Every credit decision stays with the loan officer, so onboarding teaches them to check what the agent prepared, not repeat it. Week one: how the agent gathers income evidence, calculates affordability and drafts memos and decline reasons. Week two: live applications, with a senior officer reviewing each decision. Sign-off rests on accurate income checks and specific reasons.
What does the lending agent prepare?
In a retail or small business lender that runs agents, the lending agent typically collects payslips, bank statements and bureau data, classifies income and expenditure, calculates affordability against the lender's policy, drafts the credit memo, and drafts decline reasons where an application fails (depth 1 and 2). It never approves or declines. In the EU, creditworthiness assessment of individuals is listed as high-risk under the AI Act, which brings human oversight duties; in the United States, ECOA and Regulation B require specific reasons when credit is denied, including when a model is used.
So a new officer arrives to applications that are mostly assembled. Their job is to test the assembly and make the decision. How the role shifts is described in what changes in a loan officer's day when AI agents arrive. The lending manager owns onboarding; a certified senior officer coaches.
What is the plan?
- Day 1. Person (lending manager): explains the credit policy and the agent's bounds: what it calculates, what it drafts, what it never decides.
- Day 1 to 2. Person (new officer): learns income provenance. Bank statement data from an open banking feed is signed. A payslip uploaded by the applicant is declared. Bureau data is signed by the bureau. The lender's own account data is sealed. The weight a figure carries depends on its class.
- Day 2 to 3. Agent, depth 2: drafts affordability and memos on live applications, decided by the coach while the new officer follows each check.
- Day 4 to 5. Person (new officer): reworks closed applications in a sandbox workspace, recalculating income from the documents and comparing with the agent's figures.
- Week 2. Person (new officer): decides live applications, each reviewed by the coach before the customer is told.
- Week 2. Person (new officer): records reasons on every change: income misclassified, assumption wrong, reason not specific, or approved exception.
- End of week 2. Person (lending manager): signs off decisions within the officer's mandate. Exceptions stay with the senior officer or credit manager.
| Step | The agent does | Depth | Stays with a person |
|---|---|---|---|
| Gather | Payslips, statements, bureau data | 1 | Asking for missing evidence |
| Classify | Income and expenditure | 2 | Checking classification |
| Affordability | Calculates against policy | 2 | Accepting the figure |
| Memo and reasons | Drafts | 2 | The decision, the reasons sent |
| Exceptions | Nothing | none | Credit manager |
Which checks should a new officer never skip?
One-off income. Bonuses, overtime spikes and transfers between the applicant's own accounts can look like regular income. Check each income line against its source.
Expenditure for this household. Standard assumptions may not fit a household with several dependants or high travel costs.
Specific reasons. "Did not meet affordability" is not specific. The reason should name the factor, so the applicant understands it and the lender meets its obligations.
Exceptions are not corrections. If the officer approves outside policy with a manager's sign-off, record it as an exception, not as an agent error. Otherwise the agent may be changed to treat the limit as soft.
Vulnerability alongside affordability. An applicant who mentions illness, bereavement or debt problems may need a different conversation, not just a different calculation. The agent should hold such applications for a person. A new officer must notice the signal when it appears in a payslip note or a free-text field, and follow the lender's process for customers in vulnerable circumstances.
How should corrections be recorded?
Every change to the agent's figures or wording gets a reason. Errors go to the agent owner, who can change the agent's brief, sources or bounds. Exceptions are recorded with the approver's name and kept out of the agent's changes. New officers who learn this distinction early protect the policy.
Which controls and evidence should onboarding leave?
- Reworked closed files. Evidence: recalculations and differences found.
- Reviewed decisions in week two. Evidence: both officers named on each.
- Exceptions tagged. Evidence: exception reasons with approver.
- Staged mandate. Evidence: dated authority within limits.
What should the lending manager measure?
Income errors the new officer caught. Decline reasons rewritten for specificity. Exceptions requested and approved. After sign-off: a monthly sample of decisions, complaints about declines, and early arrears on approved loans.
Where does it usually go wrong?
Accepting the affordability figure. It is calculated, so it looks authoritative. The sandbox rework teaches officers where it goes wrong.
Vague reasons sent unchanged. They read well and fail the applicant. Rewrite them until a stranger could act on them.
Exceptions logged as errors. The agent learns the wrong lesson.
Compliance and procurement colleagues onboard with related plans, for a new compliance officer and a new procurement officer. The AI agent training plan template for loan officers sets the learning out, and the wider approach is in training teams for AI agents.
Where the Academy fits
Pryme Intelligence Academy's Operator track teaches how a governed agent decides and how to review its drafts, with tests set in the learner's own industry and role, so a loan officer practises on lending cases. A lender can choose to tie an officer's decision mandate to a current Certified Pryme Operator certificate, valid 12 months and checkable at source.
Questions people ask
Can the agent tell an applicant they have been declined?
It can draft the letter. The decision and what is sent are the officer's, and the reasons must be specific.
Should the new officer explain the agent's role to applicants?
Where asked, and where law or policy requires, yes. Applicants should know when an AI system is involved.
How long until a new officer handles exceptions?
After a period of accurate decisions within mandate. Exceptions need experience of the policy's purpose, not just its rules.