3 min From raw statements to an affordability score

01The problem

Most lenders decide affordability from what applicants type into a form. Stated income, stated expenses, stated obligations. Some of it is honest, some of it is optimistic, and none of it is verified at the moment the decision is made.

Checking the numbers by hand means an analyst opening bank statements, tracing deposits, and building a picture of cash flow line by line. That takes hours per file, and two analysts reading the same file often reach different conclusions.

So firms face a bad trade. Score fast from unverified form data and absorb the losses, or score carefully from statements and watch applications pile up. Volume makes it worse, because careful review does not scale with headcount alone.

02What we built

We built a system that reads affordability directly from banking transactions. The applicant connects their account or uploads statements, and the system takes it from there. Nothing depends on what was typed into the form.

• Transaction parsing that turns raw statements into structured cash flow: income streams, recurring obligations, existing debt payments, and the balance pattern between them.
• A two-stage credit model. The first stage screens out clear declines cheaply. The second stage scores the remaining applications in depth, so effort goes where the decision is genuinely close.
• A risk-based pricing engine on top of the score. Instead of one rate for everyone who passes, each approval gets terms matched to its measured risk.
• Every score comes with the inputs that produced it, so a human reviewer can see why an application landed where it did.

The rules are the lender's own credit policy, written into software. The system applies that policy the same way on the first application of the day and the thousandth.

03What changed

Raw statements now become an affordability score in about 3 minutes. What used to be an afternoon of analyst work happens while the applicant is still in the flow.

The score is built straight from banking data, so it reflects what actually moves through the account rather than what the applicant chose to report. Overstated income and forgotten obligations stop slipping through.

And it is consistent at any volume. A busy week no longer means shortcuts or backlog. Every application gets the same read, and the team spends its time on the borderline cases the model flags instead of retracing every file by hand.

Pricing followed. With a reliable measure of risk on every file, the lender prices to that risk instead of averaging it across the whole book.