Net Promoter Score

By Gojko Adzic · Published 3 July 2026

The Net Promoter Score® (NPS®) is a popular but disputed and widely misused customer satisfaction index.

This article explains the key aspects of running NPS research, its limitations and applicability, for teams that want to use this method in practice.

How to calculate the NPS Score?

To calculate the NPS score, ask customers to rate on the scale of 0 to 10 how likely they would be to recommend a product or service to their friends or colleagues (with zero typically labelled “not at all likely”, five labelled “neutral” and ten labelled “extremely likely”).

Net Promoter Scale Survey

1. How likely is it that you would recommend [company or product] to a friend or colleague?

0 1 2 3 4 5 6 7 8 9 10
Not at all likely Extremely likely
The NPS survey uses an 11-point Likert scale to capture likelihood to recommend, as a measurement of loyalty and a predictor of growth.

The responses are then grouped into 3 buckets:

NPS Score Categories
NPS responses are grouped into promoters and detractors, and the difference in the percentage between the two groups is the NPS score.

The Net Promoter score is the percentage of promoters minus the percentage of detractors. It is usually presented either on the -1 to 1 scale, or -100 to 100.

Free NPS Score Calculator
Calculate NPS Score

Use our free NPS calculator to instantly calculate the NPS Score with an evaluation of confidence and reliability, and an interpretation of how much you can trust it. No sign-up required.

Calculate NPS

Variants and frequency

Bain & Co differentiate between three types of NPS scores:

In addition, Bain & Co suggest a variant called Employee NPS (eNPS) to measure employee engagement, with the following question:

Notably, out of the different variants, only the competitive benchmark was proven as a good predictive metric for growth, and only when applied to the wider market (see the section on Applicability).

Alternative questions

In contexts where asking about recommendation is not relevant, use one of the following two questions instead:

Interpreting NPS results

NPS is an example of Likert Scale research, with “top box minus bottom box” scoring. In Quantifying the User Experience, Sauro and Lewis warn that this kind of scoring loses information because it collapses a larger scale into fewer categories, so large sample sizes are required to make up for the lack of sensitivity in the metric. They also warn that there is no well defined method for computing confidence intervals around the NPS.

Brendan Rocks evaluated several statistical interval estimation techniques for NPS scores, concluding that “Variations on the Adjusted Wald, and an iterative Score test are found to have superior performance”.

Our Free NPS calculator tool automatically applies the two recommended interval calculation methods and provides an estimate for reliability of the score.

For additional information on statistical evaluation methods for NPS, check out Statistical validation of critical aspects of the Net Promoter Score and The distribution of Net Promoter Score in socio-economic surveys.

Applicability and avoiding misuse

There are significant benefits from improving customer experiences, and plenty of real-world examples where loyal customers contribute to product success, but there is no proof that NPS is a good singular measure that can inform business strategy and decisions, particularly when used on a transactional (experience) level.

NPS is widely popular in the industry, mostly due to simplicity and the seductive nature of a single number being used as a prediction mechanism for growth and success. This is despite the fact that academic replication attempts mostly failed (see Arguments against NPS). Beyond academic replication attempts, there are several key issues that affect the accuracy of the data collected with NPS surveys (see Problems with NPS). In practice, many of those issues significantly reduce the accuracy and the validity of the results.

Sven Baehre, Michele O’Dwyer, Lisa O’Malley and Nick Lee analyzed data from seven US sportswear brands over five years, concluding that the Competitive benchmark type of NPS scoring, under the condition that it involves not just current but also potential customers, is effective as a predictive metric for future growth. They could not prove the effectiveness of NPS in any other context.

NPS measured among current customers only did not predict growth, and neither did static NPS levels. Only quarter-to-quarter changes in NPS predicted the following quarter’s sales. Even then, the effect was modest. NPS “can explain only a fraction of future sales growth by itself”, improving the model fit by just .028, and a question about brand consideration predicted growth equally well.

