Credit decisions come first. A lender's programme should train underwriters, collections leads and complaint handlers on the exact decisions their agents will hold for them, using the lender's own past applications. It should include a common grounding for leadership, a reviewer track with an exam and supervised practice, drills on declines and vulnerable customers, and renewable sign-off.
Why does lending need its own programme?
Lending decisions touch people's lives and are among the most closely watched uses of automation. Under the EU AI Act, creditworthiness assessment and credit scoring of natural persons are listed as high-risk. In the US, ECOA and Regulation B require specific reasons when credit is denied, including when a model is used. GDPR Article 22 limits decisions based solely on automated processing that have significant effects.
The shared consequence for training is simple. A person who reviews an agent's credit work must understand it well enough to stand behind the outcome and explain it. This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.
Which roles need which depth of training?
| Role | What they review | Training |
|---|---|---|
| Credit committee and executives | Appetite, red lines, the record | Common grounding, half a day |
| Underwriters | Held approvals, every decline, exceptions | Reviewer track with practicum on past applications |
| Collections team leads | Arrears letters, payment plans, forbearance | Reviewer track with drills on vulnerable customers |
| Complaints handlers | Draft final responses | Reviewer track, focused on fairness and reasons |
| Credit risk and model risk | Agent design and certification | Builder track |
Model risk teams are often left out because they already know models. They still need to learn how an agent's depth is earned per agent, and why the certificate lapses after 90 days.
What may the agent do, and what stays with the lender's staff?
- Depth 1: read the application, bureau data and documents.
- Depth 2: draft the affordability summary, the decision rationale and customer letters.
- Depth 3: request missing documents and set up agreed payment plans within the lender's rules.
- Depth 4: disbursing a loan moves money; it is held for a named underwriter unless the agent is certified at that depth and the amount sits inside the lender's own limit.
- Stays with a person: every decline, every forbearance decision for a customer in difficulty, and every exception to policy.
An example: an affordability check on uploaded statements
Take a consumer lender where an agent reads uploaded bank statements to assess affordability. A statement uploaded by the applicant is declared provenance; a feed from the applicant's bank, where available, is sealed. In a practice case, the agent's summary shows comfortable income, but the figures came only from a declared statement whose layout differs from the bank's usual format.
The trained underwriter checks the provenance line before the summary and asks for verified data. The untrained one reads a tidy summary and approves. Practice should be built from cases like this, drawn from the lender's own files, with reasons recorded for every decision.
What controls keep reviewers sharp after training?
Training is the start, not the control. Lenders should calibrate reviewers monthly by having two people decide the same sample of held cases and discussing every disagreement. They should track how often each reviewer overturns the agent, since both very high and very low rates point to a problem. And they can make the right to approve at each depth depend on a current certificate, so a lapsed reviewer cannot release a disbursement.
Pryme Intelligence Academy's Operator track fits underwriters and collections leads: a 40-question exam passed at 32 of 40, then two weeks of real review decisions with no missed red line, signed off by an assessor.
For the schedule, use the AI agent training plan template for lenders. Before you begin, read what lenders get wrong about AI agent training. Lenders running several branches can scale with a train-the-trainer programme, and wealth managers face a similar split between drafting and advice. The wider method is in training teams for AI agents.
Questions people ask
Should collections staff train before underwriters?
Train whoever will see agent work first. Many lenders start agents in collections letters because the drafts are routine, which makes collections leads the first cohort.
Can a trained reviewer approve a decline the agent drafted?
Yes, if they have checked the reasons against the evidence and can explain them to the customer. The decision is theirs, and the record shows it.
How do we keep training current when credit policy changes?
Treat a policy change as a trigger for a short update module and a fresh practice set. Reviewers should not approve under new rules using judgement trained on old ones.