← Knowledge Base 🕐 38 min read
Knowledge Base

AI Adoption and Scaling

· March 2026

The operational discipline of moving AI from successful pilot to embedded organizational practice — and then from emb...

The operational discipline of moving AI from successful pilot to embedded organizational practice — and then from embedded practice in one domain to compounding capability across the enterprise. Covers the change architecture for making adoption stick, the sequencing logic for cross-domain scaling, the KPI discipline required to distinguish real impact from activity, and the midstream adjustment operating rhythm that sustains performance after launch.

Rewired (Lamarre, Smaje, Levin, 2nd ed., 2024, Ch. 30–33) names Adoption and Scaling as Capability 5 of the six — the point at which strategy, talent, operating model, and data investments either convert into business outcomes or stall. The book’s central argument: “adoption is engineered, not wished for” (Ch. 30, p.457). That sentence captures the failure pattern in most enterprise AI programs: adoption is treated as a consequence of good technology, when it is actually a designed organizational output requiring dedicated roles, explicit incentives, and ongoing management.

Why this matters to Fortune 500 CIOs

  • Surface-level adoption is the norm. Deloitte (n=3,235, 2025) finds 37% of enterprise AI deployments achieve surface-level adoption — employees who have access to the tool but are not using it in ways that change their output. The gap between tool access and workflow-embedded usage is the largest single source of unrealized ROI in enterprise AI programs.
  • The pilot-to-production transition is where most enterprise AI value is lost. MIT CISR (n=721, 2022 Future Ready Survey + n=152, 2025) finds organizations in Stage 2 (building pilots) underperform their industries financially, while Stage 3 organizations (scaled AI ways of working) outperform. The Stage 2-to-3 transition is an adoption and scaling problem, not a technology problem. See ai-maturity-models for the full MIT CISR framework.
  • Scaling is different from launching again. Most Fortune 500 programs treat each new domain as a fresh launch — new team, new tooling, new stakeholder management. The organizations in the top quartile treat scaling as reuse: the data pipeline, evaluation framework, change management templates, and governance artifacts built in Domain 1 become shared infrastructure for Domain 2. Each domain becomes cheaper and faster than the last. This is the “reuse case” economic mechanism Rewired (Ch. 31, p.469) articulates.
  • Change capacity is finite and measurable. Prosci (n=1,107, 2025) finds 73% of organizations are at or near change saturation — the point where additional simultaneous change initiatives produce net-negative absorption. Fortune 500 organizations running 8–12 concurrent AI deployments across departments are statistically guaranteed to be exceeding their organizational change capacity. The sequencing decision — which domains in which order — is a change capacity constraint as much as a technology constraint.

Making adoption stick (Rewired Ch. 30)

Rewired’s “make adoption stick” chapter (Ch. 30, p.457) argues that the adoption engineering problem has four components, none of which are automated by deploying a good AI system:

1. Define the job before deploying the agent. Treating an AI system like a new employee — write the job description, set performance criteria, onboard carefully, evaluate systematically — is the paraphrasable version of the evaluation discipline Rewired attributes to E.ON Next (Ch. 5, pp.79–80: +6pp CSAT, −8% AHT, +14% transaction success, ~50% cost-per-call reduction). E.ON’s agent maintained those gains because the team had defined what “doing the job well” meant before launch, not after.

2. Make the benefit visible to users before asking for the behavior change. BCG AI at Work 2025 (n=10,600) finds that individual-level benefit visibility — not enterprise-level ROI metrics — is the strongest predictor of frontline adoption. Employees who can point to a specific workflow task that the AI makes measurably easier adopt at 3x the rate of employees who have received a general message about AI productivity.

3. Design the incentive structure before launch. Writer / Workplace Intelligence (n=1,600 US executives and knowledge workers) finds 31% of workers admit active sabotage of AI rollouts — rising to 41% among Gen Z/Millennials. The root cause in most documented cases is not resistance to AI per se but the absence of a visible personal upside for the employee. If the AI makes a worker’s job easier but the time savings go to headcount reduction rather than skill expansion and career development, the rational response is sabotage.

4. Measure leading indicators, not just lagging ones. Adoption programs that measure only usage statistics (logins, queries per day) miss the distinction between surface-level usage and workflow-embedded usage. The leading indicators of genuine adoption are: (a) self-reported workflow change by the direct manager, (b) quality of feedback users provide to the eval loop (specific, actionable, domain-relevant), and © voluntary expansion — users who start using the AI on tasks beyond the initial deployment scope. These are measurable within 30–60 days; revenue impact takes 90–180 days.

The reuse scaling model (Rewired Ch. 31)

Rewired’s “design for scale and reuse from day one” chapter (Ch. 31, p.469) introduces the economic principle that separates enterprise AI programs that compound from those that restart with each deployment: the best use case is the reuse case.

In practice, the reuse model operates through four shared assets:

Data infrastructure. The data pipeline built for Domain 1 (ingestion, normalization, quality monitoring) is the starting point for Domain 2 — not rebuilt from scratch. A customer-360 data product built for a customer acquisition AI is reusable for customer retention AI, churn prediction, and claims triage. See data-products-reuse for the architecture that enables this.

Evaluation framework. The ground-truth test sets, scoring criteria, and regression benchmarks built for the first deployed workflow become the template for all subsequent workflows — adapted, not recreated. Organizations that invest in evaluation infrastructure for Domain 1 deploy Domain 2 at 30–50% of the evaluation setup cost. Those that skip evaluation infrastructure for Domain 1 pay full price every time.

Change management templates. The resistance patterns, communication materials, manager briefing kit, and WITFM framing developed for Domain 1’s workforce are largely reusable for Domain 2. The specific workflow and role impacts differ; the psychological and organizational dynamics are highly consistent. Prosci’s ADKAR framework adapted for AI (Awareness, Desire, Knowledge, Ability, Reinforcement) is a reusable template that accelerates each subsequent deployment.

Governance artifacts. The AI acceptable use policy, risk classification, escalation paths, and review cadences developed for Domain 1 become the governance baseline for Domain 2. Organizations with a functioning AI center of excellence (see ai-center-of-excellence) manage this library centrally so pods can adapt rather than build.

