← AI Adoption Cycle 🕐 14 min read
AI Adoption Cycle

30-Minute AI Workflow Readiness Assessment

Brandon Sneider · April 2026

Pick one specific workflow. Score each question: **Yes = 2 points / Partial = 1 point / No = 0 points.**

Executive Summary

  • Organizational AI maturity and workflow AI-readiness are independent variables. A company can be in McKinsey Stage 1 overall and still have one workflow that is AI-ready today. Conversely, a company can have an enterprise AI strategy in place and identify zero workflows that are actually ready to deploy.
  • This 20-question assessment evaluates a specific workflow — not the company — in five dimensions: Data Foundation, Decision Architecture, Human Oversight Design, Adoption Readiness, and Scale Architecture. Each dimension draws on a distinct body of evidence from the corpus.
  • Score 29–40: the workflow is ready for AI deployment. Score 16–28: ready with gaps that must be remediated before production approval. Score 0–15: not ready — proceed to remediation roadmap before any investment.
  • Rewired’s “digital and AI backbone” framing (Lamarre et al., 2024, Ch. 25 p.391) anchors the data foundation section: “without a governed, reusable data architecture, AI investments produce recurring bespoke-integration costs and bounded outcomes.” This assessment operationalizes that principle at the workflow level.

How to Run This Assessment

Pick one specific workflow. Score each question: Yes = 2 points / Partial = 1 point / No = 0 points.

Complete the assessment in a 30-minute session with the workflow owner, the relevant IT/data contact, and — if it exists — the AI or transformation lead. Do not complete it with only one role present; the gaps surface in the disagreements between roles.

Maximum score: 40 points.

Section 1 — Data Foundation (8 points possible)

Evidence base: wiki/data-readiness.md; research/07-adoption-challenges/ai-data-reset-decision-framework.md; Lamarre et al., 2024, Ch. 25 p.391 (“No data architecture, no AI advantage”)

Q1. Is the input data for this workflow accessible from a single system of record without manual export or cleaning?

  • Yes (2): All input data is in one system, accessible via API or direct query, with a documented schema.
  • Partial (1): Data is in 2 systems; minor cleaning required; some inconsistency in field formats.
  • No (0): Data requires manual aggregation from 3+ systems, significant cleaning, or undocumented formats.

Corpus anchor: Only 7% of enterprises say their data is completely ready for AI (Cloudera/HBR, n=230, March 2026). This question identifies whether this workflow is in that 7%.

Q2. Has input data for this workflow been audited for quality in the last 12 months?

  • Yes (2): A formal data quality assessment was completed; findings were documented and actioned.
  • Partial (1): An informal review was conducted; no formal audit; some quality issues are known but not quantified.
  • No (0): No data quality audit has been conducted; quality issues may exist but are not tracked.

Corpus anchor: Organizations that conduct formal data readiness assessments before AI project approval achieve 47% success rate vs. 14% without (Pertama Partners, 2025–2026).

Q3. Does this workflow require crossing 2 or fewer data domains (distinct business entities with separate source systems and ownership)?

  • Yes (2): The workflow operates within a single data domain (one source system, one ownership function).
  • Partial (1): The workflow requires joining 2 domains; basic entity resolution is needed but manageable.
  • No (0): The workflow crosses 3+ data domains — patient/procedure/payer, or order/inventory/supplier/finance — with no existing semantic integration layer.

Corpus anchor: Domain count is the primary decision variable for data reset necessity. 3+ domains without a semantic layer produces the Data Mirage failure pattern in 60–70% of cases (Pertama Partners, n=2,400+, 2025–2026; research/07-adoption-challenges/ai-data-reset-decision-framework.md).

Q4. Has the tech debt embedded in this workflow’s data systems been inventoried and priced?

  • Yes (2): Tech debt affecting this workflow is documented, estimated (time and cost), and included in the deployment business case.
  • Partial (1): Some tech debt is known informally; not formally priced or included in the business case.
  • No (0): Tech debt has not been assessed; the business case does not include a remediation line item.

Corpus anchor: IBM IBV (n=1,300, Nov 2025) finds 18–29% of total AI implementation cost through 2027 will be absorbed by tech-debt remediation. Business cases that omit this project +39% ROI; post-mortem reality is −14%.

Section 2 — Decision Architecture (8 points possible)

Evidence base: wiki/workflow-redesign.md; research/07-adoption-challenges/workflow-level-ai-readiness-checklist.md; Lamarre et al., 2024, Ch. 5 p.82–94

Q5. Can the correct output for a given input be defined in writing and verified by a reviewer in under 5 minutes?

