Most of the new work in consumer lending sits around decisions agents are not allowed to make alone. Lenders using governed agents need underwriting reviewers, collections and hardship reviewers, AI operations analysts, agent owners and model risk and compliance staff. Credit decisions about people are heavily regulated, so each agent needs people who check, explain and answer for it.
Why does lending need so many people around its agents?
Credit decisions affect people's lives, and regulators treat them that way. Under the EU AI Act, creditworthiness assessment and credit scoring of natural persons are listed as high-risk in Annex III. In the US, ECOA and Regulation B require specific reasons when credit is denied, including when a model is used. In the UK, the FCA's Consumer Duty expects good outcomes for retail customers.
In practice, that means an agent may gather documents, check affordability data and draft a decision, but people review, explain and own the outcome. That is where the jobs are.
Which jobs exist around lending agents?
- Underwriting reviewer. Clears applications the agent held: income that does not match bank data, a thin credit file, a borderline affordability result. Writes the reason for each decision.
- Collections and hardship reviewer. Reviews payment plans the agent drafted, handles customers showing signs of vulnerability, approves forbearance outside standard terms.
- AI operations analyst. Reads the record across thousands of decisions, finds where the agent holds too much or approves too freely, and proposes rule changes with evidence.
- AI agent owner. Answers for one agent's scope, sources, limits and certification. Often a credit or collections manager.
- Model risk and compliance roles. Check that the agent's design, testing and monitoring meet the lender's model risk policy and conduct rules. US banks follow SR 11-7, UK banks PRA SS1/23 and Canadian federally regulated lenders OSFI Guideline E-23.
Builders who design and certify the agents also exist, often in a central team.
What does an underwriting reviewer actually decide?
Take a reviewer we will call Tariq at a mid-sized personal loan provider. The agent has processed a morning's applications and holds forty. One applicant's declared income is much higher than the salary visible in their bank data. The bank data came from an open banking feed, so it is signed by an identifiable party; the income on the form is declared.
Tariq checks the file and finds a second job paid into another account. He requests a statement, then approves the loan at a lower amount than requested, recording why. For a declined applicant, he checks that the reasons the agent drafted are specific and accurate, because the customer is entitled to them. Every decision is filed.
What stays with people in lending, whatever the agent's depth?
- Final decisions on applications outside policy.
- Treatment of vulnerable customers.
- Explaining a decline in terms the customer can act on.
- Changes to the lender's credit policy and limits.
Disbursing funds is depth 4 work, moving money, and lenders set their own limits on it. Those limits decide what the agent may do, whatever its certificate allows.
How do you move into these jobs?
Credit analysts, underwriters, collections agents and complaints handlers are well placed, because they already know the policy. Read adverts for duties such as "referrals", "exceptions", "quality assurance" and "decision review". The new AI operations jobs in consumer lending describes the roles in more depth. Country detail is in AI operations jobs in consumer lending in Australia and in Singapore. Related sectors include AI agent jobs in payments, and the guide to AI agent jobs and careers covers the rest.
Where the Academy fits
Pryme Intelligence Academy's Operator track trains the review work these jobs depend on, with a two-week practicum marked to nineteen in twenty correct and no missed red line. Some lenders using Pryme may make approval rights depend on a current certificate. The Architect track suits people leading regulated, multi-agent deployments, and is assessed by a written design and a board review.
Questions people ask
Can an agent decline a loan on its own?
That depends on the lender's limits and the law in its market. Many lenders keep a person on declines, and in the EU and UK, data protection law gives people protection against decisions based solely on automated processing that significantly affect them.
Is collections work safe from agents?
The routine reminders are moving to agents. Hardship, vulnerability and negotiated plans need people, and reviewing an agent's plans is a growing part of the job.
Do I need statistics to work in model risk?
For model validation, usually yes. For agent ownership and review roles, policy knowledge and careful judgement matter more.