£150K AI programme approved by the board in a single meeting.

The company had strong engineering, decent data hygiene and a leadership team that broadly agreed AI mattered. What they didn't have was a shared, costed view of where AI actually belonged in the roadmap versus where it was fashionable to talk about.

The board had already deferred the AI conversation twice. A third deferral wasn't going to look good, but neither was walking in with a slide deck full of adjectives.

Three weeks of structured interviews across product, engineering, customer success and operations produced a candidate list of use cases. Each was documented in the same one-page format: problem, evidence, effort, cost, risk, dependency, expected outcome.

Twelve use cases made it through basic viability screening. Four were flagged as quick wins — meaningful, buildable within a quarter, low integration risk. The rest were sequenced into a phased roadmap with explicit 'not yet' criteria.

Governance was drafted alongside the roadmap: model use policy, data handling, review cadence and the small internal group that would own decisions between board meetings.

The reason the board approved was not the slide count — it was the docket. Every recommendation had a cost, a risk and a named owner before it entered the room. 'Interesting idea' had already been filtered out during week two.

The team asked for a leaner audit template for the next round. That's fair; the second time is always shorter because the language is already agreed.