PrymeIntelligence

Banking and fintech·Answer

Can an AI agent agree a payment holiday or forbearance plan?

An AI agent can agree a standard payment holiday within a pre-approved matrix, but vulnerable customers and bespoke forbearance stay with a named person.

Yes, but only inside a pre-approved matrix. An AI agent can agree a standard, short arrangement when the customer's circumstances match the bank's written rules and there are no signs of vulnerability. Anything bespoke, longer, larger or involving a customer in serious difficulty stays with a named person at the bank, because the plan shapes the customer's finances for months.

Why is forbearance more than an administrative change?

A payment holiday looks simple: the customer pays nothing for a period. But it changes the loan. Interest may continue to build, so the balance grows and later payments rise. The arrangement may be reported to credit reference agencies. And for a customer in real difficulty, a short pause can delay the help they need.

So forbearance is a decision about a customer's finances, not a form to process. Done well, it gives a customer time to recover. Done badly, it hides a deeper problem and leaves them worse off. The term is defined in what forbearance is. In the UK, the FCA Consumer Duty expects firms to deliver good outcomes for retail customers, which bears directly on how forbearance is offered.

What are the conditions?

An agent can agree a plan only when all of these hold:

  • the request falls inside a pre-approved matrix of plan types, lengths and amounts;
  • the customer's account and circumstances meet the matrix's eligibility rules;
  • there are no vulnerability indicators, such as illness, bereavement, mental health concerns or signs of coercion;
  • the customer has been told plainly what the plan will cost them and how it affects their record;
  • it is not a repeat arrangement beyond the matrix's limit.

If any condition fails, the plan becomes a draft for a person. The matrix itself is a governed document; see the repayment plan limits matrix for one structure.

What may the agent do, and what stays with a person?

Step The agent Stays with a person
Taking the request Gathers the reason, income and outgoings Nothing routine
Checking eligibility Tests the request against the matrix Cases the matrix does not cover
Explaining the effect Shows interest, new payments and record impact Answering questions the script does not cover
Agreeing a standard plan Agrees it at depth 3, inside the matrix Nothing, if every condition holds
Bespoke plans Drafts the proposal Agreeing it
Vulnerable customers Stops and hands over The whole conversation
Reviewing at the end of the plan Contacts the customer and drafts next steps Any extension beyond the matrix

The agent's playbook for the intake is in how to handle loan payment holiday requests with an AI agent, and the later review is in how to review forbearance arrangements with an AI agent.

A worked example

Take a building society with a matrix that allows up to three months' reduced payments on a personal loan for customers with a temporary drop in income.

A customer writes that their hours have been cut for six weeks. The agent gathers their income and outgoings, checks the matrix, and confirms eligibility. It explains that interest will continue during the reduced period and that the remaining payments will rise by a stated amount. The customer agrees, and the agent sets up the arrangement at depth 3.

A second customer asks for the same thing, but mentions that they have stopped work because of a serious illness. The agent stops. It does not agree a plan, because a three-month pause may be the wrong help. It hands the conversation to a specialist with the details so far, and tells the customer who will contact them and when.

What controls make it safe?

  • The matrix is the bound. The agent cannot agree anything outside it, and the matrix is owned by a named person in collections or credit risk.
  • Vulnerability stops the agent. Indicators are defined, tested and acted on at once. The agent does not weigh them against eligibility.
  • Clear disclosure, recorded. The customer's understanding of the cost is captured in the record before the plan is agreed.
  • Reversal is limited. Ending a plan early or changing its terms is a person's decision, because a payment reversal or schedule change has its own consequences; see whether an AI agent can reverse a payment.
  • Outcome review. A person reviews a sample of agent-agreed plans and tracks how those customers fare.

The broader limits for agents in banks are set out in AI agents in banking: what they may and may not do.

Questions people ask

Can the agent refuse a payment holiday?

It can explain that a request falls outside the standard options and pass it to a person. A refusal of forbearance is a decision with real effects on the customer and should be made by someone who can consider alternatives.

Can the agent agree a plan on a mortgage?

The same pattern applies if the lender has a matrix for mortgages. Because the sums are larger and arrears can lead to possession, many lenders keep all mortgage forbearance with specialists.

Does the agent need to tell the customer it is an AI?

Customers should know when they are dealing with an AI system. Under the EU AI Act this is a transparency obligation, and it matters more when the customer is discussing financial difficulty.

Decide what your agents may do. Prove it. Enforce it.

Pryme Intelligence certifies each agent on your own cases, checks every action against five bounds at the moment it happens, and files every verdict.