fundamental redesign. The other 94% are either waiting for better tools or relying on frameworks they know are incomplete.
The invisible work category is the workflow redesign signal. When 31% of a developer’s day is consumed by work that dashboards don’t track (validating AI output, fixing AI-introduced bugs, managing context switching), the workflow has already been redesigned — by AI — and the measurement infrastructure hasn’t caught up. Workflow redesign from the executive level is catching up to changes that AI already made on the ground.
Developer involvement in metric design is the adoption prerequisite. 49% of developers want involvement in defining the metrics used to assess their work; 55% want improvement data separated from performance evaluation. Organizations that build evaluation frameworks without developer input will see behavior optimization toward the visible metric rather than toward organizational outcomes — the classic Goodhart’s Law failure.
Cross-reference: DORA Four Key Metrics (deployment frequency, lead time, change failure rate, MTTR) provide a workflow-complete measurement framework that predates AI but remains more comprehensive than AI-native dashboards tracking only generation velocity. Workflow redesign conversations should include a metrics redesign component.

Microsoft Work Trend Index 2026 — The Organizational Architecture Gap

Source: research/01-ai-native-landscape/microsoft-work-trend-index-2026.md · Microsoft/Edelman Data x Intelligence, n=20,000 AI users, 10 countries, May 5, 2026 · MEDIUM-HIGH / TIER 1

The most operationally useful finding in Microsoft’s 2026 dataset is not an adoption number — it is a decomposition of where AI value is actually blocked. Random forest permutation importance analysis across 20,000 AI users: organizational factors explain 67% of AI impact variance; individual factors explain 32%.

Workflow redesign is blocked at the reward layer. Only 13% of AI users have been rewarded for reinventing how they work, even when results didn’t immediately materialize. Organizations signal through compensation what they actually value — and most organizations are still signaling for output, not for the redesign that generates durable output.
The Blocked Agency cohort quantifies the cost of the organizational gap. 10% of AI users have high individual capability but low organizational readiness — employees who can redesign work but whose organizations have not structured conditions for that capability to translate into value. This group is where workflow redesign investment has a documented ROI: the tools and skills are present; the organizational permission is absent.
53% of Frontier Professionals pause before tasks to explicitly allocate AI vs. human work (vs. 33% non-Frontier). This deliberate task allocation behavior is the individual analog of workflow redesign — applied at the task level in real time. Organizations that train this behavior as a standard practice are building workflow redesign into the daily rhythm rather than reserving it for strategy exercises.
Leadership alignment is the upstream prerequisite. Only 26% of AI users say leadership is clearly aligned on AI. Without that alignment signal, individual workflow redesign efforts operate without organizational reinforcement — and the 13% reward rate reflects that misalignment downstream.

Agentic telemetry supplement (Pillar 13): The same WTI dataset cross-referenced with Microsoft 365 platform telemetry (March 2025–March 2026) shows active agents grew 15-fold YoY, and 49% of 105,000 sampled Copilot conversations involved cognitive work (analysis, problem-solving) vs. 15% information-retrieval — a meaningful shift toward higher-leverage task types that require workflow redesign to capture.

Source: research/13-multimodal-sources/microsoft-work-trend-index-2026-agentic-telemetry.md

Gartner I&O AI ROI Survey — The Workflow Embedding Imperative (n=782, Nov–Dec 2025)

Gartner’s infrastructure-level survey isolates the workflow redesign variable with precision not available in enterprise-wide studies.

The primary differentiator between the 28% that succeed and the 72% that don’t is workflow embedding. Organizations that integrate AI into existing operational systems and processes outperform those running standalone AI experiments — the same workflow redesign conclusion reached by BCG, PwC, and McKinsey from different angles.
Agent-led workflows and auto-remediation fail at the highest rates because they were scoped as replacements for human judgment in high-variability environments — without redesigning the surrounding process to provide the structure AI needs to function reliably.
ITSM succeeds because the workflow already provides structure. Large historical ticket datasets, defined outcome metrics, and human review as a natural workflow step are conditions that existed before AI arrived — AI was embedded into a designed process, not dropped into an ambiguous one.
The implication for executives: the question is not “which AI tool should we use” but “which of our workflows is already structured enough to receive AI without a full process redesign first.” Starting there produces the 28% success profile. Starting with ambition-first scoping produces the 20% failure profile.

Source: research/07-adoption-challenges/gartner-io-ai-roi-stall-2026.md · Gartner, n=782 I&O leaders, November–December 2025 · MEDIUM-HIGH / TIER 1

Celonis 2026 Process Optimization Report — The Process Infrastructure Prerequisite (n=1,649, June–July 2025)

Celonis’s third annual report (independent fieldwork by Insight Avenue, n=1,649, $500M+ revenue, five regions) documents the operational precondition for workflow redesign that most enterprise AI research bypasses: 76% of organizations don’t yet have processes worth redesigning.

73% describe their processes as “running well enough to get by” but with “significant room for improvement.” Only 24% say their processes need no improvement. This is the upstream constraint: you cannot redesign what you cannot see, and most organizations lack the operational visibility to know what they’re actually working with.
The most cited red flag of broken processes is difficulty adopting AI (20%) — above bottlenecks (19%), missed SLAs (15%), and rising costs (12%). AI failure is surfacing pre-existing process failures that were previously invisible.
52% start process improvement with tools that don’t produce machine-readable operational data — process mapping workshops (27%) or BI dashboards (25%). Neither delivers the contextual data AI agents need to navigate business logic. Only 5–6% start with process mining or intelligence platforms.
The context gap is quantified: 82% say AI can only deliver ROI if it has the context of how the business runs. 72% report that different departments have contradictory views of the same process. This multi-model fragmentation is what makes workflow redesign fail even when the individual components are sound.
Process/Ops leaders confirm the visibility problem: 69% say their processes can’t run freely across multiple systems. 68% step in regularly to bridge gaps between departments. These are the manual coordination costs that workflow redesign is supposed to eliminate — but they can’t be eliminated without first making them visible.

