The metric that makes everyone comfortable is measuring the wrong thing.

The quarterly review opens with the good news. NPS is at 42, up six points on last quarter. There’s a small round of nodding. Someone says “great work, team” and genuinely means it. The slide has a big green number on it, and green numbers end conversations rather than start them, so the meeting moves on.

Three slides later, in a different deck, owned by a different team, churn for the same customer base is up 14%. Nobody in the room connects the two numbers. They live in different meetings, reported by different people, and by the time anyone notices the pattern, the quarter that produced it is long finished.

This isn’t an oversight. It’s what happens when an organisation measures the thing that’s easiest to defend in a room, rather than the thing that actually explains what’s happening to the business. And no metric gets defended in more rooms, with less scrutiny, than NPS.

The drunk man and the streetlight

There’s an old parable, sometimes attributed to a Mulla Nasreddin story, sometimes to nobody in particular, about a man searching for his lost keys under a streetlight. A passerby offers to help and asks where he dropped them. “Over there,” the man says, pointing into the dark car park. “Then why are you looking here?” “Because this is where the light is.”

Psychologists and statisticians call this the streetlight effect: the tendency to search where visibility is good rather than where the answer actually is. It isn’t stupidity. It’s a completely rational response to an irrational constraint. The dark car park is where the keys are. The streetlight is where you can actually see something.

NPS is a streetlight, and it’s a particularly well-lit one. One question. One number, minus ten to plus ten. Comparable across your entire industry because everyone else asks the same question the same way. Simple enough to explain to a board in one sentence, simple enough to track on a single line going back years. Almost nothing else in a business gets measured this cleanly. That cleanliness is precisely the problem: it’s easy to confuse “measurable” with “meaningful,” and NPS is one of the most measurable, least meaningful numbers most organisations own.

A number nobody can explain

Here’s the part that CX professionals rarely say out loud in the room, though almost every one of them will admit it privately, usually after the second drink at a conference. Nobody actually knows why the score moved.

It went up four points this quarter. Somebody will offer a story in the QBR: the new onboarding flow, the marketing campaign, the weather being nicer during the survey window. The story is plausible. It is also, more often than not, invented after the fact to explain a number that arrived with no explanation attached. NPS tells you the temperature. It has never once, in the history of the metric, told anyone what’s causing the fever.

This is the real irony of NPS, and it’s a bigger problem than the score being unreliable. It’s that the people running it for a living already know it’s unreliable, and use it anyway, because a number that shows up cleanly every quarter beats a diagnosis that takes actual work to produce. A single score with no attached cause isn’t a measurement. It’s a mood ring with a spreadsheet behind it.

Comfortable is not the same as correct

The uncomfortable part isn’t that NPS is a bad question. “How likely are you to recommend us to a friend or colleague” genuinely captures something: a customer’s willingness, at that moment, to put their own reputation behind you. The problem is what happens next, which is that the organisation quietly starts treating that one number as a stand-in for “is this relationship healthy,” and those are not the same question, even though they share a survey.

A customer can give you a 9 and still be quietly shopping your competitors. They liked the sales call. They haven’t tried to cancel yet. They haven’t hit the moment that actually tests the relationship, the billing error, the outage, the third call about the same unresolved issue. NPS is usually captured early, often right after onboarding or a purchase, precisely when goodwill is at its peak and the relationship hasn’t been tested by anything yet. It’s measuring the honeymoon and reporting it as the marriage.

The score feels safe because it’s defensible. Nobody gets asked a hard question about a single number that’s trending up and matches the industry benchmark. That’s precisely what should make everyone suspicious of it.

Nobody actually recommends anyone

There’s a second problem sitting underneath the first one, and it’s baked into the question itself. “How likely are you to recommend us to a friend or colleague” asks a customer to predict a piece of social behaviour that, for the overwhelming majority of categories being measured, essentially never happens.

