Underwriting is the process of assessing the risk of a loan, an insurance policy or another commitment and deciding whether to accept it, on what terms and at what price. In lending, the underwriter weighs the applicant's ability and willingness to repay against credit policy. In insurance, it is the risk of a claim. The underwriter owns the decision.
Why does underwriting matter once AI agents act?
Underwriting is a judgement with a name attached. Credit policy covers the common cases; underwriters handle what policy did not foresee, and they answer for their decisions to credit committees, auditors and regulators. Agents can do most of the preparation: gathering documents, checking them against policy, highlighting gaps and exceptions. The judgement on top remains the underwriter's.
It also matters downstream. A lender must be able to explain a decline, which is only possible if the underwriter's reasoning was recorded at the time, a point taken up in can an AI agent explain why a loan was declined? And underwriting quality shows up years later, in arrears and in decisions about whether to pass a debt to a collection agency. A loose approval today becomes a hard collections conversation later.
How does underwriting work in practice?
- Receive the application file, complete or with known gaps.
- Check it against credit policy: eligibility, limits, required evidence.
- Analyse the risk: income, commitments, history, security if any.
- Identify exceptions: anything outside policy and the case for allowing it.
- Decide: accept, decline, or accept on different terms.
- Set conditions: what must happen before funds are released.
- Record the rationale, so the decision can be explained and reviewed.
An agent can carry steps one to four at depth 1 or 2. Steps five to seven sit with the underwriter, often with a second-line check above a set amount.
What does that look like in an example?
A microfinance lender offers working-capital loans of around 500 to market traders. Its agent assembles each file: the trader's repayment history on earlier loans, group guarantee details, a photo of the stall, and sales notes from the loan officer's visit. It checks the file against policy and flags one exception: the trader has been at this market for four months, short of the six the policy asks for. The credit officer reads the file, notes strong repayment on two earlier loans at a previous market, approves with a smaller first loan, and records why. The small amount did not make the decision less real, a point made in agents in microfinance: small loans, same rules.
How is underwriting different from credit scoring?
Credit scoring is one model producing one number. Underwriting is the decision that uses that number alongside everything else, including judgement on exceptions the model cannot see. Some lenders let scores decide simple cases automatically under tested rules; even then, the rules were set by underwriters and the referred cases go to them.
Which related terms should you know?
- Credit policy: the lender's rules for what it will lend, to whom and on what terms.
- Underwriting exception: an approval outside policy, with a recorded reason.
- Maker-checker: a second person reviewing decisions above a threshold.
The wider limits on agents in lending are set out in AI agents in banking.
Questions people ask
Can an agent underwrite small loans automatically?
Lenders can auto-decide within rules they have tested and approved. An agent's judgement is not such a rule. If an agent is to decide, it should be certified for that capability on the lender's own cases and bounded tightly.
Does insurance underwriting raise the same questions?
Largely yes. Under the EU AI Act, risk assessment and pricing for life and health insurance is listed in Annex III as high-risk, which adds obligations on top.