Most AI risk sits in the space between the tool and the decision it informs.
It starts earlier than most organisations look: in how the data was collected, what it leaves out, and who answers for what the tool recommends. We look at all three, before the next tool goes live.
Every review starts in the same three places.
- DATA
Where did it come from, and what does it leave out?
- DECISIONS
Which decisions does AI shape, and who signs them off?
- PEOPLE
Who is affected, and could they challenge the outcome?
Every finding is marked in one of four states.
- HeldIn place, owned and working.
- WeakExists, but informal or inconsistent.
- GapMissing, and able to cause harm.
- UnseenNot yet assessed. Often where the largest risk sits.
From tools adopted team by team to decisions you can explain.
Compliance is decided before you build. Most find out at the audit.
Problems rarely show up where they start. That is why we begin with the data.
If something was never recorded, no amount of governance afterwards can prove what it would have shown. A dataset built only from weekday records cannot tell you what happens at night, and every decision made on it quietly fails the people it never saw. That is not a model problem. It is a collection choice, made long before the model.
And the data you did not collect yourself still counts. When you rely on a tool trained by someone else, their blind spots become your decisions.
You cannot audit your way back to data that was never collected.
See it. Trace it. Specify it. Hold it.
- 01
See
We list where AI already touches your data and your decisions, including the tools nobody approved, and map who really decides.
- 02
Trace
We follow real decisions back to the data underneath them, and mark where ownership, data quality or human judgement is missing, before it causes harm.
- 03
Specify
We turn each obligation into a requirement your teams can build and test: what the data must hold, where it is captured, and who signs off.
- 04
Hold
Your team runs it. We return at set intervals to check the structure still holds as new tools arrive.
WHERE WE SIT
- Not a technology vendor. We do not sell or build tools.
- Not a law firm. We work alongside your legal advice, and turn it into something your teams can build.
- The layer in between, where most AI governance fails.