Source: research/07-adoption-challenges/celonis-2026-process-optimization-report.md · Celonis/Insight Avenue, n=1,649, June–July 2025 · MEDIUM-HIGH / TIER 2

Skill Discovery as Workflow Redesign at the Agent Layer — Enterprise Implications (May 2026)

The same three-layer logic that drives enterprise workflow redesign (capture tacit knowledge → codify processes → train staff) is now being implemented at the agent layer through skill discovery research (12 arXiv preprints, Jan–May 2026). This is not an analogy — it is the same pipeline applied to AI agents instead of human workers.

Layer 1 — Mine: MemRL (arXiv:2601.03192), SDPO (arXiv:2601.20802), and the distilllabs production case extract skills from raw interaction traces. The distilllabs result (327 noisy traces → Qwen3-1.7B beats 744B teacher, April 2026) demonstrates that this mining phase is deployable today with off-the-shelf tooling.
Layer 2 — Curate: SkillRL’s SkillBank (arXiv:2602.08234), SkillOS (arXiv:2605.06614, Google), and the SSL Representation (arXiv:2604.24026) architecture provide centralized, quality-gated skill libraries. SkillsBench establishes the stakes: curated skills add +16.2 percentage points; self-generated skills average −1.3pp net negative (Chandra-verified Table 10, n=7,308). Curation is not optional.
Layer 3 — Distill: SKILL0 (arXiv:2604.02268) and Skill1 (arXiv:2605.06130) internalize curated skills into model parameters, eliminating retrieval latency at inference. This is the enterprise “train the workforce” phase — except the workforce is a fine-tuned specialist model.
The critical workflow redesign insight: Enterprise AI deployments that provide structured skill context to agents consistently outperform those relying on raw prompting or RAG-only architectures. The SkillsBench +16.2pp finding for curated vs. self-generated skills mirrors the McKinsey / BCG finding that workflow-redesigned AI deployments outperform tool-only deployments by comparable margins. The mechanism is the same: structured procedural knowledge reduces ambiguity at execution time.
Microsoft Waza CLI (GitHub, May 2026) provides the operational toolchain to move from “writing skills” to “operating skills” at scale — 20+ commands for skill scaffolding, evaluation, benchmarking, and quality grading. Organizations ready to operationalize this layer should evaluate Waza against their existing MLOps stack.

Source: research/21-benchmarks/skill-discovery-continual-learning-2026.md · 12 arXiv preprints (Jan–May 2026) + Microsoft Waza + distilllabs · MEDIUM (preprints, unreviewed) · May 2026

Foxit / Sapio Research: The Verification Burden as a Workflow Design Failure (March 2026)

The most common reason AI programs fail to deliver productivity is not accuracy — it is an unredesigned verification step. Foxit and Sapio Research quantified the mechanism in a March 2026 survey of 1,400 US and UK workers.

Executives perceive 4.6 hrs/week saved; net gain after validation: 16 minutes
End users perceive 3.6 hrs/week saved; net outcome after review: −14 minutes
The “verification burden” — reviewing, fact-checking, correcting AI outputs — consumes nearly all time savings when organizations deploy AI without redesigning the checking step
Organizations that redesign verification (accuracy investment, explicit check-intensity policies by task risk, HITL scope definition) recover the time savings. Those that don’t remain at 16 minutes.

The verification burden is the mechanism behind multiple corpus findings: Workday’s 40% rework consumption, ActivTrak’s work-density increase without throughput gain, METR’s −19% experienced-developer result. All three measure the same phenomenon from different angles. The Foxit data is the only one that directly measures the pre/post time allocation that produces the net result.

Source: research/07-adoption-challenges/foxit-sapio-document-intelligence-ai-productivity-2026.md · n=1,400, US + UK, Sapio Research independent · MEDIUM-HIGH / TIER 1 · March 2026

BCG Henderson Institute: Six-Segment Job Architecture (April 2026)

Source: research/04-consulting-firms/bcg-ai-reshape-jobs-not-replace-2026.md

BCG’s 165M U.S. job analysis provides the clearest published taxonomy of what “workflow redesign” actually means at role level. The six segments define the type of redesign required:

Enabled (23%): AI adds tools to existing workflows — minimal redesign. The highest-risk segment for organizations that stop here and call it “AI deployment.”
Rebalanced (14%): Routine tasks automate; higher-complexity work expands. Active redesign required to redirect redirected capacity to higher-value work. Without deliberate redesign, the capacity freed by automation flows to additional low-value volume rather than value-generating work.
Divergent (12%): Two-tier role architecture emerges. Junior positions require explicit redesign around judgment and learning, not task completion. Senior positions gain capacity but need new scope definition.
Amplified (5%): Demand expands alongside capability — the target state for strategic roles. Only achievable through deliberate workflow redesign + market positioning, not just tool adoption.
The 23% “Enabled” category is where most organizations currently operate. The workflow redesign literature (BCG Widening AI Value Gap, McKinsey State of AI Nov 2025, HBR Analytic Services n=385) consistently finds this is where value stalls. Moving roles from Enabled → Rebalanced or Amplified requires active role architecture work, not additional tool rollout.