  • Yes (2): Decision criteria are explicit, rule-based, and documented. A reviewer can assess correctness independently.
  • Partial (1): Criteria are mostly explicit but require some contextual judgment not captured in writing.
  • No (0): Correct output depends on tacit knowledge, relationship context, or judgment not present in the input data.

Corpus anchor: “Decision Clarity” is Criterion 2 in the six-criterion workflow AI-readiness scorecard (research/07-adoption-challenges/workflow-level-ai-readiness-checklist.md). Workflows that fail this criterion require a different deployment strategy — AI as a draft, not AI as a decision.

Q6. Does this workflow process enough volume to make AI acceleration compound materially?

  • Yes (2): 5,000+ transactions/month, stable volume, 12+ months of history.
  • Partial (1): 500–5,000 transactions/month, reasonably stable.
  • No (0): Under 500 transactions/month, or highly variable volume that spikes unpredictably.

Corpus anchor: Volume threshold determines whether cost-per-unit AI deployment justifies investment. BCG’s AI cost-advantage research (n=1,250, Sep 2025) identifies procurement (5–25% savings in 3–6 months) and inventory (5–15%) as high-volume workflows where AI investment compounds fastest.

Q7. Is this workflow currently documented in writing, at the task level — not just described in broad terms?

  • Yes (2): Step-by-step process documentation exists, is current (updated within 12 months), and covers exception paths.
  • Partial (1): High-level documentation exists; task-level steps are not fully written; knowledge lives primarily in experienced staff.
  • No (0): Process is undocumented; knowledge is held by individuals who are not required to document their work.

Corpus anchor: Stanford Enterprise AI Playbook 2026 (n=51 deployments) finds 27% of prior failures were attributed to “critical knowledge never captured.” AI cannot reliably execute an undocumented process — it requires constant human guidance that collapses the productivity case.

Q8. Has the workflow been redesigned (not just studied) to accommodate AI execution before this assessment?

  • Yes (2): The workflow was explicitly redesigned to define what AI handles, what humans handle for exceptions, and what activities are eliminated.
  • Partial (1): Some steps were discussed for modification; no formal redesign was completed; the workflow is still the pre-AI version with AI layered on top.
  • No (0): The workflow is unchanged from before AI was considered. This assessment is the first structured review.

Corpus anchor: McKinsey (n=1,993, Nov 2025): 55% of high performers fundamentally redesigned workflows vs. 18% of others — the single most predictive behavioral variable across 25 attributes. Lamarre et al., 2024, Ch. 5 (reimagining workflows with agentic AI) and Ch. 30 p.457 (make adoption stick) both treat redesign as a prerequisite, not a Phase 3 activity.

Section 3 — Human Oversight Design (8 points possible)

Evidence base: wiki/hitl-deployment-pattern.md; research/07-adoption-challenges/hitl-as-adoption-architecture.md; Lamarre et al., 2024, Ch. 34 p.511

Q9. Is the consequence of an incorrect AI output bounded and reversible within the current oversight design?

  • Yes (2): Errors are minimal in consequence, fully reversible, and error rate is tracked with clear thresholds.
  • Partial (1): Errors are manageable; some are reversible; a monitoring mechanism exists but is not automated.
  • No (0): Errors could produce irreversible regulatory, legal, safety, or significant financial consequences without a reliable detection mechanism.

Corpus anchor: Error Cost is Criterion 4 in the six-criterion workflow scorecard (research/07-adoption-challenges/workflow-level-ai-readiness-checklist.md). A Score 1 on this criterion is the only workflow characteristic that warrants stopping regardless of all other scores.

Q10. Does the review step require genuine judgment rather than a click-to-approve?

  • Yes (2): Human review involves actual evaluation of the AI output against defined criteria; average review time is measured and above a minimum threshold; rubber-stamp reviews are flagged.
  • Partial (1): Review is expected to involve judgment, but no mechanism monitors whether it actually does; review time is not tracked.
  • No (0): Review is a click-to-approve step with no quality monitoring; no evidence that reviewers engage with the output content.

Corpus anchor: Thomson Reuters monitors validation gates for rubber-stamping and flags reviews completed in under two seconds. Dietvorst et al. (Management Science, 2016): the adoption benefit of human modification authority only activates when the review carries real time and real authority.

Q11. Are escalation criteria defined in writing, specifying exactly when the workflow should route to a human rather than proceeding autonomously?

  • Yes (2): Escalation criteria are written, system-enforced, and automatically trigger routing. SLA violations generate alerts.
  • Partial (1): Escalation criteria exist informally; routing depends on reviewer discretion rather than system enforcement.
  • No (0): No written escalation criteria; the human handles whatever the AI cannot. “Use your judgment” is the only rule.

