The Number That Recommends Itself

A metric stops being a metric the moment it becomes a goal. Nowhere is this law more visible than in NPS.

The trap of clarity is usually described in the abstract. Goodhart's Law, the gap between a metric and its target — these phrases are true enough, but hard to hold in your hand. If you want to see the trap in one concrete case, marketing and customer experience offer few better examples. A single question, a single formula, a single number — and a story of how it became a global standard, then turned against the very thing it was meant to measure.

Imagine explaining an entire organization with one number. No ambition has been realized more successfully than in the Net Promoter Score. The question is just one: "How likely are you to recommend this brand to a friend or colleague?" Respondents answer on a scale from 0 to 10. Those who answer 9 or 10 are Promoters, 7 or 8 are Passives, 0 through 6 are Detractors. Subtract the percentage of Detractors from the percentage of Promoters, and you have the NPS. No lengthy questionnaire, no layered framework of metrics — one question, one calculation, and the entire customer relationship collapses into a single figure.

That simplicity was the appeal. It was also, as it turned out, the exact condition under which the clarity trap runs cleanest. The law that a metric stops being a metric once it becomes a target works faster and more completely the simpler the metric is. A metric with many variables offers many places to distort — but it also leaves many traces of that distortion. A metric built on a single question narrows the point of distortion to one place, and moving that one place is enough to move the whole number. NPS is close to a perfect specimen of this.

Can One Question Really Predict Growth?

NPS was first proposed in 2003, in a Harvard Business Review paper by Fred Reichheld. The title alone was a provocation: "The One Number You Need to Grow." The claim was that a single measure of recommendation intent correlated strongly with a company's growth. The title was already the conclusion — and that conclusion turned out to be the paper's most effective piece of marketing.

Knowing the context helps explain why the claim spread so fast. Until then, measuring customer satisfaction usually meant long questionnaires and a patchwork of metrics. Many questions, multiple scales, and a connection to actual revenue or growth that always stayed blurry. Interpreting survey results took expertise; convincing executives of that interpretation took more. Against that backdrop, the proposal that "one question is enough" felt like a kind of liberation.

Picture it from an executive's chair and the appeal sharpens. Instead of a complex dashboard, you get one number — a number you can line up across departments, branches, countries, quarters. Comparing this month to last required no statistics, just whether the number went up or down. And because the scale was universal, companies from entirely different industries could be measured against each other with it. An airline, a bank, a coffee chain — all ranked on the same scale from 0 to 10. That universal comparability, more than anything else, may be the real reason NPS spread faster than any other customer metric.

NPS quickly became a standard fixture across companies worldwide — in board reports, in bonus formulas, on the individual scorecards of call-center agents. Few metrics built on a single question have been used this widely, at this many levels, for this long.

So far, this follows the familiar arc of every story about the clarity trap. Complexity meets a simple number; the simplicity becomes the appeal; the appeal becomes adoption. The trouble starts later — the moment the number stops being something you report and becomes something you're targeted on.

How to Game a Single Number

Once NPS was tied to performance reviews and compensation, companies stopped managing the number and started manufacturing it — with real precision. No one says "let's fake the score" in a meeting. Instead, the same thing gets discussed under names like "improving survey response rates" or "optimizing the final moment of the customer journey." The intent is identical; the language just changes enough that no one in the room would call it manipulation.

• A service rep asks directly, before hanging up: "We'd appreciate a 10."
• Survey invitations are timed for right after purchase — the moment satisfaction peaks — to inflate the average.
• Customers who look likely to become Detractors simply never receive the survey.
• Customers who've already complained get the survey late, or not at all.
• The most loyal customers get surveyed repeatedly, tilting the sample in the company's favor.
• Surveys go out right after a customer tries to cancel and gets talked into staying with a discount — right when they're feeling good about the save.

Each of these, taken alone, looks trivial. One rep's script. A day or two of timing. A handful of names added or dropped from a send list. None of it resembles fraud. But repeat these adjustments across an organization, every quarter, across multiple teams, and something changes. The NPS score drifts further and further from the customer experience the company is actually delivering. Not manipulation so much as optimization — which is exactly how the gap between the number and reality is allowed to widen.

Executives look at a high NPS and read it as a healthy customer relationship. On the ground, a structure is already in place that keeps dissatisfied customers from ever being heard.

What's worth noticing is that most of the people making these adjustments had no bad intent. The rep was just trying to hit his own scorecard target. The CX manager was just trying to make the quarter. Each acted rationally within their own role. It's the accumulation of those rational acts that opens the distance between the number executives see and the experience customers actually have.

