Churn is a system problem, not a product problem.

What we build

• Churn risk, scored before the click A risk score per active subscriber, fitted on your own cancellation history — the order gaps, the skips, the pauses, the support contact, the failed charge three weeks ago. It ranks who is leaving and names the pattern it is reacting to, so an intervention can reach someone in week six rather than meeting them on the cancel page.
• Cancel flow, properly built Not the platform default. A real interface: reason capture first, then a save offer matched to the reason. “Too much product” gets a cadence change, not a discount. “Too expensive” gets a smaller size. The match starts as a prediction rather than a question — the system has already read the delivery gaps, the skips and the tickets and has a likely reason ranked before the customer picks one, which is what makes the offer specific instead of a blanket 20%. Discounting everyone is how you train churn.
• Pause, skip & swap Made obvious rather than buried, and surfaced to the subscriber the score says is about to need one. A pause is worth vastly more than a cancel and most brands hide the button.
• Quantity & cadence self-service The three things that actually change: how much, how often, which size. A household that grew, a customer travelling for six weeks, someone who halved their usage. If they can’t change it themselves, they cancel instead — and the account usually knew first, because four unopened refills show up as a delivery pattern long before they show up as a complaint.
• Prepaid & annual upgrade paths Built as a flow and offered at the point the data reads as highest satisfaction — after a delivered order, a clean payment history and no recent ticket — not on the pricing page.
• Replenishment timing model Consumption fitted per SKU and per customer segment from your own order gaps and skip behaviour, then cadence and reorder prompts set against the predicted run-out date rather than the platform default. Most “too much product” cancellations are a cadence that was never right.
• Cohort & NRR reporting So you can finally see retention by cohort, by SKU, by acquisition channel — and voluntary churn separated from involuntary. The risk score is reported next to it with the reason it gave, so a save offer can be argued with rather than taken on faith.
• Winback architecture Segmented by the reason they gave and by the reason the data suggests when the two disagree, sequenced over 30/60/90 days.

Benchmarks — subscription churn

Sources: Recurly Consumer Goods Churn Benchmark, Recharge DTC consumer panel.

Voluntary and involuntary churn are listed separately because they are two different repairs. The involuntary half is a billing problem — that one is payment recovery.