Closed scope
Every development has an acceptance criterion defined before starting and a date on which it is checked. No open scope, no indefinite billing.
Three routes, sequential by design. Each answers a different question, and none makes sense before the previous one has been answered.
A programme run inside the company, with its full management team, on its own cases. Not training with worked examples: the actual work is the material.
Available training is canned — the same content for a component manufacturer, an insurer and an agency. It teaches tools. A management team's problem is not which tool to use, but which decisions of their business change now that the tool exists.
Once the board can evaluate, the question changes: it stops being what is possible and becomes where to start. Most AI projects fail by choosing the wrong problem, not by executing it badly.
Structured sessions with the actual owners of each process, not only with the board. We record how the process really runs, including the exceptions nobody documents.
Each pain becomes one or more candidate initiatives, described in enough detail to be estimated. The resulting catalogue is cross-cutting: it includes what no single area asked for because it affects several.
Each initiative is scored against explicit criteria — economic impact, urgency, technical feasibility, data dependency, organisational resistance — and placed on an impact-priority matrix. The score is arguable; arbitrariness is not.
Every prioritised project is planned three times: a minimum version solving most of it at the lowest cost, a standard one, and an ambitious one that takes on the whole problem. The board chooses with all three in front of it.
Order is set by dependencies, not by score alone: some projects cannot start until another has tidied a data source. The roadmap makes those dependencies explicit.
The projects the diagnostic prioritises get built. Each is designed around the company's real process, with its exceptions and its regulations, not around a sector template. This is where the laboratories come in: a development enters through the lab that owns its domain, and returns evidence to it.
Every development has an acceptance criterion defined before starting and a date on which it is checked. No open scope, no indefinite billing.
Code, data and generated knowledge belong to the company, delivered with enough documentation for another team to maintain them.
Which indicator must move, and how it will be measured, is agreed before the first line is written. If it cannot be measured, it does not get built.