A promise to pay is the customer's commitment, recorded accurately, not a deal the agent negotiates. The agent records promises inside your policy on date and amount, pauses reminders until the date, checks whether the money arrived, and follows up gently if it did not. Promises outside policy, and customers who keep missing them, go to a person.
What is a promise to pay, and who owns the process?
In collections, a promise to pay is a customer's commitment to pay a stated amount by a stated date, usually towards arrears. It is lighter than a repayment plan: one payment, a short horizon, no change to the loan's terms. Handled well, it gives a customer breathing room and the lender a clear next step. Handled badly, it becomes a cycle of promises made under pressure and broken, each followed by more contact.
The head of collections owns the promise policy: how far ahead a promise may be dated, the smallest amount that counts, how many broken promises trigger a review. A named collections team leader takes cases outside policy. The financial support team takes customers showing difficulty. For the wider frame, see our guide to AI agents in banking.
What does the agent do at each step, and at what depth?
- Agent, depth 1: reads the account. Arrears, recent payments, open promises and their history, flags. Evidence: facts used.
- Agent, depth 1: screens for difficulty. Previous broken promises, signs of stress in the conversation, other arrears. Signals send the case to the support team before a promise is discussed. Evidence: screening result.
- Agent, depth 1: listens for the customer's offer. The customer names the amount and date. The agent does not push for more or sooner. Evidence: the customer's words.
- Platform: checks the promise against policy. Date within the allowed window; amount at or above the minimum; open promise count. Evidence: verdict citing the policy version.
- Agent, depth 3: records the promise and confirms it in writing, in the customer's own terms. Evidence: promise record and confirmation.
- Agent, depth 3: pauses reminders until the date. Collections contact stops for the promise period. Evidence: suppressed contacts logged.
- Agent, depth 3: sends one reminder before the date, within your contact rules. Evidence: message sent.
- Agent, depth 1: checks for payment on the date. Kept: confirm and thank. Broken: a single, neutral follow-up asking what happened and offering help. Evidence: outcome recorded.
- Person, team leader or support specialist: takes promises outside policy and repeated breaks. Evidence: decision and reasons.
| Step | The agent does | Depth | Stays with a person |
|---|---|---|---|
| Read | Gathers arrears and promise history | 1 | Nothing |
| Screen | Checks for difficulty | 1 | Support team for any signal |
| Listen | Records the customer's offer | 1 | Nothing |
| Check | Nothing; the platform tests policy | none | Nothing |
| Record and confirm | Logs the promise, confirms in writing | 3 | Promises outside policy |
| Pause and remind | Stops chasing, sends one reminder | 3 | Nothing |
| Check payment | Confirms kept or broken | 1 | Nothing |
| Repeated breaks | Prepares the case | 2 | Team leader or support decides |
Which controls matter, and what evidence do they leave?
The customer sets the terms. The agent records what the customer offers. It does not counter-offer to raise the amount. Evidence: the customer's words alongside the record.
No promises in return. The agent must not say a default will be removed, fees waived, or a referral cancelled if the customer pays. Those are decisions it cannot make, and the risk is covered in the promises your support agent must never make. Evidence: transcript sampling.
Contact paused means paused. Every channel, every sender, including people. Evidence: no contact logged in the promise window beyond the agreed reminder.
The agent does not take the payment. It points the customer to your payment channels. Taking card details in conversation brings it into card data scope; paying on the customer's behalf is a different question, answered in whether an AI agent can pay a customer's bills. Evidence: no payment instructions created by the agent.
Broken promises do not escalate pressure. The follow-up asks and offers help; it does not threaten. Evidence: approved follow-up wording.
What should you measure?
- Promises kept, as a count against promises made, for agent-recorded and person-recorded promises.
- Customers with two or more broken promises in a period, and how many reached the support team.
- Contacts during promise windows. Target only the agreed reminder.
- Average days to promised date, to check the agent is not nudging dates closer.
- Complaints about collections contact.
- Promises changed before the date, which can show customers committing to amounts they cannot meet.
Where does it usually go wrong?
Promises are extracted, not offered. The agent asks "can you pay the full arrears by Friday?" and the customer says yes to end the conversation. Ask what they can pay, and when.
Broken promises restart the cycle. Another promise is recorded the same day, without asking why the last one failed.
Difficulty is missed. A customer juggling several debts, perhaps including deferred-payment purchases, makes promises they cannot keep. Agents approving such purchases face the same question from the other side; see whether an AI agent can approve a buy now pay later purchase.
Reminders ignore the promise. One system records the promise, another keeps sending arrears letters.
Promises are reported as plans. A one-off promise is recorded as an arrangement, changing how the account is treated and reported.
Questions people ask
What if the customer can only pay a small amount?
If it is below your minimum, the agent passes the case to a person, who may accept it or discuss a plan. A small payment the customer can make beats a larger one they cannot.
Can the customer change a promise?
Yes, before the date. The agent records the change and checks the new terms against policy. Frequent changes are worth a conversation with a person.
Should the agent record promises from third parties?
Only from someone authorised on the account, such as a registered attorney. Otherwise it can take a message but not record a promise.