When people talk about AI in lending, two very different things often get blurred together: answering questions and making decisions. The line between them matters more than almost anything else about how AI should be used in a building society.
It helps to think of lending questions in three kinds. The examples are illustrative.
Questions the criteria answer
Illustrative example
“What's the maximum loan-to-value on this product?”
The criteria say. The answer doesn't depend on judgement, only on finding the right line in the right document. Doing that quickly and consistently is useful, and it's a sensible job for AI, provided it answers only from the society's own criteria.
Questions that depend on the case
Illustrative example
“Would this applicant's second income be accepted?”
This sounds like a lookup, but the answer depends on facts about the case, such as how long the job has been held, and sometimes on discretion.
This middle ground is where AI tools are most likely to overstep. A tool that turns “it depends” into a confident yes or no is doing more than answering. I think the right response here is the rule, what it depends on, and a clear point where a person takes over.
Decisions
Illustrative example
“Should we lend to this person?”
This needs judgement, context and someone who can stand behind the outcome. In a regulated business, someone has to be able to explain a lending decision and own it. That's an underwriter, not a system.
Signs a tool has crossed the line
If you're looking at an AI tool for lending, a few things suggest it is deciding rather than answering:
- It gives a yes or no on whether an application will be accepted.
- It scores or ranks applicants.
- It uses words like “approved” or “eligible” about a specific case.
- It can't tell you which parts of its answer depend on discretion.
None of these is automatically wrong in every context. But each one moves the tool from helping people find information towards making judgements, and that deserves a much harder look.
Where we draw it
This is the first principle we're building AdAstr around. The underwriter's judgement stays with the underwriter. Our lending-criteria assistant helps people find the rule faster. It doesn't make the decision.
If you'd draw the line somewhere different, I'd genuinely like to hear why. Get in touch for a short conversation: hello@adastr.co.uk