Corpus anchor: Bain’s three-phase agentic deployment model (Foundation → Orchestration → Scale) requires escalation architecture in the Foundation phase, before multi-agent orchestration begins. Deploying without this produces the governance failure documented in the FinCo case (research/04-consulting-firms/mckinsey-methodology-critique.md).

Q12. Is AI output traceable after the fact — can a non-participant identify what the AI decided, when, with what input, and which model version?

  • Yes (2): Complete audit trail exists: input, model version, output, reviewer identity, timestamp. Quality is monitored automatically.
  • Partial (1): Partial logging exists; some components are traceable; full reconstruction of a decision is possible but requires manual effort.
  • No (0): No audit trail; AI outputs are not logged with input context; post-hoc review of AI decisions is not possible.

Corpus anchor: Lamarre et al., 2024, Ch. 34 p.511 — the “right to deploy AI” gating framework requires auditability as a precondition for legal and compliance sign-off. Auditability is also Criterion 5 in the workflow AI-readiness scorecard.

Section 4 — Adoption Readiness (8 points possible)

Evidence base: research/07-adoption-challenges/ai-change-management-best-practices.md; research/04-consulting-firms/mckinsey-rewired-2nd-edition-synthesis.md (Capability 5)

Q13. Has a named business leader (not IT) been designated as the owner of this workflow transformation?

  • Yes (2): A business unit leader owns the workflow redesign outcome and has authority to change the workflow, allocate time, and evaluate results.
  • Partial (1): A business leader is “involved” but IT leads the project; business-side authority is advisory.
  • No (0): IT owns the project. Business unit access is required for testing but business leaders are not accountable for outcomes.

Corpus anchor: Lamarre et al., 2024, Ch. 4 p.63 — “have business leaders lead the reimagination of their domain.” The Air Canada case: the domain owner is a business leader, not the CIO. McKinsey Manifesto: concentrating on 1–3 domains with business-leader ownership delivered 20% EBITDA uplift.

Q14. Do employees who will use the AI tool know why this workflow is being changed and what their role will be after deployment?

  • Yes (2): Employees were involved in workflow redesign; they understand what the AI handles and what they are responsible for; no ambiguity about role boundaries.
  • Partial (1): Communication has occurred but is top-down; employees were not involved in design; questions about role impact remain unanswered.
  • No (0): Employees have not been informed; the change is being managed as an IT deployment, not a workflow change.

Corpus anchor: Dr. Sam Zolfagharian (AI For the C-Suite, Apr 2026): “You’re bringing these AI tools to replace me. So I’m not going to do anything.” Top-down deployment without role clarity produces the sabotage pattern that Arya Bolurfrushan documents (research/13-multimodal-sources/ai-for-the-c-suite/2026-04-14-ayra-bolurfrushan-most-companies-are-thinking-about-ai-compl.md).

Q15. Is there a structured training program covering both how to use the tool and when not to use it?

  • Yes (2): A training program exists covering tool mechanics, workflow context, quality standards, escalation judgment, and known failure modes. Completion is tracked.
  • Partial (1): Training covers tool mechanics only; workflow integration, quality standards, and failure modes are not covered; completion is not tracked.
  • No (0): No training program; deployment approach is “here’s the tool, figure it out.”

Corpus anchor: BCG AI at Work 2025 (n=10,600): employees with 5+ hours of AI training become regular users at 79% vs. 67% with less. Workday/Hanover Research (n=3,200, Jan 2026): 40% of AI time savings consumed by rework among under-trained users.

Q16. Is there a defined measurement cadence — at 90 days, 6 months, 12 months — with named financial metrics and named reviewers?

  • Yes (2): Measurement cadence is defined with specific metrics (unit cost, throughput, error rate, P&L line items), a named reviewer for each gate, and a decision rule for what happens if targets are not met.
  • Partial (1): Some metrics are defined; cadence is informal; financial impact measurement is “when we get around to it.”
  • No (0): No measurement cadence; success will be assessed qualitatively; no pre-defined criteria for stopping or scaling.

Corpus anchor: Rewired Ch. 32 p.483 (track what matters — KPI discipline tied to P&L line items, not activity counts). Pertama Partners: 54% success rate with pre-defined financial metrics vs. 12% without.

Section 5 — Scale Architecture (8 points possible)

Evidence base: research/04-consulting-firms/mckinsey-rewired-2nd-edition-synthesis.md (Capability 4 and 5); wiki/ai-maturity-models.md

Q17. Is the deployment designed so that infrastructure built for this workflow can be reused for the next 2–3 AI use cases in this domain?

