Building an 18-month AI adoption roadmap that survives contact with reality

Roadmaps fail in SMEs for a predictable reason: everything is phase one, because everything feels urgent. The fix is exit criteria — a phase is not finished because time passed, but because a condition was met.

Phase 1 — Prove (months 0–6)

• One internal use case with a measured baseline.
• Governance basics in place: register, policy, oversight rule.
• Exit criteria: measurable time or cost saving, and at least three people using it weekly without prompting.

Phase 2 — Embed (months 6–12)

• Integrate the proven use case into the system of record so it is not a side activity.
• Fix the data issues the pilot exposed — now you know which ones matter.
• Train a second cohort and appoint an internal champion per team.
• Exit criteria: the saving persists for two consecutive quarters without special attention.

Phase 3 — Extend (months 12–18)

• Add a customer-facing use case, with human oversight from day one.
• Review vendor concentration and cost as usage scales.
• Reassess readiness scores and re-plan from the new baseline.

One rule keeps roadmaps honest: nothing from the next phase starts until the current phase meets its exit criteria.

Reassess every six months. Readiness scores move, tooling changes, and a roadmap that is never revised is a document rather than a plan.