The challenge
Thornbury sells consumable refills on subscription. Their reporting showed new subscribers growing steadily and revenue holding flat, which the team read as a conversion problem and treated with more acquisition spend. The actual failure was at the second billing cycle: a large share of subscribers cancelled before the shipment that would have made them profitable. Because the dashboard counted new subscriptions rather than surviving cohorts, every month looked like growth while the base quietly replaced itself.
What we did
Rebuilt reporting around cohorts, not months
We replaced the subscriber count with a cohort retention view showing what share of each month's joiners survived to cycles two, three and six. The first version of that chart ended the acquisition-spend argument in a single meeting — it made a problem that had been invisible for two years legible in one screen.
Repriced against contribution, not revenue
First-cycle discounting was deep enough that a subscriber who cancelled at cycle two had never contributed anything. We modelled contribution per subscriber across the cycles rather than revenue per order, which set a defensible floor on the introductory offer and identified which acquisition sources were reliably buying customers who never reached profitability.
Rebuilt the gap between shipment one and two
Nothing happened between the first delivery and the second charge — the highest-risk window in the whole lifecycle. We built a sequence around usage: how much product a typical customer has left by week three, how to tell if the cadence is wrong, and how to change frequency instead of cancelling. Offering an interval change rather than a save-offer discount recovered more subscribers and cost nothing in margin.
Made the cancel flow diagnostic
The existing flow offered a discount to anyone who clicked cancel, which paid customers to leave slowly. Replacing it with a reason-first flow routed people appropriately — too much product to a frequency change, wrong product to an exchange — and produced the data showing that over half of cancellations were a cadence mismatch rather than a value judgement.
The result
Cycle-two cancellation fell 31%, which moved lifetime value 126% because the surviving subscribers reached the cycles where the economics work. Contribution per subscriber rose 44% once the introductory offer was set against margin rather than revenue. CAC payback compressed from nine months to two — not because acquisition got cheaper, but because subscribers now survived long enough to repay it. Acquisition spend is roughly where it started.
“We were about to increase paid budget to fix what we thought was a top-of-funnel problem. The cohort chart took ten minutes to build and showed we'd have been pouring money into a bucket with a hole in it. That meeting probably saved us a year.”