ADP Research — The Engagement-Productivity Inversion as a Workflow Redesign Signal (n=39,000, March 2026)

ADP’s 39,000-worker study surfaces a counterintuitive finding that directly maps to workflow redesign: daily AI users are simultaneously more engaged and more likely to feel less productive. This is the behavioral fingerprint of workflow redesign failure.

Daily AI users are 4x more likely to feel less productive than non-users, despite being twice as likely to be fully engaged. This inversion is the result of AI adding work (more tasks visible, more output possible) without redesigning what gets measured or how work is bounded.
The redesign implication: when AI expands task throughput without workflow boundaries, workers experience productivity loss even as engagement improves. The fix is not less AI — it is redefining output expectations and workload caps for AI-augmented roles. Without this, AI adoption produces the paradox the data documents.
6x engagement multiplier for job-secure workers confirms that psychological safety is a workflow redesign prerequisite: workers who trust the redesign (their job is not at risk from AI) deliver the engagement dividend; those who don’t stay in the paradox.
3.3x productivity multiplier for job-secure workers provides the quantified return on addressing the redesign prerequisite. Organizations that skip the job-security communication step lose 3.3x of the workflow redesign productivity gain.
50% of the global workforce uses AI multiple times weekly (ADP, 36 markets) — the scale at which unresolved workflow redesign failures compound into systemic productivity drag.

Source: research/07-adoption-challenges/adp-people-at-work-ai-productivity-paradox-2026.md · ADP Research, n=39,000+, 36 markets, March 2026 · HIGH / TIER 1

Genpact / HFS Research — The “Governable Autonomy” Framework for Agentic Workflow Redesign (n=545, April 2026)

As enterprises move from assistive AI to agentic AI, the workflow redesign challenge escalates. Genpact/HFS Research’s survey of 545 Fortune 2000 senior executives documents the specific redesign failure modes at the agentic stage:

33% cite unprepared business processes as the primary adoption obstacle — the agentic-era expression of the workflow-redesign-before-tooling principle. Organizations that bolt agents onto existing workflows without redesigning the end-to-end process are the 80% still requiring human final approval for every agent action.
HFS “governable autonomy” as a redesign framework: Four decisions determine whether agentic workflows scale: (1) who owns accountability when agents act without asking, (2) what metrics capture agent-native outcomes (not task productivity), (3) where human decision rights begin and end, (4) whether the process was designed for agents or retrofitted.
67% are measuring agentic AI with productivity metrics from earlier automation waves — the workflow measurement version of the redesign gap. Organizations cannot identify governance failures or optimization opportunities in agent workflows if they’re tracking task speed rather than downstream outcome accuracy.
The 17-month scale timeline is achievable specifically for organizations that start with process redesign. Those starting with tool deployment and retrofitting governance are building the 33% obstacle into their deployment architecture.
Top skill demand signals (42% workflow orchestration, 39% data engineering, 36% monitoring/observability) are all workflow architecture roles, not AI model roles. The redesign bottleneck is operational, not algorithmic.

Source: research/01-ai-native-landscape/genpact-hfs-autonomy-requires-trust-agentic-ai-2026.md · Genpact/HFS Research, n=545 Fortune 2000, April 2026 · MEDIUM-HIGH / TIER 1

Ranganathan & Ye (UC Berkeley Haas) — The Workload Intensification Mechanism (February 2026)

Eight-month ethnographic study at a 200-employee U.S. tech company. 40+ interviews across engineering, product, design, research, and operations. No AI mandate — enterprise subscriptions provided, voluntary adoption only. Published HBR, February 9, 2026. TIER 1.

The central finding inverts the workflow redesign premise: most organizations assume AI reduces work and frees capacity. This study documents the opposite mechanism — AI consistently expands work across three dimensions without any organizational directive:

Task expansion: AI lowered the cognitive cost of attempting unfamiliar work, causing workers to absorb scope that would previously have justified headcount. Product managers wrote code; researchers took on engineering tasks. Simultaneously, the engineers who now had to review and correct AI-assisted work from colleagues absorbed undocumented overhead — informally, in Slack and side-conversations.
Boundary blur: AI lowered the friction of starting any task so far that workers filled previously-recovery moments (lunch, transit, between-meeting windows) with prompts. Downtime stopped functioning as recovery. Work became ambient — always advanceable — without deliberate intention.
Multitasking inflation: AI enabled running multiple threads in parallel (coding manually while AI generates an alternative version; running agents in background; reviving deferred tasks). Workers felt momentum; the reality was continuous attention switching and a growing cognitive load even as perceived productivity was high.

The self-reinforcing cycle: AI accelerates tasks → speed expectations normalize → workers rely more on AI → reliance widens scope → wider scope expands total workload → workers accelerate further. Short-term: apparent productivity surge. Medium-term: workload creep, cognitive fatigue, decision quality decline, burnout.

The structural implication: Workflow redesign is not just a matter of configuring AI correctly — it must include explicit cessation norms. Without “AI practice” governance (structured pauses, sequencing rules, human connection rituals), the natural trajectory of voluntary AI adoption is workload intensification, not relief. An AI rollout plan that ends at deployment and leaves workers to self-regulate is not a complete plan.

Corroborates: Workday/Hanover (40% of time savings consumed by rework), Foxit/Sapio (net 16-min gain after verification), Suh & Oh Bank of Korea (efficiency captured as on-the-job leisure rather than output — but this study finds the opposite capture mechanism for knowledge workers who don’t reduce pace).