… the methodological concerns raised by academics are valid, and only the more recently developed brand health measure of NPS (using an all potential customer sample) is effective at predicting future sales growth

– Baehre, O’Dwyer, O’Malley and Lee in The use of Net Promoter Score (NPS) to predict sales growth: insights from an empirical investigation

Evert de Haan, Peter C. Verhoef and Thorsten Wiesel claim that customer feedback metrics can predict retention (they make no claim about the link between retention and growth). For NPS itself, de Haan and colleagues argue that it’s useful for predicting the most risky customers (top churners) and for customer management, but that other metrics (such as the basic top-2-box) are better for competitive positioning.

Based on the research data available at the time when this article was last updated, NPS could be used as:

NPS is not accurate or reliable enough to be used as a singular measure of customer experience, and treating it as a competitive benchmark between employees or branches is not reasonable. In particular, companies should avoid linking employee compensation to NPS scores.

Origins of NPS

Net Promoter Score was invented by Frederick F. Reichheld, while working at Bain & Company, around 2001. It became widely popular after Reichheld introduced it in the The One Number You Need to Grow paper in 2003.

The original idea came from the assumption that companies waste money on complex satisfaction surveys, and that asking a single question is a good enough predictive signal for growth. Since asking a single question is much simpler and faster than complex surveys, companies would be able to use it for fast feedback and as a relative measurement over time, and across different business units.

As the inspiration for the NPS, Reichheld quotes customer research ideas of Andy Taylor, the CEO of Enterprise Rent-A-Car, who decided to poll customers each month with just two simple questions. The first asked about the recent rental experience, and the other about the likelihood that the customer would rent from the Enterprise again. Asking just two questions enabled Enterprise Rent-A-Car to quickly collect relative comparison metrics for its five thousand branches in the United States, and provide almost real-time feedback on how individual offices or even employees were doing, “and the opportunity to learn from successful peers”. Another important aspect of Enterprise Rent-A-Car surveys was that they mostly focused on measuring the number of “enthusiastic” customers.

By concentrating solely on those most enthusiastic about their rental experience, the company could focus on a key driver of profitable growth: customers who not only return to rent again but also recommend Enterprise to their friends.

– Frederick F. Reichheld, The One Number You Need to Grow

Thomas O. Jones and W. Earl Sasser Jr. explored similar ideas in 1995, suggesting that “it is easier for a customer to respond honestly to a question” about whether they would recommend a product than about whether they would buy it again, that the relationship between satisfaction and loyalty “was neither linear nor simple”, and that it breaks down entirely in monopolies, where Jones and Sasser found that “customers remained loyal no matter how dissatisfied they were.” Providing some research proof for the idea to focus on the most enthusiastic customers, Jones and Sasser quote earlier work by John Larson who “found that completely satisfied customers were nearly 42% more likely to be loyal than merely satisfied customers.”

The theory behind NPS

Reichheld introduced NPS as a “way to measure and manage customer loyalty without the complexity of traditional customer surveys.”

The method builds on Reichheld’s earlier work on managing customer loyalty, particularly The Loyalty Effect and the 1990 article Zero Defections: Quality Comes to Services, where Reichheld and W. Earl Sasser claim that “companies can boost profits by almost 100% by retaining just 5% more of their customers” (the article it shows no source or reference data for this claim).

Defining loyalty as the “willingness of someone […] to make an investment or personal sacrifice in order to strengthen a relationship”, Reichheld suggests that loyal customers will stay with a supplier who treats them well and provides good value in the long term, even if for a particular transaction the supplier does not offer the best price or terms relative to competitors. Loyalty “drives top-line growth”, due to repeated purchases, but according to Reichheld it also reduces customer acquisition costs because “loyal customers talk up a company to their friends, family, and colleagues”. Since customers recommending the product to others “risk their reputations”, they will do that only “if they feel intense loyalty”.

Reichheld suggested measuring loyalty through likelihood to recommend more than repeated purchases because some products naturally have a long purchase cycle, so customers would not repeatedly purchase them frequently enough to make frequent surveys relevant.

