Three phases.
One compounding thesis.
150K subscriptions canceled in 2025. Most of them didn't have to. I've been designing Chatbooks' cancellation save system since V1 — each version using data from the last to get sharper. V2 cut the cancellation rate 23% relative. V3 is now in the factory.
23%
RELATIVE REDUCTION IN CANCELLATION RATE, V2 vs V1
89%
RETAINED AT 30 DAYS (V2) — vs 51% IN V1
53%
CREDIT REDEMPTION WITHIN 90 DAYS (V2) — vs 30% IN V13%
$1.1m
GCR UNLOCKED V3 (PROJECTED) 3%
COMPANY
Chatbooks
Photobook Subscription Software
MY ROLE
Senior Product Designer
Strategy, research, design, cross-functional alignment across 3 versions
COLLABORATORS
Product, Engineering, Analytics, Customer Success, BizOps
TIMELINE
2024-2026
V1 shipped June 2024 · V2 shipped May 2025 · V3 shipped Q1 2026
01
THE PROBLEM
We were just letting people leave.
The V1 cancellation flow had two screens: a reason dropdown and a confirm button. No saves. No alternatives. No friction between a subscriber and the door. In 2024, roughly 49,000 active subscriptions churned through that flow.
The more interesting data point: 40% of cancellation reasons had nothing to do with the product. Timing, budget, a baby's first year wrapping up. Temporary friction masquerading as permanent disengagement. Industry research said targeted save offers could reduce churn 10–20%. We were offering nothing.
The other problem: the reason taxonomy was too coarse to act on. "Too expensive" and "not using it enough" covered the majority of responses — which told us how many people were leaving, but nothing about what would actually keep them.
"Depending on the book you were creating, you could make the same type of edit in five different ways. No user should have to learn that."
02
STRATEGY
Don't guess. Build the data first, then the saves.
Working with Ops, Analytics, and Customer Success, we structured salvage in three releases. The logic: before we could design the right save, we needed to know exactly why people were leaving — at a level of specificity that the V1 taxonomy couldn't give us.
The first release rebuilt the reason flow and introduced a single save offer (a $5–$10 credit) primarily to establish a baseline and generate clean signal. The second release used that data to build reason-aligned save paths — a real decision tree where the offer matched the stated problem.
This sequencing wasn't the exciting answer. But it was the right one.
PHASE 1
One save. Better data.
Introduced a single save offer — a $5–$10 credit based on product.
More importantly, rebuilt the cancellation reason taxonomy with the CS team — more specific categories that would give us signal, not noise.
CANCELLATION RATE: 86% OF SUB ORDERS CHURNED (JAN–JUN 2024 BASELINE)
PHASE 2
Personalized saves
Four save paths based on cancellation reason: adjust cadence (12, 6,
or 4 books/year), connect with support, turn on notifications.
CANCELLATION RATE DROPPED TO 66% (JUL–DEC 2025) — 23% RELATIVE REDUCTION
PHASE 3
Smarter Saves & Pause
Further refinement of cancellation reasons and saves, plus a new way to switch between products to find a better fit. We also shipped subscription pause — a net-new feature.
IN PROGRESS...
03
THE DESIGN
Saves that match the reason.
The core logic of V2: the offer is a response, not a discount. Someone canceling because they have too many books doesn't need a $10 credit — they need fewer books. Someone canceling because it's too expensive should see a lower cadence before they see the door. Someone who says "I'm coming back" should be offered a way to pause, not forced to churn and reacquire later at a higher CAC.
Placement mattered as much as content. The save offer appeared after the stated reason — acknowledging it first, then responding to it. A used-car-lot approach (offers before reasons) tends to increase resentment, not saves. The Psychology background comes in handy here.
The cadence-switching option required meaningful internal alignment — nobody loves giving subscribers a way to spend less. The data made the case: customers who switched to a lighter cadence had higher print rates and better LTV because their plan actually fit their behavior.
04
THE OUTCOMES
23% fewer cancellations. And the saves were stickier.
The headline: cancellation rate dropped from 86% to 66% of subscription orders — a 23% relative reduction — across 69,000+ sub orders in the measurement period.
But the more important number was downstream. Subscribers saved through V2 weren't just retained — they were more engaged. 89% were still active at 30 days, compared to 51% in V1. They stayed active 30 days longer on average. And 53% redeemed a credit within 90 days, vs. 30% under V1. Saves that actually stuck, not saves that delayed churn by a billing cycle.
The analytics team wrote custom SQL to measure this precisely — tracking every save event, every cancel confirmation tap, and 7–30 day churn outcomes by cohort. The data was clean enough that we could directly compare V1 and V2 retention curves and make the case for V3 based on evidence, not intuition.
23%
RELATIVE REDUCTION IN CANCELLATION RATE — 86% TO 66%
89%
RETAINED AT 30 DAYS FOR V2 SAVES — vs 51% IN V1
53%
CREDIT REDEMPTION WITHIN 90 DAYS — vs 30% IN V1
05
V3 IN THE FACTORY
V2 proved the thesis. V3 builds on the biggest signal it surfaced.
The clearest finding from V2: "I'm coming back" was the #1 salvage reason and the #2 non-passive cancellation reason overall. 1,100+ subscribers — across both Monthbooks and Monthly Minis — were telling us they didn't want to leave permanently. They wanted to pause. And the product had no way to let them.
Today, that's handled manually by the Customer Success team: creating Intercom tickets, adjusting billing dates, following up. It works, but it doesn't scale. Of customers who accepted a manual pause offer, 55% reactivated within 3 months. Another 9% reactivated independently without CS intervention.
V3 — now in development for Q1 2026 — brings Skip a Payment in-app as a self-serve feature, adds cadence switching for Monthly Minis → Monthbooks, and rebuilds the save flow around flexibility rather than discounts. Projected impact: 27K–53K subscriptions saved in 2026, up to $1.1M in GCR, and 1,600+ Support hours freed annually.
What this work actually is
It's a compounding system. V1 shipped a flat offer and generated signal. V2 used that signal to build something targeted — and generated better signal. V3 is using V2's clearest finding (pause intent is #1, and we can't serve it in-app) to build the next layer. Each version makes the next one possible. That's the whole strategy.