Source: research/07-adoption-challenges/hbr-ranganathan-ye-ai-intensifies-work-2026.md · Ranganathan & Ye, UC Berkeley Haas, HBR, Feb 2026 · MEDIUM-HIGH / TIER 1

Wang et al. “Agentic AI and Human-in-the-Loop Interventions” (n=647 workers / 680,676 chats, Taobao, May 2026) — Organizational Role Design for Agentic AI

Field experimental evidence that agentic AI workflow design choice — specialize vs. integrate human roles — produces measurably different outcomes:

The integration model (same workers supervise AI and handle full human chats) generates positive spillover: better performance on human-only cases (+0.091 ratings) because offloading AI-eligible volume frees attention for complex work.
The specialization model (dedicated AI supervisors) promises earlier intervention on emotionally sensitive cases, but risks skill erosion and emotional exhaustion as workers only encounter failed AI handoffs.
Neither model dominates on all dimensions. The choice depends on the distribution of escalation types — how frequently your AI triggers emotional vs. technical failures. This is currently unmeasured in most enterprise deployments.
Core workflow redesign requirement: Agentic AI in customer-facing roles requires real-time emotional state monitoring during AI handling — not just task completion failure detection. The workflow that routes late after frustration has accumulated produces worse outcomes than no AI at all for that subset of interactions.
The positive spillover case is the legitimate ROI argument: Agentic AI improves the quality of work humans do on complex cases by reducing cognitive load from routine cases — visible only if you measure the full human portfolio, not just AI-touched tasks.

Source: research/12-agent-workers/alibaba-agentic-hitl-customer-service-2026.md · Wang et al. · arXiv 2605.14830, May 2026 · HIGH / TIER 1

INSEAD/HBS Kim, Kim & Koning — The Mapping Problem (n=515 startups, RCT, March 2026)

The most rigorous evidence on why task-level AI gains don’t automatically become firm-level results. Randomized controlled trial, Harvard Business School + INSEAD. n=515 early-stage startups, 10-week experiment, March 2026.

The mapping problem is cognitive, not technical. Treatment effects showed no variation by founder engineering background or baseline performance. The bottleneck is knowing which workflows to reorganize — not tool access, not prompting skill.
Firms that solved it generated 1.9× higher revenue, 18% more paying customers, 12% more completed tasks, and needed $224,000 less external capital — with no increase in headcount.
Only firms that automated complete production chains captured structural gains. Revenue benefits concentrated in the top 10% of treated firms — those who rebuilt end-to-end workflows rather than adding AI to individual steps.
The enterprise implication: Current AI programs subsidize tools and provide prompting training. This addresses the wrong constraint. Workflow redesign must identify where across the production chain individual AI gains compound into structural results.
Corroborates the “bottleneck shifts” evidence from Faros (98% more PRs, zero delivery improvement) and Workday (85% save time, only 14% achieve net-positive outcomes) — the gains are real at the task level; capture requires structural workflow change.

Source: research/01-ai-native-landscape/insead-hbs-mapping-ai-production-rct-2026.md · Kim, Kim & Koning, INSEAD/HBS, SSRN 6513481, March 2026 · HIGH / TIER 1

Dillon et al. — Tool Access Without Institutional Change: The 7,137-Worker Null Result (Nov 2025)

6-month cross-firm RCT across 66 large enterprises, 7,137 knowledge workers. Microsoft Copilot, Sep 2023–Oct 2024 pilot period. Three of four authors: Microsoft Research.

Email time dropped 2 hours/week (−17%). Real, persistent, statistically significant. The efficiency gain is not in dispute.
No task composition shift. Same meetings, same documents, same email threads — workers used freed-up time as slack or to shorten the workday, not to take on new work.
Firm management explains adoption 2× more than individual behavior. Firm fixed effects account for nearly twice the variance in Copilot usage as individual pre-period work patterns. A 10-fold adoption gap (6% to 70% weekly usage) across firms cannot be explained by industry or job type.
Why email changed but meetings and documents didn’t: Email is solitary — individual workers can independently develop new habits. Meeting patterns and document ownership require coordinating with colleagues and agreeing on new norms. That coordination did not happen in a one-person-at-a-time license rollout.
The co-invention condition: Point estimates suggest teams where multiple members had Copilot saved 50% more email time, but these effects were not statistically significant at sample sizes available. The literature on technology diffusion (Brynjolfsson & Hitt 2000; Bresnahan et al. 1996) predicts larger gains only with complementary process innovations — which require team-level adoption and intentional workflow redesign, not individual access.
The enterprise prescription: Tool provision is necessary but not sufficient. Organizations that achieved 70% weekly adoption in this study did something different organizationally. That delta is the workflow redesign question — and it is not addressed by license purchases, prompting guides, or individual training alone.

Source: research/07-adoption-challenges/dillon-jaffe-shifting-work-patterns-copilot-rct-2025.md · Dillon, Jaffe, Immorlica & Stanton · arXiv 2504.11436v4, November 2025 · MEDIUM-HIGH / TIER 2

Ni et al. — Skill-Tier Segmentation as Workflow Redesign (November 2025)

4-week Alibaba Taobao RCT, 5,940 agents, 2.56M chats. Qwen-based GenAI assistant. Fudan/Zhejiang/Dartmouth Tuck academic team.

Broad deployment without workflow differentiation produces an averaging effect. Average gains are real; the distribution shows the gains come almost entirely from the bottom 40% of performers.
Top performers need a different interface, not the same one. Expert agents handle multiple concurrent chats with established workflow continuity. The AI adds verification burden that increases shift-away time by 26.3% for the top quintile — degrading response continuity and customer patience.
The redesign requirement: Not prompting training. Role differentiation — deploying AI as a lifting tool for lower-skill agents, and potentially redesigning the concurrent-chat workload structure for expert agents who don’t benefit from AI suggestions in their existing workflow.
Objective vs. subjective quality divergence. CSAT scores improved for most; actual resolution rates (retrial within 3 days) did not — and worsened for top performers. A workflow that shows green dashboards while quietly degrading expert service is a lagging-indicator failure mode.

