"By day 90 they'd already bought their paint somewhere else."
A 12-location home-improvement retailer ran win-back emails on a 90-day inactivity timer, converting at 0.5%. Twelve weeks after switching the trigger to churn-risk scores and splitting the list into CLV tiers, the flow converted at 1.3% and email had grown from 9% to 16% of online revenue. Cole Bergeron built the flows; Dana Okafor built the tiers.
| Metric | Before | After (12 weeks) |
|---|---|---|
| Win-back placed-order rate | 0.5% | 1.3% |
| Email share of online revenue | 9% | 16% |
| Top-CLV-tier RPR | untracked | $3.10 vs $0.95 list average |
| Win-back trigger | 90-day timer | churn-risk score threshold |
Why was the 90-day timer failing?
Cole: Because 90 days means different things to different customers. A contractor who buys monthly is long gone by day 90; a homeowner who renovates twice a year is not lapsed at all. Klaviyo's churn-risk model, which retrains weekly, catches the contractor at the moment their pattern breaks, weeks before any fixed timer would. Same email, sent when leaving is still a decision rather than a habit: 0.5% became 1.3%.
What did the CLV tiers change in practice?
Dana: Everything about cadence and offers. The top tier was producing $3.10 per recipient against a $0.95 list average, and the old program mailed them the same 10%-off blast as everyone else, discounting people who buy at full price. Tiers let the win-back offer scale with what the relationship is worth: service-led mail for the top tier, incentives reserved for profiles the model says are genuinely at risk.
How did 12 physical locations complicate it?
Cole: In-store purchases flowed into the ESP days late at first, which made active buyers look lapsed and would have poisoned the churn model with false positives. Fixing the POS sync became a week-three detour we had not scoped. Worth it: once store transactions landed daily, the model had the full purchase picture and location-level offers became possible.
What is the honest caveat here?
Dana: Churn-risk triggers need the predictive layer unlocked, which for Klaviyo means 500+ customers with orders, 180 days of history, and recent activity. A retailer below those thresholds should run recency-based proxies first, which is exactly how theaudience-analysis service sequences it. And the revenue-share gain took the full twelve weeks; nothing about this compounds in a fortnight.
What should another retailer copy first?
Cole: Audit your purchase-data latency before touching flows. If store sales reach the ESP weekly, every timing model downstream inherits that lag. Then swap the win-back timer for a churn trigger; it is one flow filter once the data is clean. The pet-supplies engagement shows the same timing principle on the replenishment side.
Client identity anonymized. Figures are representative engagement data, with Klaviyo's predictive-analytics documentation as the reference for model behavior.
Still running win-backs on a timer?
A working session will tell you what a churn-triggered version would send this week, and to whom.
Book a working session