Following that thinking, intense loyalty drives growth, and measuring intense loyalty can help to predict growth. NPS, in theory, helps companies measure loyalty quickly and simply, in a way that can be used to benchmark and compare how customers experience a product over time, or in different environments. A simple, one question survey, is likely to result in much higher completion rates than a complex survey, providing a more complete picture. A single number makes it easily relatable, and consumable by executives.

Unlike the Enterprise Rent-A-Car research that mostly focused on tracking the enthusiastic customers, Reichheld’s research also points to the importance of tracking disappointed customers (“detractors”). The negative effects of detractors damaging the reputation of a company in their conversations with colleagues and friends offset the positive effects of enthusiastic customers promoting it, so both numbers need to be tracked in combination. Reichheld then claims that the “Net” difference between both groups is a good predictor of growth.

Eugene W. Anderson and Vikas Mittal argue in Strengthening the Satisfaction-Profit Chain that a 1% increase in satisfaction is associated with a 2.37% rise in return on investment, while a 1% decrease is associated with a 5.08% fall, providing some research proof for tracking enthusiastic and disappointed customers separately. Their numbers come from the Swedish Customer Satisfaction Barometer data covering 125 companies, but with no controls or significance tests reported, so the numbers should not be trusted blindly.

The general idea that losses weigh more than gains has been tested more formally for satisfaction itself. In The Antecedents and Consequences of Customer Satisfaction for Firms, Eugene W. Anderson and Mary W. Sullivan analysed survey responses from 22,300 customers of 57 Swedish companies, and found that falling short of expectations lowered satisfaction more than exceeding them raised it. The difference was statistically significant, and held in 83 of 114 estimates. The research looked at satisfaction, not profit, so it supports the idea to track disappointed customers separately, but it does not confirm the ROI numbers above.

Arguments for NPS

The main advantage of NPS over alternative measures is simplicity, both in terms of collecting the data and interpreting the results. Setting up and organizing NPS data collection is easy, asking a single question leads to higher response rates than with more complex surveys, and the score results are easy to calculate and understand.

The NPS index, in spite of considerable criticism in the scientific community, turns out to be a tool that is easy to implement even by those without specific statistical knowledge. For business operators, the evaluation of the number of potentially satisfied customers (promoters) is easy.

– Cazzaro and Chiodini, Statistical validation of critical aspects of the Net Promoter Score

Reichheld’s original paper quotes research by Satmetrix, who tracked likelihood to recommend scores from 400 companies in more than a dozen industries (notably, the paper does not explicitly claim that Satmetrix tracked NPS scores, but “would recommend” scores), and imply that “a strong correlation existed between net-promoter figures and a company’s average growth rate over the three-year period” for some classes of researched companies, such as airlines, where revenue data was readily available.

Evert de Haan and colleagues point to a link between NPS and customer retention, more directly as a prediction which customers are most likely to stay or churn. The standard NPS calculation correlated .170, very close to the best-performing metric (top-2-box customer satisfaction score at .184). NPS produced the best top-decile lift of any metric tested, at 2.241, which makes it a reasonable choice for identifying the customers most likely to stay. (Note that they did not research the original NPS premise, that the scores would be related to business growth).

Arguments against NPS

With a surge in popularity of the method, other researchers have tried to replicate the claims from the original Satmetrix research or prove the causal relationship between NPS and growth. Several such research efforts ended up with conclusions that the original claims do not stand up to statistical scrutiny.

In A Longitudinal Examination of Net Promoter and Firm Revenue Growth, Timothy Keiningham, Bruce Cooil, Tor Wallin Andreassen and Lerzan Aksoy compared NPS scores to large population data available from the Norwegian Customer Satisfaction Barometer and the American Customer Satisfaction Index, disputing that NPS is a better prediction of growth than other measures.

In a follow-up paper A Holistic Examination of Net Promoter, the same authors reported results of a two year research involving more than 8,000 customers of companies in three industries, failing to prove the key assumptions behind NPS. Their data suggests that likelihood to recommend is not an effective leading indicator of loyalty.

The claim that recommend intention is an effective predictor of loyalty behaviours — beyond other metrics — is not supported.