Source: research/12-agent-workers/ni-wang-alibaba-genai-customer-service-rct-2025.md · Ni, Wang, Feng, Lu et al. · arXiv:2603.29888, November 2025 · HIGH / TIER 1

Leading AI from the Middle — The Manager/Frontline Translation Layer (May 2026)

Multi-source synthesis: BCG AI at Work 2025 (n=13,000), Gartner manager surveys (n=1,973 and n=2,986), McKinsey Global AI Survey 2025 (n=1,993), NBER w34836 (n=~6,000 firms). Published May 2026.

Manager adoption (78%) outpaces frontline (51%) by 27 points. The gap is not motivation — 65% of employees are excited about AI (Gartner, n=2,986). It is translation: someone must convert executive strategy into team-level habits.
The adoption multiplier role belongs to the VP/Director layer. The executives who set direction rarely have visibility into frontline friction; the ICs who execute lack authority to redesign workflows. Middle managers are the only layer with both proximity and authority.
Redesign authority is the prerequisite. Research confirms the same finding as Dillon et al. (n=7,137) and the Ni/Alibaba RCT (n=5,940): broad tool rollout without workflow changes shifts metrics without moving the bottleneck. The manager’s job is to identify the bottleneck and redesign around it, not just mandate adoption.
Practical prescriptions: identify one workflow the team runs weekly that fits AI well (structured, repetitive, clear output); pilot with 3-5 people; document what changed; then scale. Not a 90-day transformation program — a repeating 2-week cycle.

Source: research/01-ai-native-landscape/leading-ai-from-the-middle.md · Multi-source synthesis · MEDIUM-HIGH / TIER 1–2

Everyday AI Ep 760: The 5-Move Change Management Playbook — Workflow Redesign as Move 2 (April 2026)

Everyday AI Ep 760 (Jordan Wilson, April 21, 2026) synthesizes change management patterns from high-performing organizations into a five-move framework. Move 2 — rebuild AI-native SOPs from scratch — directly maps to the corpus’s workflow redesign finding:

McKinsey finding cited: workflow redesign showed the largest EBIT impact among 25 AI implementation attributes tested.
BCG cited: 70% of AI’s value derives from people and processes, not technology. This is consistent with the corpus-wide Stage 1 vs. Stage 3 outcome gap.
The framing: “Bolting AI onto legacy standard operating procedures just makes broken processes run faster, not better.” — Jordan Wilson.
Enterprise case: Moderna achieved 80% internal AI adoption via a 2,000-person weekly enablement forum (not one-time training). BBVA scaled from 250 senior leaders to 83% bank-wide weekly AI activity using the same approach.
Warning against “year-long pilots”: consistent with the corpus finding that quarterly assessment cycles are insufficient given model capability doubling rates.

Credibility note: MEDIUM as podcast synthesis source. The underlying McKinsey and BCG citations carry their own higher credibility; the Moderna and BBVA case studies are self-reported through the episode, not independently verified.

Source: research/13-multimodal-sources/everyday-ai/2026-05-22-ep755-780-enterprise-ai-mining.md · Everyday AI podcast, Jordan Wilson · Ep 760 (Apr 21, 2026) · MEDIUM (podcast) / TIER 1

Stanford Enterprise AI Playbook (n=51 cases, Apr 2026)

The most direct empirical evidence for the workflow-redesign-first principle comes from Stanford Digital Economy Lab’s structured study of 51 successful deployments across 41 organizations.

77% of the hardest challenges practitioners cited were invisible and intangible costs: change management, data quality, and process redesign. Not the model, not the compute.
61% of successful deployments had at least one prior failed attempt. First attempts consistently failed when teams applied AI to broken or unredesigned workflows, when projects were led by technical teams without business ownership, or when organizations assumed the model would fix problems that required redesigning the work itself.
“Technology wasn’t the bottleneck — organizational adoption was the failure point.” — Executive, Professional Services Company (Stanford DEL interview)
The escalation-based operating model (AI handles 80%+ autonomously, humans review only exceptions) delivered 71% median productivity gains, versus 30% for approval-based models. The difference is entirely organizational design, not technology.
Root cause analysis from the 61% with prior failures: 35% failed because the organization wasn’t ready to adopt; 27% because critical knowledge was never captured or stored; only 16% because the technology broke.

Source: research/04-consulting-firms/stanford-enterprise-ai-playbook-51-deployments-2026.md · Stanford Digital Economy Lab, Pereira/Graylin/Brynjolfsson · April 2026 · MEDIUM-HIGH / TIER 1

IBM IBV Strategic Ascent — The 75%+ Incremental Ceiling (n=810, Q3 2025)

Source: research/12-agent-workers/ibm-ibv-agentic-strategic-ascent-2025.md · IBM IBV, n=810 C-suite, 20 countries, Q3 2025 · MEDIUM / TIER 2

The IBM IBV “Strategic Ascent” report provides quantitative evidence for why workflow redesign is the variable separating incremental AI gains from net-new value:

75%+ of AI investments target process optimization, not new capability creation. This is the population that is encountering the workflow redesign ceiling — efficiency gains bounded by the existing process architecture.
32x performance advantage for organizations with high AI adoption vs. minimal adoption — a correlation finding consistent with the workflow-redesign-as-multiplier thesis across the corpus.
Three barriers preventing the redesign shift: data readiness (49%), trust deficits (46%), skills shortages (42%). None resolve by purchasing additional tooling without redesigning the workflows that consume the data and require the trust.
IBM’s companion May 2026 research (n=2,000+) puts a number on the redesign multiplier: when all six foundational capabilities (change management, AI governance, data governance, real-time integration, system integration, financial integration) are present, agentic workflow adoption is 5.4x more likely. Change management — the workflow-redesign enabler — is the highest-leverage single lever.

