AIMI is our own instrument. Its credibility with a technical buyer comes from being explicitly mapped to published research rather than asserted — present it that way, and never imply it is an industry standard.
The seven domains
| # | Domain | What it covers | Mapped to |
|---|---|---|---|
| D1 | Stance & governance | A communicated position on what AI may be used for, on what data, with what approval and audit trail | Clear AI stance; AI governance |
| D2 | Context & data accessibility | Whether standards, ADRs and interfaces are discoverable and machine-accessible to assistants | AI-accessible internal data |
| D3 | Engineering fundamentals | Version control discipline, batch size, test strategy, rollback capability | Version control practice; small batches |
| D4 | Verification & review | Review capacity, SLOs, depth, provenance-aware review, automated gates | The verification tax |
| D5 | Platform & pipeline | Pipeline reliability and speed, environment access, self-service | Quality internal platforms |
| D6 | Measurement & value | Baselines, telemetry contract, counter-metrics, attribution, cadence | Value stream management |
| D7 | Skills & user-centricity | Role-differentiated capability, explain-before-accept, atrophy controls | User-centric focus |
The five levels
Ad-hoc experimentation
Individuals use AI; no shared stance, no measurement. Shadow usage; nobody can state the policy; no baseline exists.
Standardised tooling
Licences, policy and access standardised; usage reported. Seat dashboards exist; value still unmeasured; review process unchanged.
Measured workflows
Baselines and counter-metrics in place; AI provenance captured; the constraint is known and named.
Governed, audited delivery
Guardrails enforced by the system; provenance audited monthly; evidence available to auditors; workflows redesigned around verification.
Continuous optimisation
Value stream managed continuously; interventions hypothesis-driven; the capability runs without external help.
A team scores a level only when every domain meets that level. Report the limiting domain alongside the level — "Level 2, limited by D4 and D6" is the sentence that sells the next phase, and it is also the honest one. A domain cannot be scored above 2 on interview evidence alone.
From score to roadmap
| Current | Do next, in order | Why |
|---|---|---|
| 1 → 2 | Publish a stance; consolidate tooling; capture a retrospective baseline before any further rollout | You cannot amplify what you cannot see |
| 2 → 3 | Telemetry contract, baseline per team, counter-metrics, provenance tagging | Usage reporting is the trap — adoption is an input |
| 3 → 4 | Verification system redesign, platform SLOs, automated guardrails, provenance audit | The J-curve dip is dominated by verification and pipeline adaptation |
| 4 → 5 | Value-stream management, hypothesis-driven intervention, internal ownership of the cadence | What converts individual gains into organisational performance |