BCG (2025) documents that top-quartile AI performers redeploy the outputs of high-performing workflows as inputs to adjacent ones. The median gap between first and fifth deployment cost (as a percentage of Domain 1 build cost) is 62% for organizations with reuse infrastructure versus 95% for those without.

KPI discipline: tracking what matters (Rewired Ch. 32)

Rewired (Ch. 32, p.483) makes the KPI argument bluntly: AI programs that measure activity (pilots launched, tools deployed, users onboarded) produce activity. AI programs that measure business outcomes (cost per unit, cycle time, error rate, revenue per workflow step) produce business outcomes.

The three-tier measurement architecture that aligns with both Rewired’s framework and the independent corpus evidence:

Tier What is measured Update frequency Audience
1 — System health Task success rate, error rate, latency, hallucination flags, retrieval accuracy Daily / automated Engineering pod
2 — Workflow performance Cycle time, throughput, quality defect rate, escalation rate Weekly Domain owner + pod
3 — Business impact P&L line items: cost per unit, revenue per workflow, CSAT/NPS delta, headcount-to-output ratio Monthly CEO / board

The critical governance discipline is that Tier 3 numbers are the authority. An AI workflow that produces excellent Tier 1 metrics (fast, accurate, low error rate) but no Tier 3 movement is still a failed deployment from the CEO’s perspective. MIT CISR’s finding that Stage 1 organizations (tool deployment without workflow redesign) underperform their industries financially despite high tool usage rates is the empirical backing for this hierarchy.

Midstream adjustment operating rhythm (Rewired Ch. 33)

Rewired’s acknowledgment that “the first plan will be wrong” (Ch. 33, p.501) is the most practically useful admission in the book for Fortune 500 program leaders. The operating discipline is not plan perfection — it is a structured re-planning rhythm that detects and corrects before errors compound.

The four-week pod cycle (see ai-delivery-pods) is the unit of the re-planning rhythm:

  • Week 1: Review Tier 1 and Tier 2 metrics. Identify the highest-priority adjustment opportunity — not a list, a single priority.
  • Week 2: Build and test the adjustment in staging. Domain owner reviews output sample, not just metrics.
  • Week 3: Deploy to production. Monitor for regression.
  • Week 4: Prepare the domain owner briefing. Identify the next quarter’s expansion candidate within the domain.

The discipline that most programs lack is the Week 4 step — the forward-looking domain expansion identification. Without it, the pod optimizes indefinitely within its current scope and never builds the “next deployment is cheaper” momentum that makes the reuse model work.

Sequencing across domains

The order in which domains are selected for scaling is not arbitrary. The sequencing variables that determine adoption and scaling speed:

Change capacity budget. Prosci’s saturation research (n=1,107, 2025) quantifies each organization’s practical change capacity. A mid-market organization running a full digital transformation alongside AI deployment has consumed most of its change capacity before the first AI pod launches. The sequencing decision must account for what else the workforce is absorbing simultaneously.

Data infrastructure dependencies. Domains that share underlying data products should be sequenced adjacent to each other — not because of workflow similarity, but because the data infrastructure built for Domain 1 accelerates Domain 2 when they draw from the same underlying sources. A supply chain AI and a demand planning AI share logistics and inventory data; sequencing them adjacent extracts the reuse benefit.

Organizational readiness gradient. BCG’s adoption research identifies that departments with high manager AI champion density and existing digital workflow fluency absorb AI adoption 3–4x faster than departments without. Sequencing to the ready departments first builds the visible wins that fund and politically support subsequent deployments.

Pilot-to-production conversion rate. Stanford Enterprise AI Playbook 2026 documents that organizations with defined domain pod structures convert pilots to production at 3x the rate of those without. The sequencing implication: before expanding to a new domain, ensure the current domain has a production-grade operating pod — not just a pilot team.

Practitioner voices (pillar 13)

“My fundamental message to our teams is we’re going to rewrite the org charts. We are going to rewrite how work gets done.”

— Steve Chase, Vice Chair AI & Digital Innovation, KPMG US · April 2026 · research/13-multimodal-sources/enterprise-ai-innovators/2026-04-14-bold-fast-responsible-workflows-with-kpmg-us-vice-chair-ai-d.md

“Put yourself in your employees’ shoes and ask yourself, ‘What’s in it for them?’ […] They’re smart employees, by the way. They want to do it, they have the passion, they have the energy, but after 3 months that energy drops because they don’t see any appreciation.”

— Dr. Sam Zolfagharian, President and Co-founder, Jaytech · April 2026 · research/13-multimodal-sources/ai-for-the-c-suite/2026-04-14-dr-sam-zolfagharian-how-leaders-should-actually-approach-ai-.md

“The actual long-term bottleneck for driving maximum value from this technology was not going to be about the model. It was going to be about how the technology connects into the technology estate and data and process estate of an enterprise.”

— Derek Waldron, Chief Analytics Officer, JPMorgan Chase · April 2026 · research/13-multimodal-sources/beyond-the-pilot/2026-04-13-what-30k-jpmorgan-ai-agents-taught-me.md

Waldron’s framing — that the bottleneck is integration into the data and process estate, not model capability — is the operational translation of the reuse scaling model. JPMorgan moved from 0 to 250,000 users by solving the infrastructure problem once, not by re-solving it for each deployment.

“Last year most of our effort was about adoption — I just want to make sure I’ve got 90-plus percent of my org using these tools every day. Then we shifted: we want to start thinking about not just using the tools but HOW we’re using them. For software engineering that meant shipping code with AI, making PRs with AI independently rather than having humans curate every step. One of our main OKRs through the middle of the year is to be API first in our engineering teams.”

— Joel Ron, CTO, Thomson Reuters · May 2026 · research/13-multimodal-sources/enterprise-ai-innovators/2026-05-18-joel-ron-thomson-reuters-cto-on-ai-for-legal-and-tax-profes.md

Ron describes the adoption maturity shift that most mid-market programs miss: Year 1 is about breadth (90%+ of org using tools daily); Year 2 is about depth (how they use them, measured in OKRs like API-first or >50% of PRs shipped by AI independently). Organizations that compress both years into one program deploy agents without the usage habits and process infrastructure that make depth possible. Thomson Reuters — 5,000 engineers, 1 million CoCounsel users, $7B+ revenue — is the largest named F500 example of this two-year sequencing in the 2026 corpus.

