Win back
Memory-personalized win-back
She didn't leave because the product failed her. She left over cost, said so once in a chat, and every win-back blast since has ignored it.
What it does
Most win-back programs send the same discount to everyone who lapsed, on a campaign calendar, and most of it gets deleted. But the reason each customer left is usually sitting in their history: a price objection mentioned in passing, a goal that stalled, a feature that never clicked. The agent builds each win-back from that customer's own memory — her progress, her stated blocker — and reaches out when the moment is right rather than when the calendar says so. Someone who left over cost gets a pause or a lower tier, not a one-size-fits-all discount plea. And because the message lands in a dialog instead of a one-way blast, the customer can answer, object, and be converted in the same conversation.
How it works
- 1Trigger. a lapse condition fires — subscription cancelled and a cool-off period passed, repeat-purchase window exceeded, or a reactivation-propensity signal — and the agent picks its own timing within the window instead of batching everyone on the same day.
- 2Decision. the agent reads the customer's memory for the exit reason and what they valued: progress made, blockers stated in past conversations, price sensitivity, channel preference. It selects the angle that addresses the actual reason they left, and a judge gates the drafted message for relevance, tone, and policy before anything sends.
- 3Action. a personal, specific message goes out on the customer's preferred channel (email, WhatsApp, SMS, or push), referencing their own history and offering the remedy that fits — pause, downgrade, a restart from where they stopped — and opens a two-way conversation rather than linking to a static landing page.
- 4Follow-through. replies are handled in-dialog through to a secure checkout hand-off or a clear no — the agent never takes payment in the chat itself; the outcome is written back to memory; if the customer comes back on their own first, the pending outreach is cancelled. Reactivation and revenue are measured per cohort against a holdout, so the program proves its own ROI.
Configuration
How the agent is wired for this use case.
- Subscription/commerce platform · read lifecycle events, set up the pause, downgrade, or restart on confirmation
- CRM · read exit reason and preferences, write the outcome back to memory
- Persistent memory · read progress, stated blockers, and price sensitivity to select the angle
- Messaging channel · send the personalized two-way message on the customer's preferred channel
- Secure checkout · hand off the payment via a secure link; the agent never takes payment in the chat
What you need
The inputs this use case runs on. Your channels stay yours; the agent supplies the judgment.
Signals
subscription and purchase lifecycle events (cancelled, lapsed, repurchase-window passed), engagement-decay or reactivation-propensity scores
Data
persistent customer memory (progress, stated blockers, exit reasons, preferences), purchase history, channel consent and contact preferences
Guardrails
judge gating on every unprompted message; suppression of recently contacted or do-not-contact customers; frequency caps; stale-outreach cancellation on organic return; per-cohort holdout for honest measurement
Metrics it moves
- reactivation-rateup, because the message addresses the real reason each customer left instead of a generic incentive
- recovered-revenueup, tracked per cohort from campaign to order against a holdout
- opt-out-ratedown, since relevant, well-timed messages get answered instead of reported
Related use cases
See it on your own customer journey
Bring one drop-off, one churn cliff, or one silent segment. We will show you what a proactive agent with memory and judgment does with it.
Book a demo