Twenty questions, four dimensions, twelve minutes. You get a score out of 60, a band, and the one thing worth fixing first. Nothing is transmitted anywhere — the whole thing runs in your browser.
Not what you intend, not what the policy says. The score is only useful if it's honest, and almost everyone scores lower than they expect — which is the point of taking it.
We'll send your result, the four-tier data classification card, and the one-page AI policy you can hand to information security. No sequence, no upsell drip — three emails over three weeks, then nothing unless you ask.
| Band | Score | What it means |
|---|---|---|
| Exposed | 0–15 | AI is happening around you without structure. The immediate risk isn't falling behind — it's that something goes wrong on a project you're accountable for and you have no defensible position on how it was used. |
| Ad hoc | 16–30 | You use it, it sometimes helps, none of it is repeatable. Where most project managers sit in 2026. The gains are real but invisible, and they don't survive you going on leave. |
| Deliberate | 31–45 | You have habits and some rules, and you're ahead of most of the profession. What's usually missing is a verification step that runs every time, and evidence you could put in front of a sceptical stakeholder. |
| Systematic | 46–60 | Rare. You're in a position to set practice for other people, which is the highest-value thing anyone can do with this right now. |
Fluency — can you actually get useful work out of it.
Governance — do you know what's safe to put in.
Workflow — is it in your process or improvised each time.
Evidence — can you prove any of it to someone who asks.
It's only useful compared against itself. Retake it in thirty days after working on your weakest dimension. A jump of 8–12 points in a month is normal, and that delta is the number worth reporting — not the absolute score.
The prompt library is 30 prompts built for the artefacts you already produce — status narratives, risk logs, scope memos, stakeholder updates, closure reports. Every one carries a note on what good output looks like and what the model reliably gets wrong.