And that distance becomes less visible the higher up you go. The call-center supervisor knows exactly why the rep says what he says. The CX team wrote the send logic, so they know about the timing adjustments. But above them — the executive who receives only the quarterly NPS figure — has no way of knowing the path that number traveled to reach the report. The higher the number climbs, the smoother it looks, and the more the fingerprints on it disappear. The target gets hit. The real relationship the number was supposed to stand in for may have quietly gotten worse in the meantime. But by the time anyone in a position to notice receives anything, all they receive is the number.

Does the Correlation Actually Hold?

A deeper problem existed independent of any gaming. NPS's academic foundation itself became a matter of dispute.

After Reichheld's 2003 paper, other researchers re-tested the correlation between NPS and actual company growth. The results didn't support a relationship as strong as the original paper claimed. The relationship varied by industry, varied by competitive environment, and shifted even when the same concept was measured with slightly different methods. The certainty promised by a title like "The One Number You Need to Grow" didn't survive the scrutiny of the studies that followed. In some industries the relationship was clear; in others it barely showed up at all. Reality turned out to be far rougher than a single number, applied uniformly across every industry and every competitive landscape, could account for.

On reflection, this isn't surprising. For recommendation intent to translate into actual growth, the person hearing the recommendation has to be able to act on it — to actually switch brands. In a market with low switching barriers, one recommendation might convert directly into a new customer. In a market with high switching costs or few alternatives, even the strongest recommendation intent runs into other, larger obstacles before it becomes an actual move. Expecting one question to predict growth with equal power across every kind of market may have been an unreasonable ask from the start.

By the time these academic doubts surfaced, NPS had already secured a fixed spot on the management dashboards of thousands of companies worldwide. It was wired into compensation. Organizational processes had been built around the number. It was a permanent line item in executive reporting. Even as its original foundation wobbled, the metric had already gained enough institutional momentum to survive. Almost no company waited for the paper's claims to be validated before adopting it. Simplicity drove adoption first; verification trailed slowly behind. Few cases illustrate as clearly as NPS that a metric's legitimacy and a metric's survival are two entirely different questions. The academic doubt didn't shake the metric — the metric had already grown too sturdy to be shaken by academic doubt.

What a Single Number Erases

Set aside the gaming and the question of evidence, and NPS still has a quieter problem left. The act of compressing everything into one number flattens genuinely different textures of customer experience.

Take a dissatisfied customer stuck with a brand because cancellation is too complicated. That customer might still give a high score on the survey — the immediate feeling isn't bad enough to justify the friction of a low score and the conflict that might follow. Or take the opposite: a customer with high recommendation intent but low actual purchase frequency — someone whose goodwill hasn't followed their wallet. On the NPS score alone, these two customers are indistinguishable. Two entirely different relationships with the brand get folded into the same figure.

The question itself — "how likely are you to recommend" — deserves scrutiny too. Actually recommending something to someone is never determined by satisfaction alone. It depends on your social relationship with that person, on whether the product is even the kind of thing that comes up in conversation, on whether someone happened to ask at the right moment. A satisfying product with no reason to come up in conversation generates no recommendations; a mediocre product mentioned at the right moment in the right conversation might generate several. NPS tries to capture all of this texture in a single answer between 0 and 10. Simplification does produce clarity. But the price of that clarity is that the complexity of the actual situation gets pushed entirely outside the frame of the question. Two customers can give the same score for completely different reasons — and those different reasons are exactly the information that would tell you how to treat each of them next quarter. The number never asks why.

This distinction blurs more often than you'd expect in practice. "We track NPS" and "our goal is to raise NPS" sound almost the same, but they produce entirely different behavior inside an organization. The first treats the number as a signal and keeps asking what's behind it. The second treats the number as the destination and looks for the fastest route there. The distance between those two sentences is exactly the distance this piece has been circling all along.

Reichheld's paper was titled "The One Number You Need to Grow." If that title had been right, this piece wouldn't need to exist. What actually happened was closer to the opposite. The attempt to explain growth with a single number produced, at the same time, a sophisticated apparatus for protecting that number — and an academic argument over whether the number ever meant what it claimed.

There's a second layer of irony here. The metric was loved, at first, for making something complicated simple. What it eventually produced was a landscape more complicated than the one it replaced — the complexity of auditing how the number was made, the complexity of re-verifying what it actually correlates with, the complexity of re-segmenting the customers a single number failed to describe. A metric that arrived to remove complexity ended up creating a new layer of it, and staying anyway.

The number is still on the dashboard. It will appear again in next quarter's report. But what that number is actually telling you can no longer be answered by the single question that produced it. Answering that now requires a second question: when, to whom, and how was this number even asked? What it takes to actually know whether you're growing turned out not to be one number, but the habit of continuing to interrogate it.

The next time someone reports an NPS score in a quarterly meeting, the real question isn't "what's the number." It's "how was this number made." If that question isn't on the agenda, the organization isn't managing the number anymore. The number is managing it.

Ask them to recommend you, and you'll get an answer. Ask if that answer means anything, and the silence starts.