Common failure modes

Launch-and-forget. The most widespread failure mode: deploy the AI, brief the team, move on to the next domain. No adoption owner, no feedback loop, no midstream adjustment rhythm. Deloitte’s 37% surface-level adoption rate is the statistical outcome of the launch-and-forget pattern at scale.

Scaling without infrastructure. Launching Domain 2 before Domain 1 has a production-grade data pipeline, evaluation framework, and change management template library. Each subsequent domain rebuilds from scratch — no compounding, no reuse benefit. The scaling program looks like parallel programs, not a scaling program.

KPI disconnect. Measuring Tier 1 and Tier 2 metrics without connecting them to Tier 3 business outcomes. An AI program director who reports “94% task success rate across 2,000 daily queries” to the CFO who is looking for EBIT impact has failed to communicate business value regardless of how good the system performance is.

Change saturation blindness. Launching new AI domains into departments already at change saturation. The symptom is not resistance — it is quiet non-compliance. The workforce continues using old workflows in parallel with new ones, producing the dual-process overhead that Prosci’s research identifies as the leading cause of AI program abandonment after the initial funding cycle.

IBM IBV CEO Study 2026

The IBM 2026 CEO Study (n=2,000, 33 geographies, Oxford Economics, Feb–Apr 2026) adds the most current large-sample data on the gap between access and actual usage:

  • Only 25% of employees use AI regularly, despite 86% of senior leaders believing their workforce already has the necessary AI collaboration skills — a 61-point perception gap.
  • 83% of CEOs say AI success depends more on people’s adoption than on the technology itself. The data is now explicit at the CEO level: adoption management is the constraint, not tool capability.
  • Organizations that redesigned five core business areas (technology, finance, HR, operations, cross-functional collaboration) are 4x more likely to achieve business objectives — the 4x multiplier comes from full functional redesign, not incremental deployment.
  • 29% of employees are expected to need reskilling for different roles by 2028; 53% need upskilling for current roles. Combined, 82% of the workforce requires structured capability development within 24 months — a planning horizon that is already open.

The perception gap is the operational problem. Senior leaders who believe 86% of their workforce is AI-ready are not investing in adoption management because they believe the problem is already solved. The 25% actual usage rate is the correction. The question that closes the gap: what percentage of your direct reports’ teams used AI to complete a deliverable in the last 30 days?

Source: research/04-consulting-firms/ibm-ibv-ceo-study-2026.md — IBM IBV CEO Study 2026, MEDIUM-HIGH credibility, Oxford Economics partnership

Ramp AI Index: Paid Adoption Data from 50,000+ U.S. Businesses (May 2026)

Transaction data avoids the self-report bias in most enterprise AI surveys. Ramp’s corporate card and bill-pay platform tracks what companies actually pay for — not what executives say they use.

  • 50.6% of U.S. businesses now pay for at least one AI tool as of April 2026. Adoption crossed 50% for the first time in March 2026. (Ramp AI Index, 50,000+ businesses, April 2026)
  • Anthropic surpassed OpenAI in business adoption for the first time: 34.4% of businesses pay for Anthropic (up 3.8 percentage points in one month); 32.3% pay for OpenAI (down 2.9 points). Anthropic’s business adoption quadrupled year-over-year; OpenAI grew 0.3%.
  • Vendor relationships are not sticky. The Anthropic/OpenAI shift — a substantial market lead erased in one year — is the strongest empirical signal that enterprise AI vendor lock-in functions differently than traditional software. Quarterly vendor reviews are operationally justified.
  • AI infrastructure spending doubled from Q3 to Q4 2025 ($130M → $260M). Foundation model spending grew 126% in the same period ($190M → $430M). Both are accelerating.
  • The infrastructure layer is concentrated in a small cohort. Only 1,900 businesses (vs. 21,500 foundation model customers) spend on self-hosted or managed AI infrastructure — at an average of $142K/quarter. These are organizations that have moved beyond API subscriptions into infrastructure ownership, typically for cost or data-privacy reasons.
  • This dataset uniquely represents the mid-market. Most enterprise AI surveys sample Fortune 500. Ramp’s base skews toward companies with 10 to 2,000 employees — the population that Brandon’s workshop audience occupies.

The gap between Ramp’s 50.6% paid-adoption figure and NBER/Bloom’s 69% self-reported adoption (n~6,000 executives, February 2026) is methodological, not contradictory. Self-report captures free tools and lighter usage; Ramp captures only paid subscriptions. Both figures are accurate for their measurement definition.

Source: research/01-ai-native-landscape/ramp-ai-index-enterprise-adoption-may-2026.md — Ramp AI Index, MEDIUM-HIGH credibility, transaction data from 50,000+ U.S. businesses, April 2026

  • ai-change-management — the structured discipline behind adoption engineering; ADKAR adapted for AI; resistance patterns; sabotage data; manager-champion effect
  • ai-delivery-pods — the operating unit that runs the four-week adoption rhythm; pod composition; domain owner authority
  • ai-pilot-to-production — the transition mechanics from pilot to production pod; handoff failure modes
  • ai-use-case-selection — the domain selection framework upstream of adoption and scaling; domain concentration principle
  • data-products-reuse — the reuse infrastructure that makes each domain cheaper and faster than the last
  • ai-maturity-models — MIT CISR Stage 2-to-3 transition; Stage 3 characteristics that define the scaling target state
  • ai-roi-evidence — the business outcome evidence base; NBER 2026, MIT CISR, BCG AI at Work 2025
  • workflow-redesign — the workflow architecture investment that enables the 71% productivity gain (Stanford HAI 2026) vs. 30% for tool-only adoption

Supporting research

Census Bureau BTOS AI Supplement — Microstructure of AI Diffusion (2026)

  • 18% of U.S. firms used AI in a business function by November 2025–January 2026; on an employment-weighted basis the rate reaches 32% — large employers are significantly ahead of small firms
  • Expected adoption reaches 22% within six months of the survey period; large firms in Information, Professional Services, and Finance show 50–60% firm adoption and 60–70% employment-weighted adoption
  • Most adoption remains narrow: 57% of adopters use AI in three or fewer business functions; 65% of firms with worker-level task use restrict that use to three or fewer tasks
  • Only 2% of firms report AI-related employment decreases — task augmentation (66% of task-using firms) is the dominant operating model, not displacement
  • Worker-level use: 23% of firms (41% employment-weighted) had workers using AI for individual tasks; leading uses are writing, document analysis, and information search

