Customer service software company Groove tracked what its cancellations had in common and found the tell wasn't a support ticket or a bad review. It was the first session. Customers who went on to cancel averaged a first session of just 35 seconds and logged in 0.3 times per day. Customers who stayed averaged a first session of 3 minutes 18 seconds and logged in 4.4 times per day. After building alerts around these "Red Flag Metrics" and intervening early, Groove cut its monthly churn rate from 4.5% to 1.6%, a 71% reduction (Groove HQ).
That gap matters because it shows up before a customer ever fills out an exit survey or gives a low satisfaction score.
Usage data outperforms sentiment surveys
Most retention programs are built around asking customers how they feel: NPS, CSAT, quarterly check-in calls. But a study of more than 75,000 customer service interactions found that Customer Effort Score, a measure of how much friction a customer experienced getting what they needed, predicted loyalty and churn risk better than either satisfaction scores or Net Promoter Score. The researchers' core finding was that reducing customer effort, not exceeding expectations or "delighting" customers, is what actually keeps people from leaving (Harvard Business Review).
Sentiment is a lagging indicator. A customer who scores you a 9 out of 10 in January can still cancel in March if the product quietly stopped fitting into their routine. Behavior, logged sessions, session length, feature depth, is a leading one.
What customers say when you finally ask
When customers do explain why they left, the answer is rarely price. In a survey spanning 76 million unique subscriptions across 2,200 global merchants, 51% of consumers said they canceled a subscription because they were "not using it enough," making low product usage the single leading driver of cancellation, ahead of cost (Recurly, 2026 State of Subscriptions). The product didn't fail on features. It failed to stay in the customer's routine long enough to justify the line item.
That's the same signal Groove was catching weeks earlier through login frequency. The cancellation is just the paperwork on a decision the usage data already made.
Why the economics reward catching it early
The financial case for building this kind of early-warning system is not subtle. Acquiring a new customer is estimated to cost 5 to 25 times more than retaining an existing one (Harvard Business Review). And the compounding effect of small retention gains is larger than most operators assume.
Frederick Reichheld and W. Earl Sasser Jr. found that reducing a company's customer defection rate by just 5% raised profits by 25% to 95% depending on industry, including 85% more profit in one bank's branch system, 50% more in an insurance brokerage, and 30% more in an auto-service chain (Harvard Business Review, "Zero Defections").
The same logic shows up at the account level. In a Bain & Company analysis of Dell's customer base, detractors, the customers who gave the company low loyalty scores, made up 15% of the base and cost Dell an estimated $68 million. Converting half of those detractors to average customers was projected to add more than $160 million a year to the bottom line (Bain & Company). The money isn't in preventing every cancellation. It's in catching the accounts that are already sliding before they become a lost renewal.
What to build this week
Skip the quarterly NPS survey as your early warning system. Pull the data you already have:
- First session length and time-to-first-value for every new account in the last 90 days
- Login frequency per account, trended weekly, not just at renewal time
- A simple threshold flag: accounts whose usage drops below their own 60-day average
Groove didn't need a sentiment model. It needed a dashboard that flagged the 35-second sessions before the 90-day mark, and a person whose job was to call those accounts. Build the same thing before the next renewal cycle, not after the cancellation email arrives.