NPS has had a remarkable run. A single survey question became a boardroom metric, a quarterly ritual, and for many teams the primary input into account health. It is also, as a health proxy, one of the weaker signals you can build on. Not because asking customers how they feel is wrong, but because of who answers, when they answer, and what their answer actually predicts.
The structural problems with survey-based health
- Response bias selects out the people you most need to hear from. The customers quietly disengaging, the ones who have already mentally moved on, are precisely the ones who do not fill out your survey. Your response set skews toward the engaged, which means your score skews optimistic exactly where it matters least.
- It is a lagging snapshot. A quarterly survey tells you how someone felt on the day they happened to respond. By the time the results are tallied, the account's situation may have changed entirely. Health is continuous; surveys are periodic.
- The respondent is rarely the decision-maker. The end user who scores you a 9 may love the product, while the economic buyer who never answers the survey is questioning the line item. NPS captures affection, not the budget decision that determines renewal.
- It is gameable and gamed. Once NPS becomes a target, teams learn to time the ask after a good experience and suppress it after a bad one. A metric that is managed for is no longer a measurement.
What behavioral signals do better
The case for signals over surveys is not that opinions do not matter. It is that behavior is a more honest and more timely expression of opinion than a survey response, and crucially, it is available for every account, not just the ones who reply.
- Coverage is complete. Every account generates behavioral signal: product usage, support tone, email latency, meeting attendance. There is no response rate to worry about, because customers are not opting in, they are simply using the product and talking to you.
- It is continuous. Signals update as they happen rather than once a quarter, so risk surfaces in days, not at the next survey window.
- It captures the people who matter. The economic buyer who skips the survey still appears in the data when they stop attending QBRs or stop replying to email. Disengagement is a behavior, and behaviors are observable even when opinions are withheld.
- It is hard to game from the inside. You cannot make an account healthy by timing a question well. The signals reflect what is actually happening in the relationship.
A survey tells you what a customer was willing to say. Their behavior tells you what they actually believe. When the two disagree, trust the behavior.
Keep the survey, demote it
None of this means deleting NPS. A direct verbatim from a customer is genuinely valuable context, and the open-text comments are often more useful than the score itself. The argument is narrower: the survey should be one input among many, not the spine of your health model. Demote it from primary metric to supporting evidence, and let continuous behavioral signal carry the weight. For the full inventory of which behaviors predict churn, see the leading indicators of B2B churn.
How Merrily reads signal instead of asking
Merrily is built on the signals-over-surveys premise. Rather than sending another survey, it reads the behavioral evidence already flowing through Slack, Gmail, meeting notes, product analytics, and billing, and turns it into a live health score with no response rate to manage and no quarterly lag. The customers who would never answer a survey are scored just as completely as the ones who would, which is the entire point. The accounts you are most likely to lose are the ones least likely to tell you, and a survey will not reach them. Their behavior already has.