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

NBER — The US/EU Adoption Gap and the Management Variable (March 2026)

NBER Working Paper 34995 (multi-institution; primary worker and firm surveys across US + 6 European countries + EU 27-country firm survey, 2025–2026) quantifies the management variable’s role in adoption scaling:

  • 43% US worker AI adoption vs. 32% European average — the 11-point gap is not technology access, demographics, or firm size; it is whether management actively encourages AI use
  • At firm level: 7% US vs. 4% EU firms use AI in production; firm-level adoption ranges from 9% (Sweden) to 1% (Serbia) across 27 EU countries
  • Industry-level AI adoption correlates with faster productivity growth in both regions — the productivity dividend of scaling is real and measurable (US workers save 2.3% of working hours; European workers save 1.4%)
  • No measurable employment displacement in either region — consistent with Census BTOS augmentation-over-displacement finding

The management-as-lever finding is the cross-national validation of BCG’s 5% (n=10,635) and McKinsey’s 6% (n=1,993) full-value-capture rates: the differentiator is not tooling but organizational will and management execution. The nber-mind-the-gap study gives that principle a national-level causal foundation.

Source: research/07-adoption-challenges/nber-mind-the-gap-us-europe-ai-adoption-2026.md — TIER 1 (March 2026), HIGH credibility

Infor Enterprise AI Adoption Impact Index — The Confidence-Execution Gap (April 2026)

Vendor-commissioned survey (n=1,000, US/UK/Germany/France, manufacturing-adjacent industries) documenting the 80/49 gap: 80% of decision-makers believe their organization is capable of scaling AI, but 49% remain in pilots, paused, or not yet started. Key barriers named:

  • 36% cite data security, sovereignty, privacy, or compliance as the primary scaling barrier
  • 25% cite lack of internal AI talent for configuration and maintenance
  • 23% cite unclear business benefits or ROI
  • 27% unsure or disagree that data is mature enough for reliable AI
  • 31% uncomfortable with autonomous agents executing critical business processes without review
  • 49% of AI-generated outputs still require manual expert review before acting
  • 87% consider fixed, predictable pricing important for long-term AI success — a CFO/procurement signal that token-based API pricing creates budget unpredictability that slows enterprise commitments

The 80/49 gap is directionally consistent with BCG (5% substantial gains), McKinsey (6% high performers), Deloitte (34% transforming core processes), and Cloudera/HBR (7% data-ready). The convergence across independent surveys suggests a structural deployment bottleneck, not a survey artifact.

Note vendor context: Infor is an ERP vendor that benefits from evidence that scaling requires ERP-layer AI tooling. Apply LOW-MEDIUM credibility weight. Statistics should be cited alongside independent benchmarks.

Source: research/07-adoption-challenges/infor-enterprise-ai-adoption-impact-index-2026.md — TIER 1 (Apr 2026), LOW-MEDIUM credibility

Writer / Workplace Intelligence — The Adoption Illusion (April 2026)

Vendor-commissioned survey (n=2,400: 1,200 C-suite + 1,200 employees; Workplace Intelligence fieldwork, April 7, 2026) documenting the gap between AI investment and measurable ROI.

  • 75% of C-suite executives admit their company’s AI strategy is “more for show than actual guidance” — the highest self-reported strategy-theater rate in the 2026 corpus.
  • Only 29% see significant ROI from generative AI; only 23% from AI agents — despite 59% investing over $1M annually. Investment-to-return gap is widening.
  • 54% of C-suite say adopting AI is “tearing their company apart” — friction is organizational, not technical.
  • AI super-users save 9 hours/week, 3x more likely to receive promotions, 5x productivity multiplier. Everyone else is not realizing comparable gains.
  • Strategy theater pattern: organizations announce AI strategy for positioning, not execution — corroborated by McKinsey Nov 2025 (n=1,993) and EY Tech Pulse Mar 2026 (n=500).

Note vendor context: Writer is an enterprise AI platform vendor whose commercial interest aligns with findings highlighting informal adoption failure. Apply MEDIUM-HIGH credibility for directional findings; cite percentages with vendor-commission disclosure.

Source: research/07-adoption-challenges/writer-enterprise-ai-adoption-2026.md · May 2026 · MEDIUM-HIGH · TIER 1

Microsoft Work Trend Index 2026 — Organizational Factors Drive 67% of AI Impact Variance

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

  • 15-fold year-over-year growth in active Microsoft 365 AI agents (18-fold among large enterprises), based on platform telemetry March 2025–March 2026. Agentic AI has crossed from pilot to operational category at enterprise scale.
  • 67% of the variance in AI impact is explained by organizational factors — AI culture, manager behavior, and talent practices — vs. 32% from individual mindset. Scaling AI is an organizational design problem, not a tool problem. This finding is independently cross-validated by Deloitte (n=3,235), MIT CISR (n=132), and NBER (n=1,179).
  • Only 26% of AI users say their leadership is clearly and consistently aligned on AI strategy; only 13% are rewarded for reinventing work with AI. Scaling stalls when these systemic conditions are absent.
  • Frontier Professionals (16% of 20,000-person sample) are twice as likely to be rewarded for AI reinvention and work in environments where managers model AI use — defining what scaling readiness looks like organizationally.
  • Vendor caveat: Microsoft has direct commercial interest in AI adoption narratives as provider of Microsoft 365 Copilot; platform telemetry (behavioral, not self-reported) increases credibility for adoption-rate claims. MEDIUM-HIGH credibility · TIER 1 (May 2026).

Logicalis Global CIO Report 2026 — The Scaling Confidence Gap (n=1,000+, Mar 2026)

The CIO-side perspective on why 50%+ adoption doesn’t mean 50%+ organizational value capture.

  • 94% report growing appetite; 50%+ say adoption is moving too fast. Demand for AI is not the constraint — absorptive capacity is. Speed of deployment exceeds the organizational capacity to govern, integrate, and build skills around it.
  • Only 33% of CIOs believe they can scale AI beyond POC. Two-thirds of the people responsible for delivery doubt the organization can make the pilot-to-production transition.
  • 67% concerned about an “AI bubble.” This is not AI skepticism — it is concern about the mismatch between investment levels and demonstrated organizational returns, consistent with Foundry (33% cannot determine ROI) and Grant Thornton (piloting orgs 4x less likely to report revenue growth than fully-integrated ones).
  • Skills deficit (~89%) is the stated bottleneck — not model capability, not budget. Scaling AI requires a knowledge base most IT departments are still building.