Supporting research

research/06-industry-verticals/kikuchi-innovation-tax-banking-genai-2026.md — University of Tokyo causal study (n=809 U.S. banks, arXiv:2602.02607): 428 bps ROE decline during AI implementation (Innovation Tax); smaller banks absorb 4× the hit; J-curve pattern confirms workflow redesign costs precede gains; cross-sectional comparisons are misleading without controlling for selection bias

PwC 2026 Digital Trends in Operations — The Horizontal Structure Gap (n=767, Jan–Feb 2026)

Source: research/07-adoption-challenges/pwc-digital-trends-operations-2026.md · PwC, n=767 US operations/supply chain leaders revenue ≥$100M, Jan–Feb 2026 · MEDIUM / TIER 1

Only 41% of surveyed operations leaders say their organizations currently operate with collaborative, horizontal structures — the organizational form that AI agents require to operate across workflows.
94% of siloed organizations expect to shift to horizontal, networked models. Expecting structural change and executing it against internal political resistance are measurably different — this gap is the workflow redesign problem stated at the organizational level.
83% believe AI agents will accelerate the breakdown of functional silos; only 37% are comfortable assigning AI agents to execute full end-to-end processes now. The 46-point gap between expectation and current deployment comfort reflects the workflow design gap: the workflows do not yet support autonomous cross-functional agents.
72% rank automating operations as a top-3 AI priority, yet only 37% are comfortable with the autonomous execution that automation requires. This produces copilot-mode deployments in agentic-mode environments — tools that advise but don’t act, in contexts designed to require action.
research/07-adoption-challenges/gallup-ai-workplace-adoption-workforce-q1-2026.md — Gallup Panel n=23,717 (Feb 2026, TIER 1): 50% personal usage vs. 41% organizational integration; 9-point gap is the shadow AI / ungoverned deployment footprint; productivity perception (65% improved) does not equal workflow transformation (~10% strongly agree fundamental change occurred)
research/01-ai-native-landscape/hbs-stanford-genai-wall-effect-expertise-transfer-2025.md — HBS + Stanford RCT, n=78 (2024, TIER 2): 63–74% time reduction is uniform across knowledge distances; quality gains diverge by domain judgment — workflow redesign must account for knowledge distance, not just tool access
research/07-adoption-challenges/cdi-public-sector-ai-adoption-index-2026.md — CDI/Public First n=3,335 public servants (Feb 2026, TIER 1): in high-guidance environments, 61% report measurable benefits from advanced AI vs. 17% in low-embedding orgs — the 3.6x benefit multiplier is entirely explained by organizational scaffolding, not tool access; workflow integration is the mechanism, not adoption rate

McKinsey State of Organizations 2026 — Workflow Redesign as the Pioneer Differentiator (n=10,018)

Source: McKinsey & Company, “The State of Organizations 2026.” n=10,018, 15 countries, 16 industries. Jun–Sep 2025 fieldwork, Mar 14, 2026. MEDIUM / TIER 1.

88% experimenting; 81% no bottom-line impact — the gap is explained by workflow redesign. “AI Pioneers” (23% of sample) redesigned workflows alongside tooling; the 81% layered AI onto existing processes.
Only 6% currently realizing full value from advanced technologies they already have — the constraint is organizational structure, not technology access.
Only 30% of organizations reallocate resources enterprise-wide based on strategic priorities (vs. historical patterns). Workflow redesign requires reallocation; the majority cannot execute it.
25% expect autonomous AI agents as autonomous teammates in the short term; those who redesign workflows first see the greatest returns — the sequencing finding from the largest survey in this area.

Source: research/04-consulting-firms/mckinsey-state-of-organizations-2026.md

Cisco / Sapio Research — Industrial AI Workflow Gaps (n=1,000+ OT decision-makers, Apr 2026)

Source: research/05-analyst-firms/cisco-state-of-industrial-ai-2026.md · Cisco/Sapio Research, n=1,000+, 19 countries, 21 industrial sectors, >$100M revenue, Apr 7 2026 · MEDIUM-HIGH / TIER 1

43% of industrial AI organizations report limited or no collaboration between IT and OT teams — the primary workflow redesign failure in physical operations. Industrial AI requires unified network, security, and process ownership; siloed IT/OT teams cannot execute the redesign.
61% live in production, only 20% at mature/scaled deployment — the 41-point gap traces directly to workflow integration failures, not technology limitations. Organizations that resolved IT/OT collaboration scaled; those that did not remain in early deployment.
The use cases leading scaled adoption — process automation, predictive maintenance, quality inspection, supply chain optimization — share a common profile: single-domain, structured inputs, defined success criteria. These are the workflow types that require the least redesign effort and produce the earliest ROI.
47% of organizations with limited IT/OT collaboration cite network instability as a top operational challenge — a direct consequence of failing to redesign the coordination workflow between the two functions.

WEF/Accenture Organizational Transformation — The 15% Who Actually Redesign (450 enterprises, Mar 2026)

Source: research/04-consulting-firms/wef-accenture-organizational-transformation-age-ai-2026.md · WEF white paper with Accenture, 450+ enterprise practitioners · MEDIUM-HIGH / TIER 1

The WEF practitioner synthesis of 450 leading adopters quantifies the workflow redesign gap from the practitioner side: only ~15% of organizations are using AI to fundamentally redesign how work is performed. The other 85% are deploying AI into existing workflows — generating “localized efficiency gains and proof of return” that don’t compound.

