The AI-Adoption Maturity Assessment

No exec owns AI; tech / data team running pilots in isolation.

Exec sponsor named but no operating cadence; AI shows up in board decks, not in a P&L.

Named CAIO / Head of AI with budget, KPIs, quarterly review cadence.

AI-adoption KPIs land in business-line P&Ls, not in a separate AI line.

AI is a stated input to corporate strategy, with explicit decisions about what not to automate.

Programme organised by model type (LLM team, ML team) — not by decision domain.

List of "AI use cases" exists; no shared framework for which decisions are even AI-suitable.

Programme organised by which decisions are in scope, with explicit boundary conditions.

Each in-scope decision has a documented human-deferral policy, error band, and escalation path.

Decision architecture reviewed quarterly against actual outcomes; decisions move in / out of automation as data demands.

No model inventory; no provenance tracking; data-access by exception.

Models tracked in a register; ownership assigned; PII handling documented.

Each production model has scheduled re-validation, drift monitoring, and a named risk owner.

Model risk is challenged by a function independent of the building team (audit, second-line, or external).

Governance pack would survive a regulator deep-dive on 30-day notice with no remediation work.

Tooling decisions made by sales conversations; no exit cost analysed.

Build/buy decisions made on instinct; no documented criteria.

Documented build-vs-buy criteria including total cost, switching cost, data-sovereignty risk.

Vendor portfolio reviewed annually; consolidation and decommissioning planned.

Active second-source strategy for any critical-path vendor; no dependency without a 90-day migration plan.

Models shipped to production; nobody trained the affected teams.

One-time training when a tool launches; no ongoing capability building.

Per-business-line AI champions exist with a forum and shared metrics.

Affected workflows formally redesigned to integrate AI, with measured throughput / quality outcomes.

Adoption measured (% decisions with AI assistance, % human override) and used to refine rollout.