Evidence Gates Instead of a Fixed Scope¶
The problem¶
A prototype lands well. Leadership asks what it would take to make it real. The default answer is a scoped programme with a number attached — twelve months, a team, a delivery plan.
That answer usually fails, and it fails for a structural reason rather than a commercial one. After a prototype, the biggest risks in an enterprise AI programme are not build risks. They are:
- Nobody owns the semantic definitions the system depends on.
- Architecture and security review have not yet been passed, and will impose changes.
- It is unknown whether the business will change how it works to use the answers.
A fixed scope prices the build and silently assumes the other three resolve themselves. When one does not — and one will not — the programme is already committed, and the overrun looks like a delivery failure rather than what it was: an unexamined assumption.
The pattern¶
Structure the phase as a sequence of gates, each buying a specific piece of evidence, each with an explicit go/no-go criterion.
| Gate | Buys | Go criterion |
|---|---|---|
| Discover | An accurate map of sources, owners and question demand | Question distribution is known and sourced from users, not assumed |
| Align | Named owners for the definitions the system depends on | Each core concept has a business owner who accepts the role |
| Enable | A measured quality baseline on the business's own questions | Competency question set exists, is owned, and scores above an agreed threshold |
| Harden | Architecture and security approval | Review passed on the target design, not on the prototype |
| Scale | Adoption inside a real operating workflow | Usage in a live process, with a measured effect on it |
The argument to leadership becomes: the next phase buys evidence, alignment and options — not another demo. Each gate is individually cheap, individually justifiable, and permitted to fail. A no-go at Align costs one phase instead of a programme.
Why it survives scrutiny¶
It is a deliberately modest argument, and it is stronger for it. It concedes the thing every sponsor already suspects — that a demo does not prove an enterprise system — and replaces an act of faith with a sequence of decisions.
It also puts the hard organisational work early, where it belongs. Semantic ownership and security review are the two things most likely to stop a knowledge platform, and a fixed-scope plan reliably schedules both of them last.
When this does not apply¶
- Where the outcome is genuinely known and the work is execution. Gates add ceremony to a problem that does not have the uncertainty they exist to manage.
- Fixed-price commercial models that cannot accommodate a no-go without a contractual fight. The pattern needs a client able to stop.
- Where the sponsor needs certainty more than accuracy. Some organisations will fund a twelve-month plan and not a five-gate sequence, for reasons that are political rather than analytical. Recognising that early is worth more than winning the argument.
Related¶
- A governed knowledge platform for revenue growth management — the prototype this case was built on
- Competency questions as an evaluation gate — the instrument the Enable gate depends on