Making the shift from reactive to predictive measures of retention means changing what you measure and when.
Stu Sjouwerman, co-founder and CEO, ReadingMinds.ai
Today’s technology offers ample opportunity for companies to track a wide range of data points about their prospects and customers. They can identify the products and services they may be interested in, the purchases made, any issues or problems that were created, returns, ratings, and more.
That’s only true, though, for customers that are sending signals. Not all are. Chances are you have a not insignificant number of customers who aren’t showing up in your churn dashboard. They’re not submitting support tickets. They’re not complaining to sales. They’re not threatening to leave.
But they may be thinking about leaving. And that’s something you need to know.
Unfortunately, silence is the churn signal we have no good mechanism to detect, and that’s the problem.
Retention strategies are generally built around customers who offer indications that something isn’t working for them. They escalate concerns. Their NPS scores drop. They open a new case. They request a meeting.
They’re actually doing you a favor. Because it’s your silent customers you need to be worried about. The ones who simply fade away.
The Feedback That Never Arrives
Here’s an interesting stat to back that up. A recent CustomerThink post from a Samsung study found that only 16% of customers completed satisfaction surveys. Perhaps more troubling: More than half of those who didn’t respond were judged to be dissatisfied. Another survey from Qualtrics found that fewer than one in three customers offer feedback. Customers are becoming increasingly less likely to offer feedback, good or bad. And you, of course, want both.
This isn’t a new phenomenon, and AI is making the consequences more acute.
Think about this: when you train your AI models on survey responses, NPS scores and support tickets, you are in effect, training loyal customers who talk with you, not those that didn’t. That means the model will automatically optimize based on a population that was self-selected but fails to identify accounts at risk of leaving.
Avoidance Is a Signal — It Just Doesn't Look Like One
At ReadingMinds, we’ve learned to look for avoidance patterns, the behavioral footprints that appear long before a customer expresses dissatisfaction. There’s no single data point that can provide this kind of evidence. Taking a look across data points can yield some key insights:
- Core product actions drop below the customer’s own historical baseline, not just below industry benchmarks.
- Reports and outputs that leadership used to see have stopped circulating.
- The customer's response time is lengthening. The contact that used to respond within an hour is now silent for days. Or forever.
- Workflows are started, then abandoned. Friction that used to prompt a support ticket is now just absorbed or worked around.
It’s like a gym member who keeps paying dues but hasn’t used the facility in months. They still show up as an active member but in reality, they’ve quit. They just haven’t told you.
The Dashboard Problem
Most churn dashboards are focused on late-stage signals, the indicators that show up after a customer has already decided to leave. Things like annual recurring revenue (ARR) contraction, dropped seats, or reduced utilization are the result of customers that have disengaged weeks or months earlier.
The problem: by the time the numbers change, you’re not managing risk anymore. You’re managing defection.
Customer satisfaction scores have the same problem. They measure moments, not trajectories. A customer that gave you a score of 7 (out of 10) in January, could have dropped to a 5, 3, or even 1 by September. But without those points appearing in your data (because they stopped responding) you’ll never know.
The sad truth is this: customers most likely to leave are the least likely to tell you so.
Moving to Predictive Measures of Retention
Making the shift from reactive to predictive measures of retention means changing what you measure and when. The framework we’ve found to be most effective involves:
1. Establishing account-level baselines, not segment averages. Don’t focus on deviations from peer groups, focus on deviations in each account’s own patterns.
2. Monitoring behavioral signals continuously, not just at renewal time. A monthly review cycle will miss the avoidance patterns that emerge gradually over time. It’s important to maintain visibility into each account’s activity on a rolling basis.
3. Initiating outreach when signals stop coming to you. The time to initiate outreach is when an account goes quiet. Brief and timely structured AI agent-to-person conversations can surface problems that simply won’t be revealed via email surveys.
4. Paying attention to emotional tone. “Everything’s fine,” can be a very misleading comment. Instead of words, look for signals like shorter responses, flatter tone, and declining engagement. That’s what really tells you something. Energy and enthusiasm are leading indicators. Formal satisfaction scores are lagging.
The goal: intervening at the point of avoidance rather than at the point of anticipated renewal. Customers who are still in the avoidance phase haven’t made a final decision. They’re disengaged but not entirely gone. That’s a meaningful signal and a golden opportunity to connect and re-engage before they depart. You can save their business if you interpret the right signals.
About the Author
Stu Sjouwerman (pronounced “shower-man”) is co-founder and CEO of ReadingMinds.ai. He is the author of "Agent Powered Growth: Deploy AI Agents that Build Your Marketing Pipeline 24/7" [Wiley, 2026], which became a national bestseller, debuting at No. 17 on the USA Today's Bestselling Book List within two weeks of its publication.