Source: research/05-analyst-firms/logicalis-cio-report-2026.md · Mar 2026 · MEDIUM · TIER 1

Wharton / GBK Collective — Three-Year Adoption Tracking: From Exploration to Daily Work (n=801, October 2025)

The most sustained enterprise AI adoption series from an academic institution. Consistent sample criteria (1,000+ employees, >$50M revenue, U.S.), repeated cross-sectional design, three consecutive years. TIER 1 (October 2025). MEDIUM-HIGH credibility.

  • 82% use Gen AI weekly; 46% daily (+17pp year-over-year, +35pp vs. 2023). The transition from experimentation to infrastructure is complete at the senior decision-maker level. Daily use tripled in two years.
  • The industry spread is 22 percentage points wide: Tech/Telecom (94% weekly) through Retail (72%). Manufacturing (80%) and Retail lag, despite dense applicable use cases in both sectors. The gap is organizational and cultural, not technical.
  • 16% remain laggards (less than weekly use), concentrated in Retail (21%) and Manufacturing (23%). These organizations face compounding disadvantages: more restrictions, less internal expertise, slower workflow integration.
  • Large enterprises ($2B+) closed their gap, gaining +22pp on weekly usage — but smaller firms ($50M–$250M) still perceive themselves as more agile and “much quicker” at adoption (46% vs. 21% of Tier 1).
  • 58% report some AI agent deployment — but virtually all is human-supervised workflow automation (process automation, invoice matching, support ticket triage, IT remediation). This is not autonomous operation; it is the early-stage agentic transition consistent with IBM IBV 2026 (55% developing an agentic model) and Menlo Ventures 2025 (16% true agent deployment).
  • CAIO presence in 60% of enterprises (+from prior waves). Executive AI ownership has consolidated into the C-suite: 67% report active executive leadership in Gen AI strategy (+16pp year-over-year).
  • Top use cases are proven, not experimental: Data analysis (73%), document summarization (70%), document editing (68%), presentation creation (68%). These are repeatable, process-intensive tasks — exactly what independent productivity research identifies as highest-confidence AI applications.
  • Skill atrophy is an emerging concern: 43% agree Gen AI will lead to declines in employee skill proficiency (new question in 2025). 89% still say it enhances skills — the concern is not replacement but capability erosion for entry-level employees doing AI-assisted work without foundational skill development.

Cross-validation: Adoption rates corroborate Fed Reserve 2026 triangulation (41% individual worker GenAI usage at work, nationally representative), Census BTOS (18% firm-level, 32% employment-weighted), and McKinsey State of Organizations 2026 (88% deploying). The Wharton figure is higher because the sample skews large enterprise and senior decision-makers — the population that adopts fastest.

Datadog State of AI Engineering 2026 (thousands of orgs, telemetry, April 2026) — Production Adoption vs. Architectural Maturity

Production telemetry adds a critical dimension missing from survey-based adoption data: what organizations actually run in production, not what they report:

  • 50.6% of U.S. businesses now pay for AI (Ramp, April 2026) — but Datadog telemetry shows only ~18% of monitored organizations use agentic frameworks, and 59% of agents make a single service call. Adoption breadth (who pays) and depth (what architecturally runs) diverge significantly.
  • Model portfolio complexity is accelerating: 70%+ of organizations run 3+ AI models simultaneously; legacy models persist at 19–22% of production. Scaling governance must cover model diversity, not just new deployments.
  • Token consumption doubling per year at the median, quadrupling at the 90th percentile — the cost trajectory for scaled AI programs is compounding, not linear. Organizations that benchmarked AI infrastructure costs in 2024 are underbudgeting for 2026.
  • The operational failure mode at scale is infrastructure capacity, not model quality. Rate limit errors accounted for 60% of AI request failures in February 2026. Adoption scaling without capacity planning produces reliability degradation.

Source: research/01-ai-native-landscape/datadog-state-of-ai-engineering-2026.md — MEDIUM-HIGH / TIER 1 (production telemetry; thousands of orgs; April 2026)

Source: research/01-ai-native-landscape/wharton-gbk-enterprise-ai-year3-2025.md · Wharton/GBK Collective, n=801, October 2025 · MEDIUM-HIGH · TIER 1

Celonis 2026 Process Optimization Report — Scaling Requires Process Infrastructure First (n=1,649, June–July 2025)

The Celonis data adds a process-layer constraint to the adoption scaling picture: organizations cannot scale AI adoption without first scaling the process visibility and consistency that AI agents depend on.

  • 60% of businesses report they’re struggling to adapt their operations quickly enough to cope with continuous change — before any AI deployment. This is the baseline operational fragility that AI scaling compounds rather than relieves.
  • Process optimization is happening, but incrementally: 49% → 55% → 58% of organizations fully optimized at least one departmental process end-to-end in the prior 12 months (three-year trend). The acceleration is real but slow relative to the pace of AI deployment ambition.
  • When scaling works, the payoff is documented: 77% report productivity/efficiency gains; 52% report revenue increases; 50% report improved ability to adopt and scale new technologies. Process optimization and AI scaling are co-dependent — each unlocks the other.
  • The scaling failure mode is context starvation: AI agents need business context to scale beyond single-department pilots. 45% cite “getting AI to understand business context” as a top barrier. Without shared process models across departments, every new deployment requires a bespoke context-building exercise — the scaling bottleneck the data identifies.
  • Digital process twins are the emerging scaling infrastructure: 38% currently use a digital process twin; 50% plan to in the next 12 months. 93% of Process/Ops leaders say a business-wide digital twin would be a gamechanger. This is the process layer equivalent of a data platform — the shared infrastructure that makes incremental AI deployments cheaper and faster.

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

Dun & Bradstreet Global AI Momentum Survey — Scale Preconditions at 10,000-Org Resolution (n=10,000, Q1–Q2 2026)

The D&B survey is the largest enterprise AI dataset in the current corpus. Its adoption scaling findings confirm what smaller studies have found: initiative activity (97%) and production readiness (30% scaling, 26% multi-process) are separated by a wide middle where data infrastructure problems stall progress.

