A win-back agent
that finds the reason to come back, per customer
A customer who used to buy regularly and then simply stopped is usually easier and cheaper to win back than a brand-new customer is to acquire, but most businesses have no system watching for who went quiet or why. We build an agent that spots the lapse, works out a likely reason from the customer's own order history, and sends a personalized offer instead of a generic 'we miss you' blast.
The role today
A customer who bought regularly and then quietly stopped is one of the more valuable segments most businesses never systematically work, mostly because nobody is specifically watching for the lapse. By the time someone notices a customer has not ordered in a while, if anyone notices at all, months have usually passed, and whatever generic “we miss you, here’s 10% off” campaign goes out treats a customer who left over a bad experience the same as one who simply forgot to reorder.
The data to do better already sits in most order histories and CRMs - what someone bought, how often, when it stopped, whether a support ticket came in around that time - but turning that into a per-customer insight at any real volume is not something a team does by hand.
What the agent takes over
The agent flags a lapse against that specific customer’s own typical purchase interval, so a customer who normally orders monthly gets flagged after a shorter gap than one who normally orders twice a year, rather than everyone being measured against one fixed day count. It then looks at that customer’s order history, support tickets and the timing of the lapse to work out a likely reason - a product that is no longer in stock, a price change around the time they left, a support issue that never got fully resolved, or simply a gap with no obvious cause.
From that, it drafts a personalized win-back message: a specific discount where the data suggests price was the issue, a reminder of a product they previously liked where it looks like simple forgetting, or a direct question about what changed where the pattern suggests dissatisfaction. The channel - email, WhatsApp, SMS - is chosen based on where that customer has actually responded before, not a single default channel for everyone.
Every attempt is tracked to a clear outcome: reactivated, no response, or opted out, feeding back into both the CRM and your analytics so your team can see win-back performance as a real number, and spot a pattern across lapsed customers - a product, a price point, a support issue - worth a closer look.
What stays with humans
Deciding what to fix if a pattern points to a real product or service problem, approving discount ranges, and any customer who replies with a specific complaint stay with your team. The agent finds the lapse and drafts the right message; a person decides what the business does with what it learns.
Guards
Lapse detection is tuned per customer against their own purchase cycle, which avoids both false alarms on naturally infrequent buyers and missed windows on frequent ones. Discount offers stay within ranges you approve, every outcome is logged, and an opt-out stops all future win-back attempts for that customer permanently. A kill switch pauses campaigns in one message if the approach needs review.
Price and timeline
| Option | Price | What it covers | Timeline |
|---|---|---|---|
| Agency runs it | from $1,800 + support plan | Lapse detection, personalized campaign, result tracking | 2 to 4 weeks |
| Full control, handover-ready | from $3,050 | Source, prompts and data pipeline handed to your team | 2 to 4 weeks |
Related
Full package breakdown on the AI agents service page, and win-back of inactive customers automation for the standalone process view. Within this group, pair it with the upsell and renewal agent for active customers, and the follow-up and nurture agent for leads rather than past buyers. See also RFM scoring model for the segmentation layer underneath. Real builds: margin-aware customer and order data at scale is in the digital goods marketplace automation case study, and a full analytics warehouse feeding this kind of segmentation is in the analytics hub case study.
Want to find out which lapsed customers are actually worth winning back, and why they left? Get in touch and we will look at your order data on the first call.
FAQ
How much does a win-back agent cost?
From $1,800 for lapse detection and a personalized campaign on one customer base, live in 2 to 4 weeks. A version tied into a full analytics warehouse for deeper segmentation usually starts at $3,200.
How does it decide someone has actually lapsed?
Against that customer's own typical purchase interval, not a fixed day count applied to everyone. A customer who normally buys monthly gets flagged differently from one who normally buys twice a year.
Does it always send a discount?
No, a discount is one option among several - a new product they have not tried, a reminder of what they liked, a direct question about what changed - chosen from what the data suggests is more likely to work for that customer.
What if a customer does not want to be contacted again?
An opt-out stops all future win-back attempts for that customer and is logged, respected the same way an unsubscribe is respected anywhere else.
Can it tell me why customers are actually leaving?
It flags patterns across lapsed customers - a product, a price change, a support issue - that show up often enough to be worth a real look, though the underlying fix is a decision for your team, not the agent.