Inclusive AI underwriting: Is this the year credit unions win the loan book banks won't touch?

Credit union loan declines come in two kinds. The members you shouldn't lend to, rightly turned away. And the members you could have lent to safely, if the bureau model had been able to see them clearly enough to say yes. That second pile, the wrong-declines, is where the entire credit union mission lives. It's the member a high-street bank refused the same week, who could have repaid, and who walked away with nothing.

In our free AI Disruption Risk Index for UK credit unions, inclusive AI underwriting is the highest-scored opportunity in the whole report, at 87 out of 100. Because AI finally changes the maths on that pile: approval rates rise without default rates rising with them.

 

Three things changed, and now the sums work

 

This has been the dream of inclusive lending for decades, stuck because the tools weren't there. Three things moved.

AI can now read the signal you already hold and a bureau score never sees: member tenure, savings discipline, share contributions, the stability of a current account you've watched for years. None of it feeds an underwriting decision today, and it's exactly what separates a safe borrower from a thin file. Explainability has caught up, so the model produces a rationale a loan officer, a member and a regulator can all read, which keeps it Consumer-Duty-grade rather than a black box no board would sign off. And the build cost has come down to credit-union size. A twelve-person firm can run a shadow agent for a single quarter and walk into the year-end board meeting with evidence instead of a hunch.

 

What it looks like at the desk

 

A loan application arrives. The bureau check comes back decline. The agent reads the member's two years of savings, current-account behaviour and prior repayment, and recommends approval at £3,600 over eighteen months. The loan officer reads a four-line rationale, agrees, and approves. The member, turned down by their high-street bank that same week, gets a loan they can actually repay. The credit union earns the interest, builds the relationship, and proves the case for inclusive credit to its own board, with a live example rather than a principle.

Nothing in that replaces the loan officer. The judgement stays human; the agent just makes the case visible. It simply stops you saying no to people you were built to say yes to.

 

The return, and why defaults don't climb with approvals

 

The report puts indicative figures on it for a firm of that size. Year-one investment of £40,000 to £70,000, against £90,000 to £180,000 of additional safe loan book in the first year. The reasoning is concrete, not a flourish: a twelve-staff firm declining roughly 200 to 400 applications a year, of which 8 to 15 per cent turn out to be safely approvable once member data is weighted in, is 16 to 60 additional loans at a typical credit-union ticket of around £3,000 over eighteen months.

The number that should reassure a board is the one that doesn't move. Defaults stay inside your existing risk appetite, because the agent is validated against your real outcomes in a shadow run before it ever decides anything live. This is growth from lending more carefully, not less, in the segment you understand better than anyone.

 

What it changes for members, mission and growth

 

Sit with what that one approval means on three counts. For the member, it's access to fairly priced credit they would otherwise be refused, often the biggest financial decision of their life made possible instead of denied. For the credit union, it's the mission made operational rather than recited in the annual report: financial inclusion you can count, member by member. And for the loan book, it's growth in a segment the neo-banks and the high-street banks don't operate in, with net interest margin defended against the rate cycle, in lending you understand better than any competitor. It is the rare move that serves the member, the mission and the balance sheet in the same decision.

 

The loan book you were built to write

 

So the question for the year isn't whether to dabble in AI. It's whether this is the year you start winning the lending the neo-banks and the high-street banks structurally will not serve, the loan book that is yours almost by definition, sitting in your wrong-decline pile waiting for a tool that can finally see it. It is also the cleanest answer there is to the neo-bank pressure, won on the ground they can't compete on.

The full opportunity, with the methodology behind the figures and the worked example in full, is in the free UK edition, run through an anonymised £28m credit union.