AI in collections: the choice leaders now face

The Building Societies Association (BSA) just said the quiet part out loud.

In its July 2026 strategy report, "Finance for a Fairer Future," it drew a line that should stop every collections leader mid-scroll: AI built to improve customer outcomes and AI built to extract value will give you different answers to the same arrears case.

That's the real question behind AI in collections. Not whether to use it. What you point it at.

What the BSA report argues

The BSA's warning is specific. Models tuned for scale, margin and low-cost conversion tend to favour standard, profitable customers while 

“people with irregular incomes, weak credit histories, thin savings or complex needs get excluded, poorly served, or charged more”.

Its conclusion is blunt. AI built for customer outcomes gives "fundamentally different recommendations" than AI built to squeeze commercial value from a customer. Same technology. Often the same data. Different objective, different result.

For anyone running collections, that lands close to home. The customers most likely to be underserved by a commercially-led model are the ones you meet every day in arrears.

Why this bites hardest in collections

Collections is where the gap between those two objectives is widest.

A customer in arrears is often in a difficult moment. The FCA’s four drivers of vulnerability – health, life events, resilience and capability - describe a large share of any collections book.

An AI optimised purely for recovery will push for payment. An AI optimised for outcomes might recommend a pause, a reduced payment plan, a referral to support. This aligns with the Consumer Duty requirement to deliver positive outcomes and avoid foreseeable harm.

The forbearance rules make it sharper still. The FCA’s CONC 7.3 requires firms to treat customers in or approaching arrears with forbearance and due consideration, and to offer sustainable, affordable options. An automated system that can't do that, or can't provide an audit trail to prove it did, is a liability.

The regulator is already testing AI in collections

This isn't theoretical. The FCA’s AI Live Testing programme explicitly names debt resolution as a use case it wants to see tested safely. Accountability comes with it. Under the Senior Managers and Certification Regime, a named person carries responsibility if AI causes consumer harm. The FCA has been clear that not understanding your own model is no defence.

So the direction is set. You can use AI in collections. You have to be able to explain what it did, why, and show it delivered a good outcome. That's a high bar for a black box.

You can't defend AI outcomes without a foundation you can audit

When we talk about AI in collections, its easy to miss one vital thing. An AI engagement layer or a clever conversational tool must be built on a solid foundation.

That something is one system for collections: the complete, auditable record of what happened on every account. Communications, arrangements, actions, decisions and the audit trail, held in one place. It lets you show, case by case, what happened on the account and when. Without it, an AI layer is making recommendations on data it doesn't fully own and can't account for.

That's what Flexys provides. Our collections software maintains a complete record of collections activity and integrates with the core banking systems that hold the balances and payments. AI can sit on top and do useful work. The history and the evidence of every action stay in one place beneath it.

That's the difference between AI that improves customer outcomes and reduces regulatory risk, and AI that quietly creates both harm and exposure. Two of the four things every collections operation is judged on.

The BSA framed this as a choice for the whole mutual sector. It's really a choice for anyone automating an arrears book. Point the technology at outcomes, and build it on a foundation that can prove you did.

Get in touch to see how outcomes-focused collections works in practice.