Underwriting a business that keeps no accounts.
Four stages, in order, each one narrowing what the next has to take on trust. None of them replaces the credit officer at the end.
Illustrative. The same scene a visiting officer sees, recorded so it can be read consistently rather than remembered differently by each assessor.
The premise
Evidence beats attestation.
A small business that files no audited accounts is not an opaque business. It is a business whose record sits in places lending has never bothered to look: the goods on the shelf, the machines running, the meter, the payment acceptance at the counter, the invoices on the desk, the returns already filed.
Each of those signals is weak on its own and easy to dismiss. Read together, and read against data the borrower does not control, they describe a business well enough to lend against — provided the loan is also secured.
The work is making that reading cheap, repeatable and consistent between one officer and the next. That is an engineering problem before it is a credit problem.
The sequence
Four stages, each narrowing the next.
Nothing here is exotic. The work is doing it the same way every time, at a cost that lets a small loan carry it.
Observe
The business is documented where it operates — premises, stock, equipment, workspace and footfall. Not a form filled in an office, but a record of the place itself, captured so it can be reviewed again later by someone who was never there.
Corroborate
What was observed is read against the rails the business already uses and does not control: its registration, its filings, its consented banking and payment history, its bureau record. A claim has to survive contact with independent data.
Test
A structured discussion with the owner, in his own language, on the record. What he says about his own business is compared against what the evidence already shows — not to catch him out, but so that agreement and disagreement are both visible to the decision-maker.
Decide
Everything assembled goes to a qualified credit officer with the conflicts surfaced rather than smoothed over. The officer approves or declines, and is accountable for that decision. The file explains itself to a regulator, an auditor and the borrower.
The system builds the case. A person signs it.
Automating the assembly of evidence is an efficiency. Automating the credit decision is a transfer of accountability we are not willing to make.
Deliberate limits
What we are choosing not to build.
Constraints are easier to hold when they are written down before the pressure arrives.
No unsecured lending to this customer
The last cohort that tried to lend on data alone to small enterprises learned that information reduces adverse selection and does nothing for recovery. We take the evidence and the property.
No black-box approval
Every decision has to be explainable to the borrower who received it and to a supervisor reviewing it a year later. A model no one can interrogate is a liability the lender still owns.
No origination paid only at disbursal
Paying for volume at the front produces volume at the front. Compensation that continues across the life of the loan makes the introducer care what happens in month eighteen.
No growth ahead of the licence
We will lend when we are authorised to lend, under the framework that applies to us, and not before.
Want the technical version?
There is a longer explanation of the architecture and the evidence model for people who want to interrogate it properly. Ask, and we will walk you through it.