We ran customer satisfaction scoring for years, the way everyone does. In a quarterly review we accepted something uncomfortable: the score had never once warned us about an account before the account told us itself. It was a report, not an instrument.
Three problems with the number
It surveys the wrong people. The customers who answer a satisfaction survey are the delighted and the furious. The large middle — quietly dissatisfied, slowly disengaging, the ones who actually churn — do not respond at all.
It is not comparable. A four out of five from one organisation and a four from another mean different things, because the expectation baselines are different. Averaging them produces a number that describes nobody.
It arrives after the decision. By the time satisfaction moves measurably, the customer has usually already decided how they feel. You are measuring an outcome you can no longer influence.
What we replaced it with
A health check built from customer-relative indicators rather than an absolute score. The two that carry most of the signal are consumption — is this customer using the platform more or less than they were, on their own baseline — and their own definition of good, captured explicitly rather than inferred from a five-point scale.
Consumption is the honest leading indicator in a platform business. An account whose interaction volume is drifting down is disengaging, and it shows weeks before anyone says so in a meeting. That is enough time to do something.
What we kept
Interaction-level satisfaction, where it is immediate and specific, still earns its place: it is close enough to the event to be actionable and it feeds quality scoring. What we stopped doing is rolling it up into an account-level number and treating that as the health of the relationship.
The general lesson is not "CSAT is bad". It is that a metric you cannot act on before the outcome is a report, and reports do not change outcomes.