The conversion mechanism: Task-level productivity gains “have not consistently translated into enterprise or macroeconomic impact” without end-to-end workflow redesign. The evidence is consistent across the WEF corpus: Deloitte (34% pursuing deep transformation, n=3,235), Roland Berger (~10% capture consistent value, n=203), BCG (only 5% have seen substantial gains, 3rd AI at Work edition, n=10,635).
The common failure pattern: Organizations “expand pilots” rather than redesign operating models. More pilots → more localized wins → no EBIT impact. The conversion requires choosing a process, redesigning it completely, and assigning a named business owner (not an AI center of excellence) to the outcome.
Named proof points from crossing the threshold:
- Unilever AI-powered internal mobility marketplace: 70% cross-functional assignments, ~500,000 hours of capacity unlocked, 41% productivity improvement — without headcount reduction
- Yum China AI hiring platform: 89% of hiring needs across 16,000+ stores; manager turnover 9.7%→7.8% in one year
- Repsol agentic operations: 22 agents live across 38 use cases; agents gather inputs, run checks, draft outputs within guardrails; humans handle review, approval, exceptions
- Moderna: merged HR and IT under single Chief People and Digital Technology Officer — the organizational design signal that human-agent workflow management cannot be split across two functions

Cross-reference: research/04-consulting-firms/deloitte-state-of-ai-enterprise-2026.md, research/04-consulting-firms/roland-berger-profitless-prosperity-ai-2026.md

Holmgren et al. JAMA — AI Scribe Productivity and the Scheduling Workflow Gap (n=1.2M encounters, Jan 2026)

Source: research/06-industry-verticals/holmgren-jama-ambient-ai-scribe-physician-productivity-2026.md · JAMA Network Open, n=1,202,734 encounters / 1,565 physicians, UCSF Health, difference-in-differences · HIGH / TIER 1

The largest published study of AI scribe financial impact (n=1.2M encounters, UCSF Health) demonstrates the workflow redesign prerequisite in a clinical setting: the tool generates real productivity gains (+1.81 RVUs/week, +0.80 encounters/week), but capturing those gains requires deliberate scheduling redesign.

+1.81 additional RVUs per week per AI scribe adopter (P<.001) — approximately $3,044 additional annual revenue per physician at Medicare rates. The gain is real and statistically robust.
+0.80 additional encounters per week — but this gain requires that the freed documentation time be reallocated into additional scheduled appointments. An organization that deploys AI scribes without adjusting scheduling templates and staffing ratios will see documentation burden fall but appointment volume unchanged.
The mechanism is workflow-mediated. AI scribes reduce documentation time by an average of 5.27 minutes per encounter (AMA data). That time savings only converts to the +0.80 encounter gain if the care team reconfigures scheduling to absorb additional visits — a workflow redesign decision, not a technology decision.
This is the clearest quantified example in the corpus of the general principle: AI can free capacity, but capacity conversion requires intentional redesign. The organizations capturing the $3,044+ annual per-physician gain are those that redesigned their scheduling workflow. Those that did not will report reduced documentation burden with no revenue change.

WEF C&T Industry Community 2026 — Organizational Redesign as the Constraint

The WEF practitioner synthesis from 20+ CSOs (Salesforce, ServiceNow, SAP, Cisco, Workday, Telefónica, and others) adds a practitioner-level consensus finding: workflow redesign is not one factor among many — it is the differentiating variable.

“The same model, dropped into two different firms, can be either a demo or a transformation. The difference is organizational.” (Brynjolfsson foreword)
AI deployed as a bolt-on tool “sprinkled across existing tech stacks through co-pilots, productivity apps or closed software environments” does not generate value; real value comes from deploying AI in ways that work across software, apps, CRMs and ERPs, breaking down silos.
Deploying AI without redesigning business processes “ends up creating more work instead of less.”
One company’s unnamed finance tool failure illustrates the principle: LLM-powered revenue assistant launched → inaccurate answers due to LLM probabilistic output vs. deterministic financial data requirements → usage declined → tool sunset. The failure was a workflow/use-case mismatch, not a technology failure.
Mid-career management roles face structural pressure when workflow redesign removes the coordination functions that historically justified those layers — a non-obvious implication of organizational redesign at scale.

Source: research/07-adoption-challenges/wef-ct-ai-work-productivity-hacks-transformation-2026.md

MIT Sloan MR — Otis et al. RCT (n=640, Apr 2026) — Workflow Redesign as Judgment Scaffolding

The Otis et al. RCT provides the mechanistic evidence for why undifferentiated AI rollout without workflow redesign amplifies existing performance gaps rather than closing them:

Top 50% performers at baseline: +15% revenue/profit with AI access. Bottom 50%: −10%. Average effect: ~0%, not statistically significant. Same AI, same recommendations — divergent outcomes determined entirely by judgment differentials in how workers filtered and applied the advice.
The enterprise implication: a company-wide rollout without workflow redesign defaults to an experiment where high performers get better and weak performers get worse. Usage-rate metrics will not detect this.
Workflow redesign that builds in judgment scaffolding — guardrails, review checkpoints, escalation protocols differentiated by demonstrated capability level — is what converts an amplification machine into a leveling mechanism.
This is the same mechanism the Stanford Enterprise AI Playbook identifies at the firm level: organizations handing AI 80%+ of a workflow while humans handle exceptions achieve 71% median productivity gain vs. 30% for human-primary deployments. The redesign step is what determines who owns exceptions and what judgment is still required.

