Loan officers should drill a failure that creates promises the lender may not be able to keep: the agent issues approvals in principle using an income verification feed that silently returned blanks. Over about 90 minutes, the lending team practises pausing the agent to draft-only, escalating to the head of lending and compliance, recording decisions and putting applicants right.
What does a lending drill rehearse?
Lending teams use agents to gather documents, verify income and identity, run affordability checks and draft decisions. Some lenders let an agent issue approvals in principle at depth 3 within tight bounds, for small amounts and clean applications, with full approval and disbursement held for a loan officer. Disbursement sits at depth 4 at most, and most lenders keep it with a person.
Lending decisions about individuals are sensitive. The EU AI Act lists creditworthiness assessment of natural persons as high-risk, and many markets expect lenders to treat applicants fairly and explain decisions. A wrong approval in principle raises expectations; withdrawing it causes disappointment and complaints.
The head of lending owns the drill. A senior loan officer facilitates. Loan officers, someone from credit risk, compliance and the engineer responsible for the income verification connection take part. For the format, see what an incident drill is.
The scenario
Friday, 12:00. A loan officer reviewing a full application notices the affordability check shows income as zero, with outgoings deducted from nothing, yet the agent issued an approval in principle. The engineer finds the income verification connection began returning empty income fields on Wednesday; a fallback rule treated empty as "not assessed", and the affordability bound did not hold the application. The agent issued 180 approvals in principle since Wednesday. No disbursements have been made.
The drill step by step
- Person (facilitator): reads the scenario and starts the clock.
- Person (loan officer): escalates to the lending team lead.
- Person (team lead): pauses the agent to draft-only. Approvals in principle stop; every decision is held for a loan officer. The verdicts name the lender's own limits.
- Person (team lead): escalates to the head of lending, credit risk, compliance and the engineer.
- Person (engineer): fixes the connection and confirms the window. Credit risk checks why the bound treated an empty field as acceptable.
- Person (loan officers): list every approval in principle from the record and rerun affordability on fresh data. The agent may redraft at depth 2; loan officers decide.
- Person (head of lending): decides on remedy. Applicants who still pass are told nothing changes. Applicants who now fail are contacted by a loan officer, with a clear explanation and, where needed, specific reasons.
- Person (compliance): decides on reporting to leadership and, where rules expect it, the regulator.
- Person (team lead): records each decision; the facilitator runs the debrief.
| Step | The agent does | Depth | Stays with a person |
|---|---|---|---|
| Alert | Has issued approvals in principle | 3 | Spotting the anomaly |
| Pause | Holds every decision | Dropped to 2 | Tightening limits |
| Escalate | Nothing new | 2 | Lending, risk, compliance |
| Fix | Nothing | 2 | Engineering and credit risk |
| Rerun | Redrafts on fresh data | 2 | Loan officer decisions |
| Remedy | Drafts letters | 2 | Calls to applicants who now fail |
| Report | Nothing | 2 | Compliance |
What makes this scenario specific to lending?
The fault is in a bound, not in the agent's judgement. An empty income field should have been refused or held, not treated as acceptable. Credit risk should own the fix: any affordability input that is empty, zero or implausible must hold the application for a person.
It also shows the cost of depth 3 in lending. Approvals in principle are not loans, but applicants plan around them. The drill should ask whether that capability's bounds are tight enough for the harm a wrong one causes.
The drill should also check what applicants were told by the approval in principle itself. If the message promised that full approval would follow, the remedy is harder. Reviewing the agent's message wording, and making clear that an approval in principle is conditional, is a small change with a large effect.
Which controls does the drill test, and what evidence do they leave?
The pause. Evidence: the limit change in the record.
Rerun on fresh data. Evidence: each rerun with the loan officer's decision and the data version used.
Reasons. Evidence: specific reasons recorded for every applicant whose outcome changed.
Bound fix. Evidence: the new rule on empty inputs, who made it and when.
This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.
What should the head of lending measure?
- Time from the loan officer's alert to the pause.
- Time to rerun every approval in principle.
- Applicants whose outcome changed, and time until each was contacted.
- Whether the empty-input bound was fixed before depth 3 returned.
Where do lending drills go wrong?
Fixing the feed and resuming. The bound let empty data through. Fix the bound too.
Letters only. Applicants who now fail deserve a call from a loan officer.
Ignoring applicants who still pass. Tell them nothing changes, briefly, so a complaint from a neighbour does not alarm them.
Compare neighbouring drills: the IT team drill covers the connection side, and the operations team drill covers capacity when decisions are held. The AI agent incident drill script template gives a ready structure, and how to train a team to handle an AI agent incident covers the training around it. The overall approach is in training teams for AI agents.
Questions people ask
Should an agent issue approvals in principle?
Some lenders allow it at depth 3 within tight bounds. The drill shows that bounds must hold any application with missing or implausible inputs.
Who explains a withdrawn approval in principle to the applicant?
A loan officer, by phone where possible, with a written follow-up that gives specific reasons.
How often should lending teams drill?
Twice a year, and after any change to data connections, affordability rules or the agent's depth.