Yes, provided it explains the reasons recorded when the decision was made. An AI agent can put those reasons into plain words, say what the applicant can do next and tell them how to ask for a review. It must not reconstruct or guess reasons afterwards. Disputes, human reviews and reconsideration stay with a named person.
Why must the explanation come from the record?
A decline explanation is only true if it describes the real decision. If the agent reads the file afterwards and works out what it thinks the reasons were, it may produce something plausible and wrong. The applicant then acts on the wrong reason: pays down the wrong debt, disputes the wrong entry, or reapplies with nothing changed.
There are legal duties too. In the United States, ECOA and Regulation B require specific reasons when credit is denied, including when a model is used. GDPR Article 22 and its UK equivalent give people safeguards against decisions based solely on automated processing, including the right to obtain human intervention and contest the decision. A made-up reason cannot meet either standard.
So the condition is simple. The reasons must be captured at the moment of decision, by the model or the credit officer, and stored with the decision. The agent explains those, and nothing else.
What are the conditions?
The agent can explain a decline when:
- the decision record contains specific reasons, recorded at the time;
- the person asking is the applicant, authenticated, or someone with authority to act for them;
- the explanation stays within the recorded reasons and the lender's approved wording for each;
- the applicant is told how to request a human review.
If the record has no specific reasons, the agent says the explanation will come from a person and passes the request on.
What may the agent do, and what stays with a person?
| Task | The agent | Stays with a person |
|---|---|---|
| Explaining the recorded reasons | Explains them in plain words | Nothing routine |
| Saying what the applicant can do | Points to general steps tied to each reason | Personal financial advice |
| Explaining credit file entries | Explains which file the lender searched and how to get a copy | Disputes about the entries |
| Requests for human review | Records the request and assembles the file | The review |
| New evidence | Records and attaches it | Reconsideration |
| Complaints | Logs and routes them | Handling them |
| Explaining a decision with fraud indicators | Gives only the approved wording | Everything else |
Fraud cases show why the last row matters. Where a decline involved suspected fraud, the lender may be limited in what it can say. The agent uses the approved wording and nothing more.
The method for writing these explanations is covered in how to explain a credit decision to an applicant with an AI agent.
A worked example
Take a lender whose approved model declined an application for a 6,000 personal loan. The decision record holds two reasons: the applicant's income could not be verified from the documents provided, and the time at their current address was below the policy minimum.
The applicant messages: "Why was I turned down?" The agent explains both reasons in plain words. For the first, it says which document was missing and that the applicant can reapply with it. For the second, it explains that the lender has a minimum time at address, without stating the threshold if policy treats it as confidential. It tells the applicant they can ask for a person to review the decision.
The applicant uploads the missing payslip and asks for a review. The agent records both and passes the file to a credit officer who was not involved. The officer reviews the case with the new evidence and decides.
What controls make it safe?
- Reasons are captured at decision time. The model or officer records specific reasons; the agent only reads them.
- Approved wording per reason. Each reason code maps to plain-language wording signed off by credit and compliance.
- No new reasons. The agent cannot add reasons, soften them or speculate beyond the record.
- A review route in every explanation. The applicant always learns how to reach a person.
- A full record. What was explained, to whom and when is filed with the decision.
Explaining is different from deciding; for that, see whether an AI agent can approve a loan. For where explanations sit in the process, see what loan origination is. The wider limits are in AI agents in banking: what they may and may not do.
Questions people ask
Can the agent tell the applicant what score they got?
That depends on the lender's policy and the law in its market. Many lenders explain the factors rather than give a number. The agent follows the approved policy.
Can the agent tell the applicant whether they would be approved if they reapplied?
No. That would be a new decision. The agent can say which recorded reasons might change if the applicant's circumstances change.
Can an applicant ask the agent to correct their credit file?
The agent can explain how to contact the credit reference agency and record the request. Corrections are made by the agency, and the lender handles its own reporting through its normal process.