Source: research/07-adoption-challenges/mit-sloan-ai-judgment-gap-rct-2026.md — HIGH / TIER 1

Section AI Proficiency Report (Jan 2026, n=5,000) — Use Case Quality as the Redesign Signal

Section’s analysis of 4,500 submitted AI use cases provides the clearest evidence in the corpus that adoption without redesign defaults to low-value use cases:

59% of enterprise AI use cases are basic task assistance (summarizing, reformatting, drafting). These substitute one input for another without changing how work flows.
Only 2% of submitted use cases are advanced applications: task automation, workflow restructuring, decision augmentation — the categories that require redesign to implement but deliver the largest ROI.
Only 15% of submitted use cases likely generate employer-level returns. The other 85% generate employee convenience at best.
Automation of multi-step tasks and processes — the workflow redesign category — ranks near the bottom of reported adoption frequency despite being the highest-ROI category.
Training programs built around prompting syntax produce a 40/100 proficiency score. Use-case-specific coaching that teaches workers which business problems in their role are amenable to AI redesign is what distinguishes experimenters from practitioners.

The pattern is structurally identical to what the WEF C&T synthesis, the Celonis Process Gap report, and the IMF 2026 analysis all identify: AI tools are being added to existing workflows rather than used to redesign them. The tool is present; the redesign is not.

BCG / Boston University — The “Digital Employee” Framing Trap (n=1,200+, May 2026)

Source: research/12-agent-workers/bcg-hbr-ai-agents-not-employees-2026.md · BCG Henderson Institute + Boston University Questrom, n=1,200+ managers, May 2026 · MEDIUM-HIGH / TIER 1 (RCT design; vendor commercial interest disclosed)

A randomized experiment with 1,200+ managers delivers the clearest evidence in the corpus that how an organization describes AI agents to employees determines the quality of human oversight — with direct consequences for workflow governance:

Managers who received AI assistance framed as an “employee” identified 18% fewer errors in AI output than managers who received identical assistance framed as a “tool.” Workflow redesign without redesigning the framing degrades the human review quality that justifies the redesign.
Individual accountability for errors dropped 9 percentage points in the employee-framing condition — accountability migrated to the AI, which cannot be sanctioned. Redesigned workflows require explicit accountability routing as a design artifact, not an assumption.
Adoption intent did not improve under employee framing. The only plausible justification for accepting the governance cost did not materialize. Framing-as-colleague is not a change-management strategy; it is a liability posture.
The architectural fix: treat agents as contractors with narrow statements of work — scoped permissions, audit logs, explicit kill switches. This is the structural equivalent of maintaining “tool” framing at the process level, not the culture level.

Source: research/07-adoption-challenges/section-ai-proficiency-report-2026.md — MEDIUM / TIER 1

Zapier / Centiment — Agentic AI Deployment in Production (n=525, Oct 2025) — Where Workflow Redesign Is and Isn’t Happening

Source: research/12-agent-workers/zapier-centiment-enterprise-ai-agents-adoption-2026.md · Zapier / Centiment, n=525 U.S. C-Suite executives at 1,000+ employee companies, October 2025 · MEDIUM-HIGH / TIER 1 (vendor-commissioned; Zapier commercial interest disclosed)

The Zapier/Centiment survey provides deployment-domain data that maps directly onto which workflow types have and haven’t been redesigned for agents:

Customer support (49%) and operations (47%) lead agent deployment — both high-volume, structured-decision environments with well-defined success criteria. These are workflows where redesign is relatively straightforward because inputs, outputs, and error costs are already documented.
Finance (24%) and sales (26%) lag significantly. Higher cost-of-error, complex data dependencies, and regulatory oversight requirements make workflow redesign harder — not the technology, but the redesign investment.
The use-case concentration (data entry/extraction 47%, document summarization 41%, report generation 36%) reveals what “workflow redesign” actually looks like at the median enterprise: structured repetitive tasks with verifiable outputs. The high-judgment redesigns (diagnosis, deal structuring, risk assessment) are not yet captured in this data.
Only 38% use human-in-the-loop oversight with approval gates. Workflow redesign without explicit accountability architecture produces deployment without governance — not the same thing.

SLM Data Curation Pipeline — Where Curation Sits in the Fine-Tuning Workflow

Source: research/21-benchmarks/slm-data-curation-pipeline-2026.md · GitHub primary data + NVIDIA/academic sources, May 2026 · TIER 1/2

The SLM curation research identifies a structural gap in the enterprise AI fine-tuning workflow: data curation — deciding which examples are worth training on — is the highest-leverage step but the least-tooled in open-source infrastructure.

SkillsBench (arXiv:2602.12670, n=7,308): curated training data produces +16.2pp improvement over baseline; self-generated/uncurated data averages −1.3pp (net negative). The workflow decision of whether to curate determines the sign of the outcome.
The OSS ecosystem has a structural gap: pre-training scale curation (NeMo Curator, DataTrove) and human annotation UI (Argilla, Label Studio) exist, but automated quality scoring of instruction-response pairs for SFT lives only in academic repos (DEITA, Cherry_LLM) and Snorkel Flow. distilabel is the practical OSS bridge closest to closing this gap.
Enterprises doing custom fine-tuning without a curation step are spending compute on data that actively degrades model performance — a workflow sequencing error, not a model quality problem.
Snorkel AI’s $135M raise (Andreessen Horowitz, March 2026) and Cleanlab’s $30M raise (Jun 2025) validate that the curation gap is commercially significant.

See also: wiki/enterprise-slm-specialization.md