Overdue payment calls: script, cadence and calling rules

Overdue payment calling is set up as a campaign: a list of contracts with the amount and due date, a reminder script, a permitted daytime window and retry rules. After that the voice AI agent calls on its own, speaks Uzbek and Russian, logs the promise to pay in the customer record and escalates only disputed conversations. Below: the script line by line, the retry cadence and the legal minimum.

Why manual payment calling does not cope

Payment reminders are a high-volume, utterly routine task. One operator makes a limited number of productive calls per shift, and the load is uneven: at month end and around contract deadlines it multiplies. The result is that not the whole base gets called, only the part someone managed to reach.

What matters is that a noticeable share of customers are not refusing to pay — they simply forgot the date. The absence of a timely reminder turns a solvent customer into an overdue account, after which the far more expensive collection procedure kicks in.

• Volume. Thousands of contracts against dozens of operators — the arithmetic does not work.
• Peaks. Month end and payment deadlines create a spike nobody is left to cover.
• Routine. The same script on repeat wears people down and drops quality by the end of a shift.
• Lost outcomes. With manual calling the result of a conversation often never reaches the system and is lost.

How to set up overdue payment calling in four steps

1. Load the data. The customer list, the amount and due date, the contract number and the language to speak. Data is pulled from the AI-CRM or uploaded as a list.
2. Describe the scenario. What to remind them of, what to ask (is the customer ready to pay, and when), what counts as success and what counts as grounds for escalation.
3. Set the calling rules. The permitted daytime window, the number of retries, the pause between them, the stop list and do-not-call.
4. Launch and review. The agent calls, speaks the customer’s language and understands the answer; the outcome, the transcript and the next step land in the record, and conversation quality is reviewed in the call analysis module.

You do not need to change your telephony: the agent connects over SIP on top of your existing PBX — the telephony layer is described on the telephony (VoIP) page. Onboarding takes from 2 hours for a small business and up to 30 days for enterprise environments with integrations and security requirements.

Which scenario each payment stage needs

The same mechanism works outside the financial sector: subscription and membership renewals, insurance renewals, tuition payments, recurring invoices in B2B services. Wherever there is a repeating payment, there is a reminder scenario.

What exactly the agent says on a reminder call

The reminder script is short — the conversation fits into a minute and a half. The order of the lines matters more than the wording: first who is calling, then the fact and the amount, then a question the customer answers in their own words.

1. Introduction and disclosure. The agent names the company and immediately states that an AI assistant is speaking. These are the first two lines, not small print at the end.
2. Fact and amount. The contract number, the amount due and the date: “under your contract, 1,200,000 soums are due by the 5th”. No judgement of the customer, no pressure.
3. The key question. “Will it be convenient to pay by that date?” — an open question instead of “press 1”. The answer is recognised as free speech in Uzbek or Russian.
4. Logging the promise. If the customer names their own date, the agent repeats it aloud and writes it into the record as a promise to pay — the next contact is counted from that date.
5. How to pay. On request the agent names the channel — the bank app, a payment service, the cash desk — and sends the link or the details as a Telegram message.
6. Closing. Thanks, a repeat of the date, and one mandatory branch: a “do not call me” request moves the contact to the do-not-call list within that same call.

Tone is set separately for each stage: before the due date it is a neutral reminder; in early arrears, polite persistence with no threats and no listing of consequences. Wording about collection, courts and penalties is kept out of the machine script — that is for a lawyer and an operator.

How many times to call if the customer does not pick up

Cadence is the set of retry rules. It is configured before launch and protects against the main mistake of mass calling: over-dialling that turns a reminder into an irritant.

The second mandatory setting is payment-status sync. The calling list is rebuilt before every run, so a customer who paid yesterday drops out of the campaign. Without it the agent calls people who have already paid — the most common source of complaints about automated calling.

How a voice agent differs from an auto-dialler announcement

Why an Uzbek voice is critical in a conversation about money

A conversation about money is the most sensitive to tone. If the voice sounds mechanical or speaks a language that is not the customer’s own, they hang up in the first seconds and the reminder never lands. Zukko.AI has its own Uzbek voice, which the global models do not provide — the agent holds a natural conversation in Uzbek and Russian and switches between them inside a single call. More in a robot that calls and speaks Uzbek.

What the law requires for payment calling

A phone number, a full name and payment history are personal data; their processing is governed by Law ZRU-547 as amended by ZRU-1125 on localisation (gazeta.uz, March 2026). The practical minimum for payment calling: a lawful basis for the call and for processing the data, the permitted daytime window, disclosure that an AI is speaking, a correct tone without pressure, and respect for a refusal. This is an overview, not legal advice — the final assessment for your case comes from your compliance team.

Which metrics to watch after launch

We do not publish “normal” industry figures — the baseline is taken in the first week of your own campaign and then compared against itself. Five indicators are worth watching; they are available in the reports and in the call analysis module.

• Contact rate. The share of contacts where a conversation actually happened. If it falls, the problem is number quality or calling time, not the script.
• Share of promises to pay. How many conversations ended with a specific date rather than a vague “I will see”.
• Promises kept. How many of those dates were actually closed by a payment — reconciled against billing; this is the money the campaign brings.
• Escalations. The share of calls handed to a human. A sharp rise means the script does not cover the objections customers actually raise.
• Refusals and complaints. Growth in do-not-call is a signal that the tone is too insistent or the cadence too dense.

What the business gets

• The reminder reaches the whole base — coverage stops depending on the number of operators.
• Fewer arrears through forgetfulness — the customer is warned before the date falls due.
• A conversation, not a recording — the agent understands the answer and drives to a specific next step.
• Nothing is lost — the outcome of every call is recorded in the customer record.
• Complexity goes to a human — disputed and conflict cases reach an operator with the full context.

In short

Payment calling is work where timing and coverage matter, not improvisation: the ideal task for a voice AI agent. It absorbs the mass routine and the peaks and leaves operators what genuinely needs a human — a dispute, a conflict, a non-standard arrangement. Launch is possible on your existing telephony: from 2 hours for a small business and up to 30 days for an enterprise. A general overview of what the voice agent can do is in the guide to automated calling.

Want to hear how the agent delivers a payment reminder on your own data and scenario? Get a demo call or explore the voice AI module.