– Keiningham, Cooil, Andreassen and Aksoy, A Longitudinal Examination of Net Promoter and Firm Revenue Growth

Additionally, the paper claims that there is a huge variance in the link between the intention to recommend and growth across different industries, doubting correlation precision and accuracy.

Keiningham and colleagues dispute the claims in the original Satmetrix research, arguing that the original sample sizes per industry were too small to be reliable. The results were not readily repeatable in the same industries by new research, and the conclusions may not be generalizable across industries. Finally, Keiningham and colleagues claim that the single metric model significantly underperforms dual-metric and multi-metric models when predicting growth, so using a single score might be an oversimplification. They measured “an average improvement of nearly 20 per cent when moving from a single-metric model to the best multi-metric one”.

Evert de Haan and colleagues compared several common metrics, including NPS, and concluded that the metric that best predicts retention depends significantly on the industry and “unit of analysis” (for example, focusing on individual customers or comparing companies against a benchmark). One of their key conclusions is that “combining CFMs [customer feedback metrics], along with simultaneously investigating multiple dimensions of the customer relationship, improves predictions even further”, directly disputing the idea that NPS is good as a singular measure. Their conclusions come from analysing the Dutch Customer Performance Index data from a study with 6,649 people providing 8,924 evaluations of 93 Dutch consumer firms across 18 industries between September 2010 and November 2012.

Neil A. Morgan and Lopo Leotte Rego tested six satisfaction and loyalty metrics against six dimensions of business performance, using the American Customer Satisfaction Index covering 569 observations of 80 firms over 7 years. Their conclusion was that metrics based on recommendation intentions (or even recommendations) have “little or no predictive value”, and that “net promoters metric has no significant relationship with future business performance at all”.

Our results clearly indicate that recent prescriptions to focus customer feedback systems and metrics solely on customers’ recommendation intentions and behaviors are misguided

– Neil A. Morgan and Lopo Leotte Rego, The Value of Different Customer Satisfaction and Loyalty Metrics in Predicting Business Performance

Notably, Morgan and Rego did not use NPS data according to Reichheld’s definition, but statistics about complaints and recommendations from the ACSI, so their argument should be viewed more in the context of the link between likelyhood to recommend and business performance more generally, not directly about NPS specifically.

Baehre and colleagues compared seven ways of scoring the same 0–10 answers, and how well they predict next-quarter sales growth. The standard NPS calculation came third. A simpler formula, just counting the share of people answering 9 or 10, did slightly better. Separately counting detractors added no value. The authors still conclude that NPS “offers reliable insights, is familiar, and is a “good enough” number to grow”, with the warning that their results come from a single industry.

Problems with NPS

There are several structural problems with using NPS score as a singular predicting and decision-making factor, severely limiting the relevance of that metric without additional research.

Loyalty is a leading indicator, not a measure of success

The original NPS paper suggests that NPS score is a leading indicator of customer loyalty, and that customer loyalty is one of the key leading indicators of growth.

As a leading indicator, the score is a quick and simple tool to point the direction, not a good measurement of ultimate success.

In practice, naive use of the metric results in business leaders treating it as a success or outcome metric. Safdar and Pacheco in the Wall Street Journal article The Dubious Management Fad Sweeping Corporate America suggest that many large companies, including “American Express Co., Best Buy Co. and Citigroup Inc.”, listed the metric as a criterion for executive compensation.

Because of readily available benchmarks, NPS became a way for comparing different companies, and a reporting metric for company success. According to the Safdar and Pacheco, “NPS was cited more than 150 times in earnings conference calls by S&P 500 companies” in 2019. Unlike accounting numbers that need to be verifiable, NPS is self-reported and not audited, with no way of actually gauging the correctness of the numbers the executives provided to investors. Unsurprisingly, the tracked metrics typically always improve:

Out of all the mentions the Journal tracked on earnings calls, no executive has ever said the score declined

Khadeeja Safdar and Inti Pacheco, The Dubious Management Fad Sweeping Corporate America

Using a leading indicator as a measurement of success, especially as NPS is a leading indicator of a leading indicator, is unjustifiable.

