Customer win-back calls
Former customers are not one uniform audience. One may have left because a project ended, another because the service failed, and another because their needs changed. A generic sales call ignores that history. It also creates risk when teams call stale records without checking consent, suppression status, or whether an unresolved complaint should go to support instead of sales.
A customer win-back phone agent can make a controlled first contact with people the business has already determined are eligible. It identifies the business and purpose, asks whether the person is willing to talk, confirms the reason for leaving without arguing, and presents only the next steps approved for that segment. The goal is a reliable call disposition—interested, not interested, wrong person, support needed, call later, or opt out—not a forced conversion.
A concrete call walkthrough
Suppose Jordan previously used a maintenance plan but did not renew. The customer system supplies the plan type, end date, and approved win-back path.
Agent: Hello, may I speak with Jordan? I am the AI phone assistant for Northline Services. I am calling because your maintenance plan ended earlier this year. Is now a good time for a brief question?
Caller: I canceled because two visits were hard to schedule.
Agent: Thank you for explaining that. I will record scheduling as the reason. I cannot resolve the earlier experience myself, but I can ask a service specialist to follow up, or I can end the call. Which would you prefer?
Jordan's response changes the workflow. If Jordan requests help, the agent records the issue and can try an attended warm transfer. The caller waits while the agent privately briefs the recipient, and the call connects only after acceptance. If nobody accepts, the agent returns to Jordan and explains the fallback. If Jordan says not to call again, the agent confirms the request and ends the promotional path rather than continuing with an offer.
For a former customer whose departure was not complaint-related, the prompt may state an approved offer or appointment option supplied by the business. The agent should not invent discounts, negotiate terms, or present an expired offer. When the contact belongs to an old prospect rather than a former customer, use lead reactivation so the script does not imply a relationship that never existed.
How to set it up on ThunderPhone
Define audience eligibility in the source system before a call is attempted. HubSpot or Salesforce can supply the customer record, prior relationship, approved segment, and suppression status. Google Sheets can support a small, manually reviewed list. A custom database can pass the minimum required context and receive the outcome through REST or webhooks.
The prompt should specify:
- the business identity, AI disclosure, and reason for calling;
- how to verify that the intended person answered;
- the approved opening for each customer segment;
- neutral questions about why the relationship ended;
- which offers, scheduling actions, or transfers are allowed;
- how to handle complaints, uncertainty, and requests to stop; and
- the exact dispositions written back after the call.
Use a verified imported production number, because demo U.S. numbers are inbound-only. Connect Cal.com or Google Calendar only if the approved next step is scheduling. Slack can alert the retention or support team when a customer raises a service problem. A Zapier MCP connection can expose selected actions from an existing workflow, while remote MCP servers are attached by URL and synchronized for tool discovery. For custom outcome handling, ThunderPhone webhooks support event subscriptions, per-endpoint secrets, retrying non-blocking delivery, and HMAC-SHA256 signatures.
Configure voicemail separately from a live conversation. ThunderPhone can let the prompt decide, hang up, or leave a configured message when voicemail is detected, and a campaign can retry voicemail outcomes according to its retry policy. A voicemail should not disclose sensitive account history or claim the customer expressed interest.
Before launch, simulate satisfied former customers, unresolved complaints, a wrong party, an expired offer, voicemail, a direct opt-out, a transfer failure, and incomplete CRM data. Reusable scenarios, graded call logs, regression suites, and A/B experiments can help test prompt changes. Simulations are billable real calls, and the interface shows the charge first.
What to measure
Keep contact quality and commercial outcomes separate:
- intended-person contact, voicemail, wrong-party, and no-answer rates;
- reasons customers give for leaving;
- requests for support or human follow-up;
- accepted transfers and booked next steps;
- customers who return after each disposition;
- opt-out requests and suppression failures; and
- missing, stale, or contradictory source records.
If the call is tied to a contract approaching its end, compare this workflow with renewal reminder calls. A win-back call addresses a relationship that has already lapsed; a renewal reminder should reach the customer before that point.
Honest limits
An AI phone agent cannot repair a damaged relationship by repeating a script. Complaints about safety, billing disputes, service failures, contract terms, or a sensitive personal situation need a person with authority to investigate and respond. The agent should not debate the customer's reason, conceal its role, or keep selling after a refusal.
Eligibility also remains the customer's responsibility. Consent, calling windows, applicable do-not-call requirements, recordings, and opt-out management must be configured for the business's obligations. Keep the initial call narrow: learn what happened, offer an approved path, and respect the answer.