A creditworthiness assessment is a lender's evaluation of whether a borrower is likely to repay a credit agreement as agreed, based on their financial position, credit history and circumstances. It decides whether credit is offered and on what terms. A credit score is one input to it; the assessment is the lender's whole judgement.
Why does it matter once AI agents act?
The EU AI Act lists creditworthiness assessment of natural persons in Annex III as a high-risk use. An AI system used to evaluate whether a person is creditworthy carries obligations: risk management, data governance, logs, transparency, human oversight, and for providers, a conformity assessment before the system is placed on the market, covered in EU AI Act conformity assessment and AI agents.
An agent that gathers the file and recommends a decision may be that system, or part of it. The question is what its output does. If a person reads its summary and decides on the full file, the agent assists. If its recommendation is followed as a matter of course, it is assessing creditworthiness in substance. Can an AI agent score a customer's creditworthiness? works through that line.
How does an assessment work in practice?
- Gather evidence: application, bureau data, bank statements, accounts for the self-employed.
- Verify it: compare declared figures with evidenced ones and note gaps.
- Analyse: income stability, existing debt, repayment history, purpose of the loan.
- Judge: likely to repay, on these terms or different ones?
- Decide and record: the outcome, the reasons and the evidence behind them.
Agents carry steps one to three well, at depth 1 or 2. Steps four and five are where a lender's accountability lives, and they stay with a person or with decision rules the lender has approved and tested.
What does that look like in an example?
A lender to small businesses receives an application for 20,000 from a self-employed plumber. Because the borrower is a natural person, the lender treats the assessment as within Annex III, which it would not for a limited company. The agent pulls two years of tax returns and twelve months of statements, matches declared income to deposits, and notes that income drops each winter. It drafts a file note recommending a shorter term. The underwriter agrees and records why.
That natural-person line is easy to miss in business lending. Sole traders and partners are people, even when the loan is for their trade.
How is it different from an affordability assessment?
A creditworthiness assessment asks whether the lender will be repaid. An affordability assessment asks whether the borrower can repay without hardship. A customer with high income and heavy spending may be a good credit risk and still be stretched by a new loan. Some rules fold affordability into creditworthiness; the questions remain distinct, and a lender should be able to answer each. The affordability assessment checklist covers the second.
Which related terms should you know?
- Credit scoring: the model that predicts repayment, an input to this assessment.
- Underwriting: the decision process that uses the assessment to set terms.
How agents fit into lending decisions more widely is covered in AI agents in banking.
Questions people ask
Does the EU AI Act ban AI in creditworthiness assessment?
No. It classifies the use as high-risk and attaches obligations. Lenders may use AI for it if those obligations are met.
Who is responsible, the tool's provider or the lender?
Both, for different things. The provider answers for how the system is built and documented; the lender, as deployer, answers for how it is used, overseen and monitored.