Recommendations are not always relevant

The relevance of NPS scores as a predictor of growth heavily depends on customers actually being able to sensibly recommend a product or service, and then measuring loyalty that way. Markets with low likelihood to recommend are not a good fit for NPS. In particular, even the original paper recognises that NPS scores do not predict growth in markets that are “monopolies and near monopolies, where consumers have little choice”.

Separately, there is a problem with the standard NPS question for product types that people don’t casually recommend to friends or colleagues. A typical example is a software system employees are forced to use as part of their job, and do not have a choice in selecting an alternative. Reichheld recognized this issue in the original paper, and recommended using alternative framing around loyalty:

Asking users of the system whether they would recommend the system to a friend or colleague seemed a little abstract, as they had no choice in the matter. In these cases, we found that the “sets the standard of excellence” or “deserves your loyalty” questions were more predictive.

– Frederick F. Reichheld, The One Number You Need to Grow

Classification buckets are not mutually exclusive

The basic assumption of NPS is that the categories of promoters and detractors are mutually exclusive, but there is no proof of this in the original papers, and there is data to show the contrary. In Where Net Promoter Score Goes Wrong, Christina Stahlkopf suggests that the likelihood to recommend a brand is highly contextual, and reports on a survey where “52% of all people who actively discouraged others from using a brand had also actively recommended it”. As an example, Stahlkopf provides an analysis of an online music streaming service where people would actively recommend the service to their friends because of customizability, but discouraged their parents from using the same service because of cost and complexity.

Small sample sizes affect statistical relevance

As a Likert Scale, NPS scores can end up heavily biased when sample sizes are small, yet the original papers suggest that data is valid even for single interactions.

Happy customers are more likely to answer

De Haan and colleagues also found that “people who provided a higher NPS are however significantly more likely also to take part in the follow-up survey”, which becomes important for surveys tracking customers over a longer period, as the data might end up being skewed towards positive responses.

Judgements are noisy

The NPS survey asks people to make a judgement about a theoretical topic, and such surveys are known to be significantly affected by “Occasion Noise”, driven by irrelevant factors such as mood or even the weather (documented by Kahneman in Noise).

Comparing NPS across groups is misleading

In The distribution of Net Promoter Score in socio-economic surveys, Capecchi and Piccolo argue that variations in how confidently and consistently people from different groups interpret and answer the same question (Respondent heterogeneity) introduces a systematic, directional bias into NPS scores that makes cross-group comparisons unreliable.

Capecchi and Piccolo observed NPS is always lower than the actual underlying customer sentiment, with the gap growing proportionally to the level of uncertainty and heterogeneity of the group of customers involved in the survey. The same underlying strong positive sentiment yields an NPS of 0.50 when respondents are relatively homogeneous, but drops to 0.25 as respondent heterogeneity increases. That is not because sentiment changed, but because more heterogeneous responses mathematically dilute the score.

This means that a customer segment with more diverse or uncertain opinions will always score lower than a homogeneous one, even if the underlying loyalty and satisfaction are identical. Because of that, comparing NPS scores across business units, market segments, time periods, or competing companies is unreliable without controlling for differences in respondent homogeneity.

Intent is not a reliable indicator

NPS tracks intent, which is unreliable. In How Valuable Is Word of Mouth?, authors Kumar, Petersen and Leone suggest that only about one third of the people expressing intent to recommend actually follow through.

The gap between stated intent and later behaviour was already documented in the loyalty literature NPS grew out of. Christina Stahlkopf conducted a research checking if customers actually promoted or told others to avoid a brand, comparing it with NPS scores from the research done by the client brand, and reported wide disagreement between the NPS data and actual consumer actions.

According to their NPS ratings, 50% of customers in our first survey were promoters, but 69% of customers had actually recommended a brand. So the NPS categorization missed a big chunk of the actual promoters… According to their NPS ratings, 16% of consumers in the survey were detractors, yet only 4% of all the respondents had actually told others to avoid a brand. In fact, we found that detractors were seven times more likely to have either recommended a brand or said nothing at all than to have disparaged it.

