Spot at-risk customers before they lapse
Almost no customer vanishes without warning. They drift past their usual reorder window, stop opening emails, or go quiet after a complaint.
Scoring those signals weekly costs nothing and catches people while a save is still cheap. AI can build the score straight from your order export.
Fastest path: one prompt, end to end
🤖 AI prompt — paste into ChatGPT / Claude
You are a retention analyst. Use MY data only.
My product + typical reorder cycle: [e.g. supplements ~40 days]
Data I can export: [orders CSV / email engagement / support tickets, list what you have]
Do this:
1. Define my at-risk signals with thresholds tied to MY cycle: days past expected reorder (e.g. >1.5x cycle), email engagement drop (no opens in N sends), unresolved or negative support ticket, declining order value across last 2 orders.
2. Build a simple 0-100 at-risk score from those signals with weights, and give me the spreadsheet formula or steps so I can run it on my export weekly.
3. Segment the output into three buckets: watch, act now, probably gone, with a recommended intensity for each.
4. Tell me which single signal is most predictive for a product like mine and why.
If a data source is missing, ask; do not guess my numbers.
Output: signals + thresholds + score formula + three buckets.
Or do it in 4 steps
- Anchor everything to your reorder cycle. "Inactive for 60 days" means nothing by itself: it's churn for coffee and normal for furniture. The core signal is days past expected reorder, so compute your median gap between orders first and measure risk against that.
- Layer the softer signals. Email engagement fading (no opens across recent sends), an unresolved or unhappy support ticket, and declining order value each add risk. None alone is proof; two or three together is a customer walking toward the door.
- Score simply, act by bucket. A weighted 0-100 score sorted into three buckets: watch (light touch, keep serving), act now (this is where saves happen), probably gone (suppress from regular sends, occasional winback only). Fancy models can wait; a spreadsheet beats doing nothing by a mile.
- Route each bucket to its play. The at-risk score is strategy; execution lives in its own notes: the winback sequence is in winning-back-customers, and the email segmentation mechanics are in the segmentation note under marketing. Wire score to play and let the triggers run (see retention-trigger-map).
Worked example (labeled): a supplement brand with a ~40-day cycle scores its customer export weekly. A customer at day 70 (1.75x cycle), zero opens in the last 5 sends, and a shipping complaint that closed without a reply scores deep in "act now."
A personal check-in with a small make-good lands before the customer has mentally switched brands; the same message at day 120 would have been a coupon to a stranger. Early beats generous.
Run the score weekly and tune thresholds quarterly; the cheapest save is the one that happens before the customer decides they've left.