  • Yes (2): Deployment uses a platform or data product approach; reusable components are identified; the next use case would start from a higher base.
  • Partial (1): Some reuse is possible but was not the explicit design goal; the next deployment would require partial rebuild.
  • No (0): Point-solution deployment; infrastructure built specifically for this workflow; no reuse path planned.

Corpus anchor: Rewired Ch. 31 p.469 — “the best use case is the reuse case.” Palantir’s 139% net dollar retention (Q4 2025 audited): after the Ontology foundation, each new use case costs a fraction of the first — the economic mechanism of compounding returns.

Q18. Is there a plan for what happens when this workflow’s AI performance degrades — including monitoring, model update governance, and rollback?

  • Yes (2): Monitoring, model update governance, and rollback procedures are documented and owned. Degradation thresholds trigger specific review steps.
  • Partial (1): Monitoring exists but degradation response is ad hoc; no formal model update governance; rollback is technically possible but not planned.
  • No (0): No performance monitoring; no degradation response plan; no rollback capability.

Corpus anchor: Rewired Ch. 33 p.501 (plan for midstream adjustments). Bain agentic AI governance framework: observability infrastructure must be in place before orchestration begins. This is the production-continuity equivalent of the pre-deployment governance requirement.

Q19. Can this deployment operate without requiring the original implementation team to be on call for the first 90 days?

  • Yes (2): Runbook documentation covers all standard operating scenarios; handoff to operations team is planned; on-call is bounded to escalation cases.
  • Partial (1): Key knowledge is in the implementation team’s heads; runbook is partial; operations handoff is planned but documentation is incomplete.
  • No (0): The implementation team is the operational team; no handoff plan; day-to-day operations depend on technical knowledge not yet transferred.

Corpus anchor: MIT CISR “4S” framework (Mar 2026) — the Stage 2→3 transition requires Synchronization (redesigned roles) and Stewardship (governance embedded in operations), both of which require an operations handoff, not continued implementation-team dependency.

Q20. Is there a clear escalation path from this workflow’s team to a named executive who can approve scope changes, additional budget, or deployment pause without a multi-month approval cycle?

  • Yes (2): Escalation path is named, tested, and fast: a specific executive is responsible, can be reached within 24 hours, and has authority to act.
  • Partial (1): An escalation path exists in theory; reaching the decision-maker requires 3+ layers of sign-off; response time measured in weeks.
  • No (0): No named executive with deployment authority; scope changes require a new business case cycle.

Corpus anchor: Stanford Enterprise AI Playbook 2026 (n=51 deployments): Executive Sponsorship cited as primary accelerator in 43% of deployments. The absence of this path is the structural reason pilots take years at one company and weeks at another — not technology complexity.

Score Interpretation

Total Score Interpretation Recommended Action
0–15 Not Ready Stop. Identify the lowest-scoring section as the primary remediation target. Do not approve investment until that section reaches Partial or above on all questions.
16–28 Ready with Gaps Proceed to pilot with remediation milestones. Sections scoring below 4/8 are production blockers if not resolved before the production approval gate (typically 90-day pilot exit).
29–40 Ready Proceed to pilot. Flag any individual No-score questions as monitoring priorities. Run the assessment again at the 90-day production approval gate to confirm no scores have degraded.

Section Score Diagnostic

When total score falls in 16–28, the section breakdown identifies where to invest:

Section Score 0–3 Score 4–6 Score 7–8
1 — Data Foundation Full data reset required before any pilot Targeted data remediation (4–12 weeks) Data ready
2 — Decision Architecture Workflow must be redesigned before pilot Document exceptions and redesign specific steps Decision architecture ready
3 — Human Oversight Governance design required Refine escalation and monitoring Oversight ready
4 — Adoption Readiness Change management program required Targeted training and communication Adoption ready
5 — Scale Architecture Platform and reuse design required Document runbook and reuse plan Scale ready

Running This Assessment with Leadership

The assessment is designed as a leadership tool, not a technical checklist. The most valuable sessions are the ones where a business leader and an IT leader disagree on the score for the same question. Those disagreements surface the actual organizational gaps: IT believes the data is clean (Q1); the workflow owner knows it requires manual export every Monday morning. Surfacing that disagreement in a 30-minute session is worth more than any individual score.

If the discussion about this assessment in your organization produces defensiveness or score inflation, name the pattern directly: the corpus finds that organizations that honestly identify pre-deployment gaps and remediate them achieve 3–4x higher production success rates than those that proceed through checklist gaps.

The goal is not a high score. The goal is an accurate score before investment is committed.

If this assessment raises questions specific to your organization’s workflow inventory or deployment sequencing, I’d welcome the conversation — brandon@brandonsneider.com.

Brandon Sneider | brandon@brandonsneider.com April 2026