How do FCA-regulated lenders use AI in collections and arrears?
AI collections in an FCA-regulated lender means pre-arrears intelligence (predicting who will roll before they do), treatment paths that route each borrower to the CONC 7 outcome most likely to work for them, and vulnerable-customer detection that is stronger than any human agent panel. The lenders shipping this see 20-30% lower roll-to-3+ rates and materially better regulator posture.
AI collections in an FCA-regulated lender means pre-arrears intelligence (predicting who will roll before they do), treatment paths that route each borrower to the CONC 7 outcome most likely to work for them, and vulnerable-customer detection that is stronger than any human agent panel. The lenders shipping this see 20-30% lower roll-to-3+ rates and materially better regulator posture.
Pre-arrears is where the P&L is won
Waiting for a missed payment before acting is the old model. Modern lenders identify a rising-risk borrower two to six weeks before their first miss — from cashflow signals, portal behaviour, engagement patterns, external triggers — and act inside the borrower's frame of reference, not as a collections call after the harm.
This is not about pushing harder. It is about intervening earlier, in a way the borrower experiences as help.
The four AI systems inside modern collections
- Pre-arrears scoring: propensity-to-miss, refreshed daily.
- Treatment routing: which CONC 7 path (forbearance, restructure, hold, plan) has highest expected outcome per borrower.
- Vulnerable-customer detection: signals from language, behaviour, product-use, external events.
- In-portal self-serve for hardship, promise-to-pay, restructure — with the Consumer-Duty overlay live.
Why this is a Consumer Duty accelerator, not a risk
Every intervention is logged, scored for outcome quality and compared against a matched control. That evidence base is what a supervisor actually wants to see. Firms operating on ad-hoc dialler scripts cannot produce it. Firms operating on the architecture above can produce it on demand.
Related questions
Does this replace agents?
No. It concentrates agent time on the borrowers who need human treatment and hides the rest in-portal self-serve.
How is CONC 7 encoded?
As machine-readable rules in the policy layer, with every model output mapped to the rule that allowed or blocked it.
Vulnerable-customer signals — what specifically?
Language cues, sudden portal-behaviour changes, external event proxies (bereavement, employment loss), product-use anomalies. Every signal is documented.
Does the regulator like this?
The lenders that show up to supervisory conversations with this architecture have materially shorter conversations than the lenders that do not.
Timeline?
Ten to fourteen weeks for pre-arrears + treatment routing after the £15k Diagnostic.
The AI-first lenders are pulling away on unit economics, not on rate. Every quarter you defer the architecture is a quarter of compounding disadvantage.
The £15k AI Diagnostic maps your lender stack, prioritises the systems that pay back fastest and produces a costed sequenced build plan.