PrymeIntelligence

Banking and fintech·Template

Repayment plan limits matrix

A repayment plan limits matrix setting what an AI agent may agree by product, balance and customer situation, with field guidance and a filled sample.

Every payment plan an AI agent agrees should sit inside a repayment plan limits matrix. It sets, by product, balance and customer situation, the longest term, the lowest monthly payment, whether interest can be frozen, how many plans a year, and the evidence required. Anything outside it goes to a person the matrix names.

What problem does the matrix solve?

Payment plans are where collections work meets customer outcomes. A plan that is too tight sets the customer up to fail. A plan that is too generous, offered inconsistently, treats similar customers differently and may cost the lender more than it should. People handling arrears have always worked to limits; the matrix writes them down so an agent can work to the same ones.

Writing them down changes the question you ask about the agent. It is no longer "does the agent have good judgement about payment plans?" but "does the agent stay inside the matrix every time, and refer everything else?". That second question can be certified on your own cases. The process the matrix supports is described in how to agree repayment plans within set limits with an AI agent.

When should you use it?

Use it to configure the agent, to brief collections staff who receive referrals, and to build certification cases: every cell of the matrix should be tested at and just beyond its edge. Review it whenever arrears policy, product terms or treatment of customers in financial difficulty changes.

The matrix covers plans for amounts already owed. Requests to move a payment date before any arrears exist follow a lighter route, covered in how to handle requests to change a repayment due date with an AI agent. The idea of a limits matrix is shared with support work, for example the refund and credit limits matrix for support agents, but the stakes here are higher because a plan shapes the customer's finances for months.

How do you fill in each field?

  • Product and balance band. Separate rows for each, because what is reasonable for an overdraft is not reasonable for a loan.
  • Situation. Standard, financial difficulty, vulnerability. Vulnerability usually means "person only".
  • Maximum term. The longest plan the agent may agree.
  • Minimum monthly payment. The floor, as an amount or as the result of an income and expenditure check.
  • Interest and fees. Whether the agent may freeze them during the plan.
  • Evidence. What must be gathered first: none, a short income and expenditure statement, or a full one.
  • Frequency. How many plans a customer may agree with the agent in a year.
  • Agent depth and referral. Depth 3 within the cell; held for whom outside it.

The repayment plan limits matrix

Product Balance band Situation Max term Min monthly payment Interest and fees Evidence Plans per year Agent depth Refer to

A filled sample: a consumer lender

Product Balance band Situation Max term Min monthly Interest and fees Evidence Plans a year Agent depth Refer to
Credit card Up to 2,000 Standard 6 months Balance divided by term Fees frozen; interest continues None 1 3 Collections officer
Credit card Up to 2,000 Difficulty 12 months From short income and expenditure Both frozen Short statement 2 2 Collections officer
Credit card Over 2,000 Any 12 months From full income and expenditure Officer decides Full statement 1 2 Senior officer
Personal loan Any Standard Add up to 3 missed payments to end of term Contractual payment Fees frozen None 1 3 Collections officer
Any Any Vulnerability Not delegated Person only Specialist team

Notice where depth drops. A standard customer with a small card balance gets a plan from the agent in one conversation. A customer in difficulty gets a more generous plan, but a person agrees it, because the terms rest on their income and spending and the conversation needs care.

What should the agent do at the edge of a cell?

Never round in the customer's favour or the lender's. If the customer asks for seven months on a six-month cell, the agent offers six, explains that a colleague can consider longer, and refers if the customer wants that. The verdict records the cell that decided it. If a broken plan is followed by a request for a new one, the frequency column decides whether the agent can help.

Limits on spending or merchant transactions are a different kind of limit and should not be mixed into this matrix; see whether an AI agent can change a customer's card spending limits and whether an AI agent can set a merchant's transaction limits. The guide to AI agents in banking covers the wider boundaries.

Questions people ask

Should customers see the matrix?

Not as a table, but the agent should be clear about what it can agree and that a person can consider more. Hidden limits that customers bump into without explanation produce complaints.

Who approves changes to the matrix?

The head of collections proposes; credit risk and conduct risk agree. Any widening of a cell at depth 3 should be checked against certification cases before it goes live.

What if a customer refuses every plan the agent can offer?

The agent refers the case to the named role with a summary of what was offered and declined, and tells the customer when to expect contact.

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.