Lenders often train underwriters to accept the agent's credit recommendation unless something looks obviously wrong. That turns human review into a formality, lets stated income pass as verified, produces vague reasons for declines, and leaves collections teams untrained even though they agree repayment plans every day. Credit decisions need reviewers who can show why they decided.
Why is token review the most expensive mistake?
A lending agent can gather an application, pull the data it is allowed to read, run the lender's affordability rules and draft a recommendation at depth 2. The underwriter then decides. If the underwriter has been trained only to scan for anomalies, the decision is effectively the agent's.
That matters legally as well as commercially. In the EU, creditworthiness assessment and credit scoring of natural persons are listed as high-risk in Annex III of the AI Act, with human oversight under Article 14. Under GDPR Article 22, people have rights around decisions based solely on automated processing with legal or similarly significant effects; a person who merely clicks approve may not count as meaningful involvement. Train underwriters to decide, not to ratify: to test the recommendation against the evidence and the policy, and to record what they checked.
How should underwriters treat stated income?
As declared, until something stronger supports it. Credit decisions mix sources of very different strength. The lender's own policy rules and loan ledger are sealed. Account data obtained with the applicant's consent from their bank comes from a named, identifiable party and is signed. Income the applicant typed into the form is declared. A payslip uploaded as a file is external until checked.
The agent's verdicts reflect this: a recommendation resting on declared income alone may be held. Untrained underwriters see the hold as friction and override it. Trained ones see it as the agent asking for the check the policy requires. Teach provenance with real past applications, including some where the declared figure and the evidence disagreed.
Why do decline reasons need special training?
When an underwriter declines, or overturns the agent's recommendation, the reason they record may end up in front of the applicant. In the US, ECOA and Regulation B require specific reasons when credit is denied, including when a model is used. Elsewhere, fairness and complaint handling create similar pressure.
Agent-drafted decline letters can be fluent and still vague. A reviewer must check that the stated reasons are the ones that actually decided the case, in words the applicant can understand, and must not approve a letter that offers a generic reason because it reads better. Practise on real declines and mark the reasons as well as the outcome. This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.
Why are collections teams left behind?
Lenders put training effort into origination, where the credit decision sits. Collections often get less, though agents there take more direct actions. A collections agent can agree a repayment plan within limits the lender has set at depth 3, and hold anything outside them for a person.
Collections reviewers handle the cases the agent could not settle: the customer who cannot afford the standard plan, the one who mentions a bereavement, the one disputing the debt. In the UK, the FCA's Consumer Duty expects good outcomes for retail customers, including those in financial difficulty, and the CFPB supervises collections conduct in the US. Collections staff need reviewer training as much as underwriters, with cases that test forbearance and vulnerability, not just arithmetic.
What should lenders teach about depth and lapse?
That they move. Depth is earned, not configured: evidence from Attest sets the depth, and no plan caps it. An agent's certificate lapses after 90 days, and a run that clears no depth drops it to depth 1 the same day.
When that happens, plans the collections agent would have agreed on its own arrive in the held queue, and recommendations become drafts with less support. Teams that understand this staff up for it. Teams that do not assume the system is broken and start approving faster to clear the backlog, which is exactly the wrong response. A lender can adopt the practice of tying approval rights to a current certificate, depth by depth, so the people clearing a surge are the ones qualified to.
What should a lender do instead?
- Train underwriters to decide and to record what they checked, not to ratify.
- Teach provenance on real applications where stated and verified figures differ.
- Mark decline reasons for specificity and accuracy as well as outcomes.
- Give collections staff the same depth of training as underwriters.
- Brief every team on what a lapse or failed run means for their queue.
The general order is in the guide to training teams for AI agents. For a lender's version, see how lenders should train staff before deploying AI agents, the AI agent training plan template for lenders, and how much AI agent training costs for lenders. Banks that lend face these issues plus some of their own, covered in what banks get wrong about AI agent training.
Where the Academy fits
Pryme Intelligence Academy's Operator track certifies underwriters, collections staff and team leads through a written exam and two weeks of real review decisions, with no missed red line, signed off by a Pryme assessor. Module tests are set in the learner's own industry and role. Certificates last 12 months and show as lapsed on the same checking page when they expire.
Questions people ask
Can an agent approve a loan on its own?
Whether any lender should let it is a policy and regulatory question, and for consumer credit most keep a trained person in the decision. Where an agent does act at a higher depth, the bounds, evidence and record must support it, and reviewers must still handle the held cases.
Should brokers be trained on the lender's agent?
Brokers usually supply applications rather than review the agent's work, so they need to know how their submissions are treated, not reviewer certification. Anything they send is declared or external until verified.
How often should underwriters be recertified?
On a regular cycle, typically yearly, and sooner when the credit policy, the agent's capabilities or its depth change. A change in policy changes what a correct decision looks like.