AI in Collection Agencies: Contact, Vulnerability and Compliance

Collections is one of the most heavily regulated consumer-facing activities, and for good reason: it involves contacting people about money they owe, frequently when they are under financial pressure. The rules on contact, conduct, and vulnerability are not constraints around the operation; they define it. This guide covers where AI helps within that, drawing on FISTA Solutions' AI agents work in financial services. It complements ai in credit bureaus and what is a guardrail policy. This article is general guidance, not legal advice.

Why are contact rules not optimisation parameters?

Because they are limits set to prevent harassment. Frequency caps, permitted hours, and channel restrictions exist because unrestricted contact causes harm to people already under pressure.

A system that finds the best time and channel within those limits is legitimate. One that treats the limits as boundaries to approach as closely as possible is operating against their purpose, and it will generate complaints, regulatory attention, and reputational damage that outlast any collection gain.

Element	Automatable	Trained staff required
Contact timing and channel within limits	Yes	—
Routine account queries	Yes	Escalation
Vulnerability signal detection	Yes, conservatively	Immediate handling
Affordability assessment	No	Yes
Dispute investigation	Assembly only	Determination
Arrangement setting	No	Yes

How should vulnerability be handled?

Detected conservatively and routed immediately. Indicators include disclosed illness, bereavement, mental health difficulties, caring responsibilities, and hardship language — and they appear in conversation rather than in structured fields.

The threshold should be set to over-route. An unnecessary escalation costs a few minutes of a trained person's time; a missed one means continued pressure on someone who cannot cope with it, which causes real harm and, increasingly, regulatory consequence.

Why must affordability be human?

Because assessing what someone can sustainably pay requires understanding their circumstances, and an arrangement set above what they can afford fails within months and leaves them worse off — with fees, further arrears, and a failed arrangement on their record.

A sustainable arrangement is worth more to the creditor than an optimistic one, which aligns the commercial and the fair outcome. Reaching it requires a conversation rather than a calculation.

What happens when a debt is disputed?

Collection activity pauses while the dispute is investigated. That is a regulatory requirement in most regimes and a basic fairness one.

Automated collection continuing through a dispute is a serious compliance failure, and it is a common one in systems where the dispute flag and the collection workflow were built separately. The pause should be structural rather than procedural.

What role does AI actually play?

Timing and channel selection within limits, so contact reaches people when they can engage rather than when it irritates. Routine query handling, so people can check a balance or confirm an arrangement without waiting. Vulnerability detection. Dispute evidence assembly. Quality monitoring of recorded interactions against conduct standards.

That last is underused and valuable: reviewing a sample of every agent's calls against conduct standards, rather than a handful, identifies problems before complaints do.

What about the recordings?

They are evidence of conduct and should be treated as such, with retention matching regulatory requirements and access controlled. Automated review against conduct standards is a compliance capability rather than a performance management one, and framing it that way determines whether agents accept it.

Who should own it?

Compliance, with operations delivering. In this sector the compliance function should hold a veto over anything touching contact strategy, vulnerability, or arrangements, because commercial pressure on those decisions is constant and the consequences of getting them wrong fall on consumers.

How is it evaluated?

Sustainable arrangements that complete, complaint volume and cause, vulnerability correctly identified and handled, dispute handling timeliness, and conduct standards adherence from call review. Contacts made and in-period collections reward exactly the behaviour the rules exist to prevent.

What goes wrong?

Contact optimisation that treats limits as targets. Vulnerability thresholds set to minimise escalations. Affordability calculated rather than discussed. Collection continuing through disputes. And performance measured on amounts collected this month.

What does it cost to run?

Low per account. The cost that matters is trained staff for vulnerability and affordability, which should increase rather than decrease as automation handles routine contact — that reallocation is the point.

What should you do first?

Review a sample of calls against your vulnerability policy and count how many indicators were missed. That number is usually uncomfortable and it is the clearest case for systematic detection, because the misses are not deliberate.

How does self-service change the picture?

Substantially, and mostly for the better. Many people in arrears find contact with an agent stressful and avoid it, which prolongs the debt and worsens the outcome for everyone. A self-service route to check a balance, propose a payment, or request a pause removes that barrier.

The design constraint is that self-service must not become a way of avoiding the vulnerability conversation. Someone setting up an arrangement they cannot afford through a form has not been helped, so affordability checks and routes to a person need to be present in the self-service path rather than only in the agent one.

How FISTA Solutions helps

FISTA Solutions builds collections systems with contact timing optimised strictly within regulatory limits, conservative vulnerability detection routed immediately to trained staff, structural collection pauses on dispute, automated conduct review across all calls rather than samples, and affordability conversations kept human, through AI agents, AI enablement, and forward deployed engineers. The record behind the approach is 150+ projects for 50+ companies with 99.9% uptime.

To collect sustainably and treat people fairly, message FISTA on WhatsApp, or read what is a guardrail policy.