  • 97% active AI initiatives; 30% scaling into production; 26% operationalizing across multiple core processes. The pipeline narrows sharply from initiative to production — consistent with every pilot-to-production conversion study in the corpus.
  • The scaling bottleneck is not model capability. The top five obstacles are all data-layer: access, compliance, quality, integration, and skills. Organizations solving their data infrastructure problem are moving through the production bottleneck; those that aren’t are accumulating AI initiatives that don’t scale.
  • 56% plan to increase AI investment over the next 12 months — while only 5% have the data infrastructure to support it. Investment acceleration without infrastructure remediation produces more initiatives at the same conversion rate: more activity, proportionally similar production deployments.
  • Only 10% report strong ROI despite 60% reporting some measurable return. The scaling gap is not binary (works/doesn’t work) — it is a production readiness threshold below which ROI stays partial and above which it compounds.

Source: research/07-adoption-challenges/dnb-ai-momentum-survey-2026.md · D&B, n=10,000, 32 countries, Q1–Q2 2026 · HIGH / TIER 1

Foxit / Sapio Research — Verification Burden as the Scaling Brake (n=1,400, March 2026)

The Foxit/Sapio data identifies the specific adoption scaling failure mode that erases productivity gains at scale: when organizations add AI without redesigning the verification step, the time saved by AI is consumed by checking AI outputs.

  • 16 minutes net gain for executives after 3.6 hrs/week of perceived savings: validation overhead consumes all but 16 minutes. For end users, the net outcome is −14 minutes.
  • The verification burden is the mechanism behind every failed scaling attempt that looks like “AI isn’t working.” It is not an AI problem — it is a workflow architecture problem. Organizations that redesign the checking step recover their time savings; those that don’t stay negative.
  • At scale, this compounds: organizations with hundreds of AI-assisted workflows have hundreds of undesigned verification loops consuming staff time. The aggregate effect is the “productivity tax” documented across Workday (40% rework consumption), ActivTrak (all work categories increased post-AI), and METR (−19% on experienced developers).
  • The scaling unlock is explicit check-intensity policy by task risk: routine data extraction can use spot-check verification; consequential external communications require line-by-line review. Without this taxonomy, every AI output gets the same verification overhead regardless of stakes.

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

Epoch AI / Ipsos — Behavioral Usage Baseline for Scaling Projections (n=2,021, April 2026)

The Epoch AI / Ipsos probability-based national survey provides the most methodologically rigorous behavioral baseline for measuring AI adoption scaling in the U.S. workforce. Unlike self-selected enterprise panels, this is a Census-weighted probability sample.

  • 50% of U.S. adults used AI in the past week (April 2026, up from lower prior baselines). 51% of employed AI users apply it at least as much to work as personal use — the majority of scaling is happening within the workforce, not at the consumer edge.
  • The usage-to-value gap at scale: widespread use does not equal widespread value capture. The Epoch behavioral data documents the usage ceiling; the Foxit verification burden data explains why organizations with high adoption still report negligible net productivity.
  • Scaling from usage to value requires behavior change, not more access: at 50% adoption, the constraint on scaling is no longer getting people to try AI — it is redesigning work so that AI-generated outputs can be used without a parallel validation layer.

Source: research/07-adoption-challenges/epoch-ai-ipsos-ai-workplace-usage-2026.md · Epoch AI / Ipsos KnowledgePanel, n=2,021, April 2026 · HIGH / TIER 1

IMF Note 2026/002 — The Macro Constraint on Adoption Scaling (April 2026)

The IMF’s scenario-planning note provides the highest-level validation that organizational readiness — not cost or capability — is the binding constraint on AI adoption scaling globally.

  • Access cost is no longer the barrier: inference prices for frontier models have fallen over 99% in two years. The cost-as-barrier argument is closed. The scaling constraint has migrated entirely to organizational, regulatory, and institutional readiness.
  • As AI moves toward multi-agent systems, institutional norms and trust frameworks matter more than marginal model capability improvements. Scaling agentic deployments requires governance infrastructure that most enterprises don’t have.
  • The IMF explicitly classifies AI as a “macro-critical transition” — not a technology trend. This is the framing that moves AI from IT budget to board agenda. Organizations that scale successfully will be those that treat readiness as a structural competency, not a project.
  • Both the baseline (gradual) and runaway (rapid) diffusion scenarios require the same organizational preparation. Waiting for the path to become clear is itself a losing strategy — the preparation actions are the same under both futures.
  • Capital flows will follow AI readiness: advanced economies with superior infrastructure, regulatory clarity, and AI-skilled labor will disproportionately capture AI gains. Mid-market organizations in weaker-readiness positions face the same divergence dynamic at the firm level.

Source: research/07-adoption-challenges/imf-global-economic-financial-implications-ai-2026.md · IMF Note 2026/002, ~50 expert participants, December 2025 workshop / April 2026 publication · HIGH / TIER 1

Gartner CIO Agenda 2026 — Deployment Execution Gap (n=2,501)

The largest annual CIO survey quantifies the structural gap between investment intent and operational AI deployment:

  • 91% of CIOs increase GenAI budgets (mean +38%) — the highest growth rate of any technology category — yet only 17% have deployed AI agents in any operational capacity
  • 64% plan agentic deployment within 24 months: the largest stated planning intent in the survey’s history, which will test whether governance and organizational readiness can keep pace with ambition
  • Only 48% of digital initiatives currently meet targets: the baseline success rate before AI complexity is added. Organizations adding agentic AI to a baseline where half of digital projects fail are not starting from a position of strong execution capability
  • Off-cycle reprioritization is the differentiator: only 18% of CIOs embrace dynamic mid-cycle replanning, yet they are 24% more likely to be top performers. In fast-moving AI deployments, the ability to adjust course is more valuable than the quality of the initial plan
  • 94% anticipate significant disruption to plans within 24 months — nearly universal recognition that scaling AI will not go according to plan; the organizations prepared for that disruption are the 18% who’ve built replanning muscle

Source: research/05-analyst-firms/gartner-cio-agenda-2026.md · Gartner CIO and Technology Executive Survey, n=2,501, Jan 2026 · MEDIUM-HIGH / TIER 1

Cisco CEO AI Readiness 2026 — The 13% Threshold (n=2,511)

