Ten or fifteen years ago, adopting Net Promoter Score was the right call. Before it, customer experience had no common currency: no single number a chief executive could track alongside revenue and margin, no way to compare a bank against a telco against an airline, no simple story to tell a board about whether customers were happier this quarter than last. Fred Reichheld’s 2003 case for “the one number you need to grow” solved a real problem. It gave CX a seat at the table it had never had, with a question simple enough that response rates held up and trend lines meant something. Organisations that built a CX programme around NPS in the years that followed were not being naïve. They were doing what any sensible, cost-conscious organisation should do: picking the best available proxy for something that mattered, because the alternative was measuring nothing at all.
The problem is not that NPS was ever the wrong choice for that job. It is that many organisations have kept using it for a bigger job it was never built to do: not just a sentiment barometer, but the operating system for the whole CX programme, the number that decides where investment goes and whether the programme is judged to be working. Held up against that job, three structural weaknesses are hard to ignore.
1. A score with no lever
NPS tells an organisation where it stands, not what to change. Two organisations with an identical score of thirty can be driven by entirely different experiences, one losing customers over slow claims handling, the other over confusing pricing, and the number alone cannot tell either of them which lever to pull. Recent independent research makes the problem worse than it looks. A 2024 review of the accumulated evidence on NPS found its own arithmetic to be unstable: because it counts only respondents who choose 9 or 10 as promoters and 0 through 6 as detractors, modest shifts in the middle of the distribution produce score swings far larger than the change in underlying sentiment they are meant to represent (1). An organisation can watch its NPS move by ten points between survey waves with almost nothing having genuinely changed in the customer experience, and the score gives no way to tell whether that movement is signal or noise. Most organisations respond in the only way the metric allows: launch an initiative, wait a quarter or two, check whether the number moved. That is not a diagnostic process. It is trial and error, run at the pace of a survey cycle.
2. What people say versus what people do
NPS asks a single hypothetical question: how likely are you to recommend this to a friend or colleague. It is asking people to predict their own future behaviour, and behavioural science has spent decades establishing that stated intentions are a weak predictor of what people actually go on to do.
Across the accumulated behavioural science literature, stated intentions explain roughly a quarter of the variation in what people actually do afterward. NPS asks customers to state an intention. The other three quarters is where the metric goes silent.(2)
This is one of the most replicated findings in behavioural psychology, and it applies with particular force to recommending. “Would I recommend this” is a low-stakes hypothetical answered inside a survey. “Did I actually recommend this” requires the same person to remember a real product weeks later, find an appropriate moment, and volunteer an opinion in a real conversation. Research testing NPS directly against this gap finds what the theory predicts: stated likelihood to recommend correlates weakly with the online word of mouth customers actually generate (3), and a 2024 study asking the question head on, does NPS predict word of mouth, found the relationship far weaker than the metric’s popularity implies (4). A high promoter score tells you something about a customer’s goodwill in that moment. It does not tell you whether they will ever act on it.
3. A sentiment reading sold as a growth forecast
NPS built its reputation on a specific promise: promoters grow companies and detractors shrink them, so track the balance and you are tracking future growth. That promise does not survive close inspection. In one of the most cited independent tests of it, a Journal of Marketing study following firms over several years found NPS was not the superior predictor of revenue growth its original research claimed. It performed no better than the ordinary satisfaction metrics it was designed to replace (5). The 2024 review referenced above reaches the same conclusion from a different angle, examining NPS’s presumed links to firm growth, negative word of mouth and loyalty, and finding the evidence for each thinner than the metric’s ubiquity suggests. None of this makes promoters worthless, or means sentiment does not matter. It means NPS measures exactly what it was built to measure: how customers feel right now. It was never built to measure, and cannot be made to measure, whether a customer stays when a competitor becomes more attractive, or whether a prospect who has never used the product chooses it over an alternative next month. A feelings reading has been quietly asked to do a forecasting job, and the two are not the same thing.
None of this makes NPS a bad decision in 2012. It makes it an incomplete one now. A metric that cannot tell you what to fix, cannot reliably predict the behaviour it claims to forecast, and was never built to reach retention or acquisition in the first place, is still a perfectly good sentiment gauge. What it cannot be, on its own, is the operating system for a CX programme that is meant to grow the business.
From sentiment to cause
The alternative is not a better survey question. It is a different kind of diagnostic, built to answer what NPS was never designed to ask: what is actually driving a customer’s choice, and what would have to change for that choice to go the other way. Causal CX answers this through structured research into three things. The Jobs a customer is genuinely trying to get done: the functional, emotional and social progress they are hiring a product or service to make, not just what they say they would like. The Blockers standing between them and choosing, or continuing to choose, you: the anxieties, habits and switching costs that hold a decision in place regardless of how satisfied someone says they are. And the CX pain points and frustrations that surface across the experience, read not as an unranked complaints list but against the jobs and blockers they actually sit beside.
Together, these produce a performance framework: a prioritised, evidence based account of the dimensions that genuinely determine whether a customer chooses you, keeps choosing you, or walks. Read an organisation’s existing journey maps and pain point data against that framework, rather than against complaint volume alone, and the picture changes. Some loudly reported pain points sit on dimensions that barely influence choice at all. Some quiet ones, rarely mentioned unprompted, turn out to be the exact reason customers leave, or the exact reason a switcher chose a competitor instead. The resulting roadmap is prioritised by causal impact on retention and acquisition, not by how often an issue is mentioned or how easily it can be fixed. This is what NPS could never provide on its own: not another score to track, but a working explanation of why the score is what it is, and precisely where to intervene to move it for real reasons rather than survey noise.
Sentiment still matters, and nothing here argues for abandoning NPS or the tracking discipline built around it. What changes is what it is asked to do. Treated as a lagging indicator to be explained, NPS remains a useful ongoing check. Treated as the primary diagnostic deciding where CX budget goes, it is being asked to do a job that a decade of independent research says it structurally cannot do.
Ready to start the conversation?
Ask what your last NPS improvement initiative was specifically supposed to fix, and whether anyone could explain, before the next survey wave landed, why it would work. Most CX leaders find that harder to answer than they expect. That is usually where a causal diagnostic earns its place.
