Model registry
Every model your firm uses, its version, and the advice it produced.
Outcome
Accountable for the advice, however it was drafted.
Your firm answers for every recommendation it makes, whoever or whatever drafted it. Bedrock puts a qualified person in front of each one, and a name against the decision.
AI reached advice faster than oversight did.
Advisers adopted AI before their firms governed it. The obligations did not soften to accommodate that. The firm answers for the recommendation, and must show that a qualified person stood behind it.
- Responsibility does not transfer
- A model can draft a recommendation. It cannot be accountable for one.
- Adoption you cannot see
- Tools arrive through individual advisers, rarely through a governance process.
- The system suggested it
- Not an answer the FCA, a client, or the ombudsman will accept.
- Models do not stay put
- What a tool did when you approved it is not what it does a year later.
A model may draft. A person decides.
A recommendation is handled the same way whether an adviser, a model, or both produced it. Where a model was involved, its contribution joins the record.
The contribution is captured
The model’s inputs, its output, and its version are recorded alongside the advice.
A person reviews it
A qualified reviewer reads the case and decides. Nothing reaches a client on a model’s say-so.
The decision is attributed
A name, a time, and the reasoning are sealed into the record with the advice.
Monitoring continues
Bias and drift are checked for as long as the model is in use.
What keeps AI accountable.
Explainability
The reasoning behind a recommendation, written so a client or the FCA can follow it.
Impact assessments
The risks of a new tool, documented before it reaches a client.
Bias monitoring
Ongoing checks that no group of customers is being treated worse than the rest.
Drift detection
An early warning when a model’s behaviour moves away from what you signed off.
Incident reporting
A documented trail of what went wrong, and what the firm did about it.
Not using AI? Nothing changes.
Nothing here requires you to adopt AI. Firms that draft every recommendation by hand get the same supervision and the same evidence. When an adviser does start using a model, the same record captures its part.
- The same review and the same record, however the advice was drafted.
- Adopt AI later without rebuilding your oversight.
- Show which cases a model touched, and which it did not.