Customer Re-Engagement: Use Signals Without Guessing Interest

Silence does not mean a customer wants another offer. Even a high AI score cannot tell the team why someone stopped or whether contact is welcome now.

Start with one observable signal and use the ladder below. The result should be review, contact, help, or stop—not a hidden assumption of renewed interest.

Move from a signal to a safe action

Signal type	Example	Safe use	Human or stop boundary
Explicit	Customer asks to revisit something later	Review or act at the agreed time	Recheck current permission
Transactional	New purchase, return, or service case	Update what the customer is eligible for	Service may stop selling
Behavioral	Approved visit or product interaction	Prioritize a record for review	Does not prove intent
Predicted	Model estimates likelihood	Support a review decision	Show the inputs, uncertainty, and errors
Weak or prohibited	Sensitive inference or generic company news	Do not use	Exclude from the decision

Copy this signal record and fill every line before contact:

Inactive definition or not-applicable reason:
Signal, source, and date:
Permitted purpose and contact permission:
Service state and reason not to contact:
Supporting evidence:
Uncertainty and exception:
Decision and reason:
Contact owner:
Help and close owner:
Customer choice and message purpose:
Stop and close state:

Klaviyo documents a vendor win-back flow using a defined inactivity period. NIST's AI RMF supports clear human-AI roles, monitoring, and override when AI helps evaluate signals. Neither source proves a cadence, prediction quality, or outcome. Klaviyo, NIST.

Apply the ladder to one record

1. Write what “inactive” means for this business; do not copy a number of days.
2. Record the signal, source, date, permitted purpose, and any reason not to contact.
3. Prefer an explicit customer request or current transaction over a model estimate.
4. Let AI prioritize review only when the supporting information and uncertainty are visible.
5. Give the customer one useful choice, plus clear help, close, and opt-out routes.
6. Record false eligibility, corrections, complaints, and whether the process stopped when it should.

A person remains responsible for contact permission, sensitive signals, service conflicts, message approval, and the final decision. AI can assist the review; it does not establish interest.

Fictional worked example: act on an agreed date

A fictional member asks the team to revisit a workflow guide after the business hires an operations manager. The agreed month arrives, and the team fills the signal record:

Inactive definition or not-applicable reason: Not applicable—an explicit dated customer request triggers this review, so no inactivity score is needed
Signal, source, and date: Customer's fictional membership note dated 15 May 2026 asks for a review in August; source checked 15 August 2026
Permitted purpose and contact permission: Fulfil the requested reminder through the permitted service-email route; do not add a sales campaign
Service state and reason not to contact: No open dispute or opt-out in the current account check; stop if either appears before contact
Supporting evidence: Original dated request, current account state, identity match, and contact preference
Uncertainty and exception: The operations-manager hire and renewed interest are unverified; ask rather than infer. Model score is not applicable because the explicit reminder is sufficient
Decision and reason: REVIEW, then one contact question, because the customer named the topic and review month
Contact owner: Account owner approves and sends the question
Help and close owner: Service owner accepts HELP; account owner records an explicit close or approved no-response close
Customer choice and message purpose: Ask whether the guide is useful now, the reminder should close, or help is needed; no claim that interest has returned
Stop and close state: Stop on opt-out, service conflict, invalid identity, explicit close, or HELP handoff; no reply moves to the approved closed-no-response state

AI surfaces the request and source; it does not infer that the member is interested again. A reply determines the next action, and no reactivation result is assumed.

Risks and controls

Risk	Why it matters	Plain response
AI score is treated as interest	Guesswork becomes contact	Show the signal and require human review
Sensitive behavior is inferred	Creates privacy and fairness risk	Use an approved signal list and minimum data
Recent service problem is missed	Contact arrives at the wrong time	Put service before selling
Every return is credited to AI	Inflates causal claims	Set the comparison and observation window first

Measure signal validity

Use valid-signal rate = reviewed records with a reviewer-confirmed eligible signal / records reviewed. Also track false eligibility, corrections, opt-outs, complaints, help requests, and observed returns. Do not claim the process caused reactivation without a valid comparison.

The lifecycle or data owner should review monthly. The business must set its starting baseline, intervention threshold, approved signals, and pause owner before AI-assisted use.

Frequently asked questions

Is website activity enough to contact someone?

Not by itself. The purpose, permission, identity confidence, and channel rules still determine whether contact is appropriate.

Can AI infer churn?

It may estimate risk under an approved method. The estimate remains uncertain and does not establish cause or renewed interest.

Is a discount the best re-engagement action?

Not necessarily. A useful resource, preference check, or service route may fit better. Any offer needs approved terms.

When should the process stop?

Stop on reply, purchase, help request, opt-out, service conflict, invalid contact, explicit close, or the approved no-response state.

What to do next

Use the ecommerce customer re-engagement template when the signal is a purchase-cycle lapse.

Evidence and limitations

No predictive accuracy, reactivation or revenue lift, lawful basis, integration, or Easy AI capability is claimed.