Made-up example. Northbridge Bank is not a real bank, and every number here is invented to show how AI EnableOS works.
2
Step 2 of 6
Evaluate
Check data, risk and feasibility. Build or buy.
1. Questions asked
The AI wrote these questions from similar real projects. The answers are made up.
Q: What data would the AI need?
A: Only the rulebook and policy updates. No customer data.
Q: What is the worst thing that could go wrong?
A: A wrong answer after a policy change. It must never see customer accounts.
Q: Build or buy?
A: Build on the bank's approved AI platform. It is already approved, so it is faster than buying a vendor tool.
Q: Who must approve the risk?
A: The model-risk team.
2. What we found
- Data it needs
- Rulebook and policy updates, no customer data needed for the lookup
- Risks
- Wrong answers after a policy change; must not see customer account data
- Build or buy
- Build on the bank's approved AI platform. Faster than buying a new tool.
3. AI recommends, a person decides
AI recommends Go
Medium risk. Go, if the assistant can only read the rulebook and never customer accounts.
- No customer data is needed for a rule lookup.
- The main risk is an old rule after a policy change, so updates must be loaded the same day.
- The platform is already approved, so build is faster than buy.
A person decided: Risk lead (sign-off 1)
4. Sign-off 1: Compliance and risk review
Who signs
Risk lead
Evidence they looked at
- List of data the assistant can read: rulebook and policy updates only
- Risk level: medium, with reasons
- Plan to load policy changes the same day
Sent back to fix if
The assistant could reach customer accounts, or nobody owns policy updates.
Stopped if
The lookup needs customer data and there is no approved way to protect it.
Signed: Approved. Condition: read-only access to the rulebook.