Yes, if it knows where the statements came from. An agent can find regular income in bank statements, separate salary from transfers, and cross-check it with payslips. The figure's weight depends on the source: data fetched from the customer's bank is strong; an uploaded PDF can be edited. Irregular income and anything that looks altered go to a named person.
Why does the source matter more than the reading?
A bank statement uploaded as a PDF is a document the customer controls. Editing one to show a higher salary is easy, and fraudsters sell templates for it. An agent that reads the PDF accurately has only accurately read what the customer chose to show.
Statement data obtained with the customer's consent directly from their bank, through open banking where it exists, is different. It comes from the institution that holds the account, and the customer cannot alter it in transit. The same transaction list carries more weight. That is why bank agents need provenance on every instruction, and why the agent records the source with every income figure it produces.
What does the agent check?
- Regularity. Salary usually arrives on a similar date each month from the same payer. The agent identifies the pattern and the payer.
- What is not income. Transfers from the customer's own savings, refunds, loans, and money from friends can look like income. The agent excludes them and records why.
- Consistency with payslips. Net pay on payslips should match the credits. Differences are flagged.
- Changes over time. A new job, a pay cut or a gap in income changes the picture. The agent reports the pattern, not just the average.
- Signs of alteration. For uploaded PDFs: fonts that change mid-line, running balances that do not add up, dates that fall on non-working days for the payer.
What may the agent do, and what stays with a person?
| Situation | The agent | Stays with a person |
|---|---|---|
| Regular salary, statements from the bank | Calculates and records verified income, depth 3 | Nothing routine |
| Regular salary, uploaded PDF | Calculates and flags the source | Accepting it for larger loans |
| Running balances that do not reconcile | Holds and records the discrepancy | Fraud review |
| Self-employed or irregular income | Summarises credits by source and month | Judging sustainable income |
| Benefit or pension income | Identifies it from the payer | Applying the lender's policy on it |
| Income from abroad | Records currency and payer | Policy on foreign income |
| Deciding whether the customer can afford the loan | Never | The affordability assessment |
Self-employed applicants often supply accounts as well as statements; preparing those is a separate task, covered in whether an AI agent can prepare financial statements.
An example: two applicants, two sources
Take two applicants for similar personal loans, each claiming a monthly salary of 3,200.
The first connects their bank through the lender's consent flow. The agent sees three months of salary credits of 3,180, 3,200 and 3,210 from the same employer on the last working day of each month. It records verified income of 3,197 a month, with the source marked as bank data.
The second uploads three PDFs. The salary lines show 3,200 each month. But the running balance after one salary credit is 1,800 higher than the previous balance plus 3,200 would allow, and the font on that line differs slightly from the rest. The agent does not conclude the document is fake. It records the discrepancy and holds the application for the fraud team. A person asks the applicant to connect their bank instead, and the real salary turns out to be 1,400.
What controls make it safe?
- Source recorded with every figure. Bank-sourced data and uploaded documents are labelled differently in the file.
- Arithmetic checks on documents. Running balances are recalculated, and mismatches hold the case.
- Exclusion reasons recorded. Every credit excluded from income carries the reason.
- No affordability decision. The agent produces the income figure; the assessment uses it.
- Sensitive spending ignored. The agent reads credits for income and does not infer sensitive traits from the debits it sees.
Income is one half of the affordability picture; the other is expenditure, covered in running income and expenditure assessments with an AI agent and checking income and affordability evidence with an AI agent. Identity checks usually come first: see whether an AI agent can verify a customer's identity documents. The general limits are in AI agents in banking: what they may and may not do.
Questions people ask
How many months of statements does the agent need?
That depends on the lender's policy for the product. The agent applies the policy and asks for more when income is irregular.
Can the agent verify income for a joint application?
Yes, for each applicant separately, with each one's consent for their own statements.
Can the agent contact the employer to confirm salary?
Only through a process the lender has set up and the applicant has agreed to. Most lenders use payslips and statements instead.