Nobody sits around at dinner recommending their insurance provider. Nobody brings up their telco unprompted at a barbecue. People recommend restaurants, films, and the occasional tradesperson who didn’t overcharge them. They do not, in the ordinary run of life, spontaneously advocate for the company that processes their electricity bill, and yet every one of those companies is asking customers to imagine a scenario that will almost never occur and rate their likelihood of doing it.

Researchers have a name for what happens when you ask people to predict a hypothetical behaviour rather than observe an actual one: hypothetical bias. People are reliably bad at forecasting their own future actions, particularly social ones, and the gap between the stated intention and the real behaviour is well documented and rarely small. NPS doesn’t measure whether anyone will recommend you. It measures how a customer feels about answering an imaginary question they’ve never had to act on, which is a different thing wearing the same number.

The scale that only counts a fifth of itself

Then there’s the arithmetic, which has its own quiet irony built in. NPS runs on a zero-to-ten scale. Anyone scoring nine or ten is a promoter. Anyone scoring zero to six is a detractor. The score is promoters minus detractors, as a percentage.

Sit with that for a second. A 7 or an 8, on a ten-point scale, is a genuinely positive response by any normal reading of the English language. In a school system, it’s a B. In almost any other context, it’s a pass, comfortably. In NPS methodology, it’s neither promoter nor detractor. It’s a passive, and it counts for precisely nothing in the final number. It doesn’t add. It doesn’t subtract. It vanishes.

Which means a company can have every single customer answer 7 or 8, a result any sane business would frame as broadly satisfied, and post a score of zero. The methodology doesn’t have a category for “content but not evangelical,” even though that describes the actual emotional state of most customers, most of the time, about most things. NPS was built to reward enthusiasm and punish complaint, and it simply has no vocabulary for the enormous, ordinary middle where the real relationship lives.

Goodhart’s law and the slow drift

There’s a third mechanism working against you here, and it has a name too. Goodhart’s law states that when a measure becomes a target, it stops being a good measure. The moment “improve NPS” becomes an actual KPI with a bonus attached, people don’t necessarily improve the experience. They improve the number, and those are two different projects that happen to overlap for a while before quietly diverging.

Send the survey the moment a deal closes, before the product has been used in anger. Coach the team to ask detractors to “update their score” after a call, which is a polite way of asking someone to resurvey a customer until the answer looks better. Exclude the accounts most likely to complain from the sample because they’re “not representative.” None of this is fraud. It’s just what happens when smart people are held accountable to a number rather than to the thing the number was supposed to represent.

The score keeps climbing. The organisation keeps congratulating itself. And the gap between what the dashboard says and what the customer is actually experiencing widens every quarter, invisibly, because the thing measuring the gap is the same thing causing it.

What the streetlight can’t show you

The metrics that would actually tell you something uncomfortable tend to look nothing like this. Repeat contact rate: did the customer have to come back for the same issue. Resolution durability: did the fix still hold three months later, or did it just get the case closed today. Actual behaviour: did they renew, expand, or refer somebody, rather than did they answer a single question kindly on a Tuesday.

These are harder to capture. They take longer to show up. They require joining data across teams who currently don’t talk to each other, which is itself a fairly reliable sign that nobody’s incentivised to look. And they occasionally produce a result nobody wants to present at the quarterly review, which is exactly why organisations tend to avoid building them in the first place.

That avoidance isn’t accidental. A metric with the power to embarrass someone in the room gets far less organisational enthusiasm than one that reliably makes everyone feel like they’re doing a good job.

The real problem

None of this means NPS is worthless, or that every board should stop asking for it tomorrow. It means the comfort is the tell. If a metric has never once produced an uncomfortable conversation, or an explanation anyone actually believes, that isn’t evidence the relationship is fine. It’s evidence the metric was never built to find out.

The organisations that actually improve are the ones willing to go looking in the dark car park, with a proper torch, even though the streetlight was working perfectly well and telling them everything was fine. It usually wasn’t. It was just well lit.