Christina Stahlkopf, Where Net Promoter Score Goes Wrong

Jones and Sasser argue that the outcome can be predicted from intent by factoring in overestimation, writing that although the intent will “generally overstate the probability of repurchase, the degree of exaggeration usually is fairly consistent”. They provide an example of “60% to 80% of automobile customers queried 90 days after buying a car say they intend to repurchase the same brand, and 35% to 40% actually do so three to four years later”. They do not claim that this rough 50% factor is universal, so if you plan to use intent as an indicator of action, it’s best to establish the overestimation factor yourself with your own customers.

The attrition is roughly the same order as the figure Kumar and colleagues measured for recommendation intent twelve years later, which suggests the problem is a property of asking about future intentions rather than anything specific to the recommend question.

NPS is easy to game, and people are incentivised to game it

NPS was, according to Net Promoter 3.0, used “by two-thirds of the Fortune 1000” in 2011. Still, the relevance of the NPS score as a predictor of growth was never confirmed by peer-reviewed research.

Reichheld and co-authors of Net Promoter 3.0 claim that one of the most common types of abuse by inexperienced practitioners is to link NPS scores to bonuses, after which employees “care more about their scores than about learning to better serve customers”.

The most common and most damaging misuse of Net Promoter Score is linking it to front-line employee compensation.

Fred Reichheld, from the Account Experience Podcast

Due to simplicity and ease of measurement, NPS scores are also relatively easy to falsify and game. For example, because NPS is an aggregate score, a common way of gaming NPS is to select only customers with good experiences for the survey. The Wall Street Journal article suggests employees of Best Buy, one of the companies quoted as actively using NPS scores to judge employee performance, were actively exchanging ideas online on how to inflate their own NPS scores.

Potential improvements

Some of the major complaints about NPS can be addressed by introducing additional metrics or questions. However, such additions diminish one of the main attractive aspects of NPS, simplicity.

Addressing the intent unreliability and misclassification

Stahlkopf recommends asking about customers’ actual past behavior, instead of intent, and calculating an Earned Advocacy Score “by taking the percentage of active advocates and subtracting the percentage of active discouragers”. This requires two separate yes/no questions instead of a single Likert scale, but according to Where Net Promoter Score Goes Wrong provides “clearer, more-detailed, and more-actionable data”.

Addressing the self-reporting issues

With a recognition that companies want to link customer loyalty metrics to executive and front-line compensation, Reichheld started recommending in 2021 complementary “earned growth” metrics, as an accounting-based, auditable counterpart to NPS. The earned-growth metric “reinforces the effectiveness of NPS” and provides a “clear, data-driven connection between customer success… and business results”. Reichheld develops the same argument at book length, with Darci Darnell and Maureen Burns, in Winning on Purpose.

Reichheld and colleagues suggest tracking two earned growth metrics:

The proposed calculation for the Earned Growth Rate includes the following steps:

Note that there is no peer-reviewed research that supports the validity of earned growth metrics or earned advocacy score.

NPS Licensing and Trademarks

While Reichheld claims he released the method as “open source” (from an interview with Adam Dorrell on the Account Experience Podcast), the names NPS®, Net Promoter Score®, and Net Promoter System® are registered trademarks, owned by NICE systems in the US. “Net Promoter” was also a registered trademark until June 2026, but it then expired.

The method itself is free to use (without any fee or permission); but using the trademarked names requires attribution even for non-commercial projects, and commercial use of the registered trademarks requires a license, according to Net Promoter System Trademarks and Licensing Page.

An umbrella term for similar research systems is Likelihood to recommend, often used to avoid the registered NPS trademark.

Try the free NPS calculator

Calculate your score from bucket counts or raw 0–10 responses, and get 95% and 99% confidence intervals computed with benchmarked statistical methods — not the naive formula most online calculators use. No sign-up — your data stays in your browser.

Use the free tool

Get notified about new articles & major updates
Share
Suggest an edit to this article Request a new topic