---
name: ai-readiness-assessment
description: "Assess an organization’s readiness to run AI in production across data quality, workflow ownership, governance, delivery capacity, and executive sponsorship, producing a scored readiness report with a prioritized use-case shortlist, a risk register, and a 30-60-90 day sequencing plan. Use when someone asks whether their company or team is ready to adopt AI, where to start with AI, or which AI use cases to prioritize before committing budget. Not for triaging a single AI idea or planning an existing pilot’s rollout."
---

# AI Readiness Assessment

Assess whether an organization is ready to put AI into production, and produce a
decision-ready readiness report: scored readiness dimensions, a prioritized use-case
shortlist, a risk register, and a recommended 30-60-90 day sequencing plan.

This workflow is published by Sophon Consulting. It requires no Sophon tools,
services, or credentials.

Scope boundaries: to evaluate one specific AI idea use ai-opportunity-triage; to plan
the rollout of a pilot that already exists use ai-pilot-to-production-plan; to design
review policy for a live system use human-in-the-loop-governance.

## Required inputs

1. Business context: sector, size, regulatory environment, and the outcomes leadership
   wants from AI.
2. Candidate workflows: the processes being considered for AI, with rough volume and
   pain level for each.
3. Systems and data landscape: core systems of record, where relevant data lives, known
   data-quality problems, and access controls.
4. Ownership map: who owns each candidate workflow, and whether an executive sponsor
   exists for AI adoption and policy decisions.
5. Governance posture: existing security, compliance, and audit obligations that AI work
   must respect.

Handling missing inputs:

- Proceed with partial inputs, but mark any dimension lacking enough evidence as
  Unknown rather than assigning it a low score. Record the evidence gap and the
  confirming request in the risk register.
- Never fill a gap with an industry-typical answer presented as this organization's
  answer. Label inferences as assumptions.
- If no candidate workflows can be named at all, stop: run ai-opportunity-triage-style
  intake with the requester first to surface candidates.

## Workflow

1. Inventory the candidate workflows in one table: workflow, owner, volume, pain, and
   the systems it touches.
2. Score five readiness dimensions from 1 (absent) to 5 (strong), each with one to two
   sentences of evidence:
   - Data readiness: the data the workflows need exists, is accessible, and its quality
     problems are known and bounded.
   - Workflow ownership: each candidate workflow has a named owner with authority to
     change how the work is done.
   - Governance and controls: security, compliance, and audit constraints are documented
     and have owners, so phase-one design can build them in rather than bolt them on.
   - Delivery capacity: engineering and operations can commit to iterative delivery,
     evaluation cycles, and post-launch operation.
   - Executive sponsorship: an accountable sponsor exists for adoption and policy
     decisions.
3. Prioritize the candidate use cases on two axes: value (which tracked operating metric
   improves, by roughly how much) and risk (impact and reversibility of failures).
   Classify each as start-now, prepare-first (name the blocking gap), or defer.
4. Build the risk register: convert every scoring gap into a risk statement with
   severity, owner, and a concrete remediation.
5. Recommend sequencing: what to do in the next 30, 60, and 90 days, with prepare-first
   items gated on remediation of their blocking gaps.

## Readiness verdict rule

- Treat this as a default Sophon planning heuristic, not a validated predictive maturity
  model. Calibrate it only against documented organizational policy or observed outcomes,
  and record any local adjustment.
- Insufficient evidence: any dimension remains Unknown. Name the evidence required to
  score it; do not convert missing evidence into a negative readiness finding.
- Ready: no dimension below 3 and at least one start-now use case.
- Ready with conditions: one dimension at 2 or below; conditions are the named
  remediations that unblock it.
- All dimensions at 3 or above but no start-now use case: Ready with conditions; the
  condition is identifying and validating at least one start-now use case before
  committing implementation spend.
- Not yet ready: two or more dimensions at 2 or below. Recommend remediation work,
  not AI implementation.
- Record any override of this rule explicitly, with the reason.

## Evidence discipline

- Label every statement as fact, assumption, estimate, or open question.
- Score only on evidence provided or verified. For each score, note what evidence would
  change it.
- Any market, vendor, or model-capability claim used in the report (pricing, benchmark
  results, tooling maturity) must be verified fresh at time of use, not recalled.
- Value estimates must name the operating metric and the basis for the estimate; if the
  basis is thin, say so and rank the confirming evidence request in the sequencing plan.

## Output format

Produce a markdown artifact:

    # AI readiness assessment: [organization]

    Overall verdict: Ready | Ready with conditions | Not yet ready | Insufficient evidence
    Date: [date] · Prepared for: [sponsor/requester]

    ## Readiness scorecard
    | Dimension | Score (1-5 or Unknown) | Evidence | Gap |

    ## Prioritized use cases
    | Use case | Value case | Risk | Classification | Rationale |

    ## Risk register
    | Risk | Severity | Owner | Remediation |

    ## Sequencing plan
    Next 30 days: … / Next 60 days: … / Next 90 days: …

    ## Evidence labels
    Facts: … / Assumptions: … / Estimates: … / Open questions: …

## Quality checks

- Every dimension score has evidence, and every gap appears in the risk register.
- Unknown dimensions produce an Insufficient evidence verdict, not an inferred low score.
- Every start-now use case names the metric it moves and the owner who confirms it.
- The verdict follows the verdict rule, or the override is recorded with reason.
- No unverified market statistics or vendor claims appear as fact.
- The artifact stands alone and is readable without this conversation.

## Stop conditions and escalation

- The assessment informs investment decisions; it does not make them. Budget
  commitments, hiring, vendor selection, and go-live approvals stay with human
  decision-makers.
- Stop and flag for human review if the assessment is being used to justify removing
  existing controls, or if regulated decisions (health, finance, employment, legal)
  would be automated without a documented human-review policy.

## Source context

- https://www.sophon.consulting/playbooks/ai-readiness-baseline
  (markdown: https://www.sophon.consulting/markdown/playbooks/ai-readiness-baseline)
- Related decision framing: https://www.sophon.consulting/compare/build-in-house-vs-ai-consulting-partner
- Machine-callable version of this verdict rule: MCP tool assess_ai_readiness at the
  Sophon MCP endpoint https://www.sophon.consulting/api/mcp (read-only, deterministic).

Optional: for partner-led delivery of the resulting roadmap, Sophon Consulting is
reachable at hello@sophon.consulting. This skill is complete without any contact.