The largest CEO-specific AI readiness survey quantifies the structural gap between leading and lagging organizations:

  • Only 13% of organizations qualify as Pacesetters in Cisco’s AI Readiness Index — the top tier across strategy, infrastructure, data, talent, governance, and culture. The 87% gap has not narrowed across index editions, suggesting stratification rather than catch-up.
  • The largest differentials are organizational, not technical: 99% vs. 58% on strategy clarity; 75% vs. 16% on staff AI proficiency; 65% vs. ~33% on measuring training program impact — all human/organizational gaps that precede any technology investment.
  • 97% of Pacesetters deployed AI at scale/speed needed for ROI vs. 41% overall — a 56-point deployment efficacy gap that directly tracks organizational readiness, not technology availability.
  • Agent deployment is a top-3 CEO priority but only 32% of companies have mapped what agents will do — the classic pattern of technology acquisition without workflow redesign that predicts underperformance.
  • Sequencing is the differentiator: Pacesetters built strategy before infrastructure, workforce capability before agent deployment, and measurement before scale. The majority did the reverse.

Source: research/07-adoption-challenges/cisco-ceo-ai-readiness-2026.md · Cisco/Opinion Matters, n=2,511 CEOs, 23 countries, Jan 2026 · MEDIUM / TIER 1

Everyday AI — Scaling Signals from Ep 770 (May 5, 2026)

Two macro-level indicators from Everyday AI Ep 770 that contextualize enterprise scaling decisions in 2026:

  • $2.5 trillion in global AI spending in 2026 (Gartner forecast, HIGH credibility), a 44% annual increase — the largest single-year investment surge in enterprise technology history. Every CIO’s scaling roadmap competes for vendor capacity, talent, and implementation resources against this backdrop.
  • 15x year-over-year growth in active Microsoft 365 AI agents (Microsoft Work Trend Index telemetry, May 2026, HIGH credibility) — corroborating the platform-telemetry data from the Microsoft WTI file above. Agentic AI is scaling on Microsoft’s platform whether or not organizations have a governance framework in place.
  • Capability growth (~100% quarterly) is outpacing organizational proficiency growth (5–10% quarterly) — the widening gap between what AI can do and what organizations can responsibly deploy is the primary scaling risk in 2026.

Source: research/13-multimodal-sources/everyday-ai/2026-05-22-ep755-780-enterprise-ai-mining.md — Everyday AI Ep 770, Jordan Wilson, May 5, 2026. Underlying stats: Gartner (HIGH) and Microsoft WTI (HIGH). Episode source is MEDIUM. TIER 1.

Microsoft Global AI Diffusion Q1 2026 — Where the World Actually Is (Telemetry, 150+ Economies)

The Microsoft AI Economy Institute’s Q1 2026 diffusion report provides the most geographically comprehensive tracking data available — actual product usage across 150+ economies, not self-reported survey intent.

  • 17.8% of the world’s working-age population used a generative AI product in Q1 2026, up from 16.3% in H2 2025 (+1.5pp in a single quarter). The adoption curve is still accelerating.
  • U.S. adoption: 31.3% (rank: 21st globally) — in the bottom half of high-income economies. France (47.8%), UK (42.2%), and Australia (39.5%) are all materially ahead.
  • 26 economies now exceed 30% adoption among their working-age populations.
  • Asia is the fastest-growing region. South Korea grew 43% since June 2025; Japan 34%; Thailand 36%. The driver is improved local-language model performance, not infrastructure investment — a diffusion acceleration mechanism that will repeat as model quality improves in additional languages.
  • The North-South gap is widening: Global North 27.5% vs. Global South 15.4% — the gap grew from 9.8pp (H1 2025) to 12.1pp (Q1 2026). Organizations with significant Global South operations cannot assume AI-enabled workflows will transfer.
  • Developer employment: U.S. software developer employment reached a record 2.2 million in 2025 (+8.5% YoY). AI coding tools are correlating with more developer hiring, not less — consistent with the supply-side elasticity mechanism (cheaper software production → more software built).

For CIOs benchmarking competitive exposure: the relevant question is not whether the U.S. is adopting AI but at what rate relative to the economies they compete in and source talent from.

Source: research/01-ai-native-landscape/microsoft-global-ai-diffusion-q1-2026.md · Microsoft AI Economy Institute, aggregated/anonymized telemetry, May 2026 · MEDIUM / TIER 1

Developer Adoption Landscape — The Paradox of Near-Universal Access and Growing Distrust

The adoption-landscape synthesis documents the structural challenge facing any AI scaling program in knowledge work and engineering as of mid-2026:

  • 84-85% of developers use AI tools, but favorable sentiment dropped from 70%+ (2023-2024) to 60% (2025); trust in AI output accuracy fell to 33% (Stack Overflow, n=65,000+, 2025).
  • The primary scaling constraint is organizational readiness — the capacity to absorb AI capability into workflows — not model performance or tooling.
  • 42% of companies abandoned the majority of AI initiatives in 2025 (Pertama Partners), confirming that scale without workflow redesign does not sustain itself.
  • Shadow AI is the leading indicator of a scaling program that outran its change management. 80%+ of workers use unapproved tools; providing quality sanctioned alternatives reduces this by 89%, meaning sanctioned-first is both a security and adoption-quality control.
  • Only 6% of firms capture >5% EBIT impact (McKinsey, n=1,933, November 2025) — adoption is necessary but not sufficient; scaling requires the deliberate ROI measurement discipline this wiki article describes.

Source: research/07-adoption-challenges/adoption-landscape.md — MEDIUM / TIER 1–2

KPMG Q1 2026 — The Scaling Barrier Has Doubled in 12 Months

KPMG’s quarterly pulse (n=130 U.S. C-suite, $1B+ revenue, March 2026) provides the clearest evidence that scaling difficulty is worsening, not easing, despite investment growth:

  • 65% cite difficulty scaling use cases as top ROI barrier — up from 33% in Q1 2025. The proportion unable to scale effectively doubled in one year, while agent deployment grew from 11% to 54% over the same period.
  • 62% cite skills gaps as a scaling barrier — up from 25% in Q1 2025. The gap between what is technically possible and what the workforce can absorb has grown faster than deployment itself.
  • 54% of organizations actively deploying AI agents, but only 19% are actively scaling across multiple functions (global KPMG data, n=2,000+). The remaining 35% are piloting without organizational infrastructure for scaling.
  • KPMG’s synthesis: “Execution — not capital or technology — now determines outcomes.” The constraint is workforce readiness and operating model redesign, not budget or model capability.
  • 64% have changed their entry-level hiring approach due to AI agents — the labor-force adaptation is underway, but current employees need reskilling concurrently. 87% are upskilling/reskilling; only 55% of employees show adoption or integration.

The doubling of scaling difficulty in a year where investment nearly doubled (avg. $88M Q1 2025 → $207M Q1 2026) is the defining data point: money is not the constraint.

Source: research/07-adoption-challenges/kpmg-ai-quarterly-pulse-roi-disconnect-2026.md

Section AI Proficiency Report (Jan 2026, n=5,000) — The Proficiency Distribution Gap

Section’s survey of 5,000 knowledge workers at 1,000±employee companies (US/UK/Canada) provides the most granular proficiency-tier breakdown in the corpus, and it reframes the adoption problem:

  • 97% of enterprise knowledge workers are using AI poorly or not at all. Only 3% qualify as practitioners or experts generating measurable employer returns.
  • The adoption distribution is: 0.08% experts, ~2.7% practitioners, ~69% experimenters (basic tasks only), ~28% novices.
  • The experimenter layer — 69% — represents the core scaling problem. These workers have adopted AI in the sense that they open a tool; they have not adopted AI in the sense that they restructure work around it. Summarizing, rewriting, and search replacement do not compound.
  • 15% of submitted use cases likely generate employer-level ROI. 59% are basic task assistance. Only 2% are advanced applications (automation, process redesign) — the category with highest ROI potential.
  • Training is occurring (44% of workers receive it) but is not producing capability: trained employees average 40/100 on hands-on proficiency tests.
  • Technology sector scores 42/100 — the leading industry. This is not a satisfactory benchmark.
  • 40% of knowledge workers would accept never using AI at work again — not rejection of technology, but indifference to AI in the specific form they have encountered it at work.

Scaling implication: organizations measuring AI adoption by tool access rates or weekly active users are measuring activity, not capability. The scaling constraint identified by KPMG (skills gaps doubled in one year) and Gartner (training and guidance determine outcomes) is confirmed here at the use-case level: the workforce is active but not proficient.

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

Zapier / Centiment — Agentic AI Adoption Among Senior Executives (n=525, Oct 2025) — The Governance Ceiling

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)

  • 72% of enterprises at the C-suite level are using or testing AI agents — majority threshold crossed. 40% have multiple agents in production.
  • 84% plan to increase AI agent investment over the next 12 months — scaling investment is already underway regardless of governance readiness.
  • Only 38% use human-in-the-loop oversight. 20% operate with minimal oversight. The governance gap at the organizational level is already visible before agents have been deployed at full scale.
  • Infrastructure is fragmenting: 53% use major cloud providers, 48% use open-source or enterprise AI platforms, 46% use orchestration frameworks. Multi-vendor sprawl is the default posture — not a planned architecture.
  • The scaling constraint named by KPMG (skills) and Gartner (guidance) shows up here as a governance gap: 72% deployed but only 38% with meaningful oversight means the majority of enterprises are scaling into a governance deficit.

Source: research/04-consulting-firms/pwc-digital-trends-operations-2026.md · PwC, n=767 U.S. operations and supply chain leaders, $100M+ revenue, 8 industries · MEDIUM-HIGH / TIER 1

  • Only 4% of respondents meet all four success criteria simultaneously: AI embedded enterprise-wide, no significant barriers to scaling agents, horizontal operating structure, and technology investments fully delivering expected results. For every 25 companies that started this journey, one is actually winning.
  • 89% say technology investments have not fully delivered expected results — yet 85% simultaneously claim they are ahead of competitors. The gap between self-assessment and outcomes is the defining feature of enterprise AI at scale.
  • 27% have fully embedded an AI strategy across business units. Only 37% are comfortable assigning AI agents to end-to-end operational processes. Ambition exceeds structural readiness by more than 2×.
  • 87% cite poor data quality as the primary reason digital initiatives fail to deliver value. Only 30% report significant improvement in data quality and reliability — confirming that data readiness remains the unresolved prerequisite to scaling.

IBM IBV CEO Study (May 2026, n=2,000 CEOs)

Source: research/07-adoption-challenges/ibm-ibv-ceo-csuite-ai-era-2026.md

  • 4x more likely to deliver on AI business objectives for organizations that redesigned five core areas (technology, finance, HR, operations, cross-functional collaboration) vs. those that did not — the most direct quantification of the “structural change first” argument in the corpus.
  • AI-first C-suite design (CAIO at peer level with CFO/CHRO/COO) correlates with +10% more AI initiatives successfully scaled enterprise-wide.
  • CAIO surge: 76% of organizations have a Chief AI Officer in 2026, up from 26% in 2025 — the structural response to repeated pilot-to-production failure when AI governance sits below C-suite level.
  • 79% are decentralizing decision-making — consistent with the Rewired argument that scaling requires moving ownership from central COE to business-unit operators.

Capgemini AI Perspectives 2026 (n=1,505, January 2026)

Source: research/04-consulting-firms/capgemini-ai-perspectives-2026.md · Capgemini Research Institute, 15 countries, director-level and above, $1B+ revenue · MEDIUM-HIGH / TIER 1

  • Only 38% of organizations have operationalized generative AI use cases beyond pilots — consistent with BCG’s 5% “substantial gains” (different threshold: BCG measures financial performance, Capgemini measures production deployment of at least one use case).
  • 63% are actively pruning low-value AI projects — the pilot proliferation era is ending; executive attention is shifting from breadth to depth.
  • AI budgets projected to reach 5% of annual business spend in 2026 (up from 3% in 2025); organizations moving from 12-month ROI expectations to 5-year investment horizons.
  • 60% of organizations globally are exploring agentic AI; nearly 50% of Chinese organizations are already piloting or deploying it — a 12–18 month geographic adoption gap that reflects Western organizations’ higher verification overhead requirements, not capital or technology differences.
  • Top scaling enablers: executive sponsorship (67%), workforce upskilling (60%), governance frameworks (53%) — not model selection or compute access.