NPS Benchmarks Need Comparable Respondents
A customer experience team presents a Net Promoter Score of 42. Someone asks whether 42 is good for the industry, and a search turns up a benchmark of 55. The difference looks like a clear performance gap. But if the team surveyed customers just after a support call while the published figure came from a market survey about whole brands, the two numbers answer different questions. The first describes a particular interaction among people who had it; the second describes brands among the people surveyed in a market. Subtracting one from the other cannot tell the team how far it trails competitors.
That distinction is the useful starting point for NPS benchmarks. A benchmark can help you decide where your company stands or whether a change has improved an experience. It becomes misleading when the respondents, question, or survey conditions differ from yours. My rule is to name the decision first, then choose the comparison that can answer it. A convenient industry average is no substitute for that match.
An NPS number describes a group of respondents
Net Promoter Score begins with a recommendation rating from 0 to 10. Ratings of 9 or 10 are classed as promoters, 7 or 8 as passives, and 0 through 6 as detractors. NPS is the percentage of promoters minus the percentage of detractors among the respondents; passives remain in the denominator but contribute to neither side of the subtraction. Bain’s explanation of the calculation gives the same categories and formula.
Consider an illustrative survey with 100 completed responses: 50 promoters, 30 passives, and 20 detractors. The NPS is 50% minus 20%, or +30. The number is a difference in percentage points, not a percentage of people who are loyal, and it does not describe anyone who was never asked or did not respond. If a report gives only “NPS +30,” ask for the respondent group, the survey period, and the distribution of all three rating groups before drawing a conclusion.
The distribution matters even when two surveys yield the same NPS. In another illustrative set of 100 responses, 40 promoters, 50 passives, and 10 detractors also produce +30. The first set has more enthusiastic respondents and more dissatisfied ones. The second has fewer of both. A manager deciding whether to improve a specific touchpoint needs more than the net difference to see where attention belongs. The score summarizes a pattern; it does not explain the reason for it.
This is also why a change in a small group deserves care. In an illustrative survey of 20 respondents, one person’s rating moving from detractor to promoter raises the NPS by 10 points: detractors fall by five percentage points and promoters rise by five. That arithmetic does not mean a real improvement of the same size has occurred across all customers. It means the reported number can move substantially when the group is small. Before calling a five or ten point change a win or a setback, look at how many people responded and whether they represent the same population in each period.
The survey type determines the comparison
There are three useful settings to distinguish. Bain describes competitive benchmark, relationship, and experience NPS as different kinds of feedback. Qualtrics describes relational and transactional NPS in similar terms: the former concerns the overall relationship, while the latter follows a specific interaction. The terms “experience” and “transactional” point to a triggered moment; they do not turn its result into an overall market measure.
A relationship survey asks customers to assess the broader company relationship without making one recent event the reason for asking. It is the sensible starting point if you want to follow how your own customers’ overall view changes. A support-call survey, by contrast, reaches people after a support call and puts that episode close to the question. A purchase or installation survey does something similar for a different touchpoint. These scores may help the teams responsible for those experiences, but a support-call score should not be treated as the company’s relationship score. Qualtrics distinguishes the overall view from feedback after a particular interaction.
A competitive benchmark has a different purpose again. In Bain’s account of double-blind competitive NPS, a third party surveys a market sample so brands can be compared with other brands in that market. This can reach beyond the customers a company would invite through its own channels. If the decision is whether the brand is gaining ground against alternatives, I would seek a competitive study built around those alternatives. An internal customer survey cannot by itself answer what people who chose a competitor would say.
The practical mistake is to line up three numbers because they all carry the label NPS. Imagine, purely as an illustration, a relationship score of +40 from current customers, a +15 score after support contacts, and a market score of +25 from a separate competitive study. The three figures do not form a league table. The +15 may signal that the support experience merits attention, but its respondents were selected by having a support contact. The +40 reflects a broader relationship among the company’s respondents. The +25 reflects the market sample and brands in that study. Each has value when used for the decision it was designed to inform.
Bain explicitly cautions that relationship and experience scores should not be compared directly, and its benchmark discussion warns against comparing internally collected customer scores with outside competitive figures. The judgment to make is therefore simple: use market research for relative market standing, a stable relationship survey for changes in customer sentiment, and a matched interaction survey for changes at that interaction. Mixing them can make ordinary differences in who was asked look like differences in performance.
A fair benchmark keeps the people and question comparable
Once the survey type matches your decision, examine who answered. “Our industry” may still contain unlike audiences. A survey of existing customers does not sample the same people as a broader competitive study. Within an industry, a company serving one kind of customer or market may face different expectations from a company serving another. SurveyMonkey GetFeedback notes that audience, industry, and culture can affect scores. An industry label helps narrow the search; it does not settle whether the reference group fits your business.
Next, read the actual question and its placement. An overall recommendation question presented in a relationship survey is different in context from one sent immediately after a transaction, even if the rating scale is identical. Wording and presentation can also affect the responses; SurveyMonkey flags differences in how questions are presented. If a benchmark provider does not describe what respondents saw, I would keep the number as background context rather than use it as a target for a team.
The collection method belongs in the comparison too. A company asking only people it can contact through its own customer list has a different route to respondents from a third party recruiting a market sample. Bain separates those approaches because the resulting groups can differ. This is especially important when a published score is called an “industry benchmark” without saying whether it came from competitive market research, a pool of participating companies’ customer surveys, or a particular touchpoint. Before using the number, find out which it is.
Time matters for the same reason. A benchmark is a result for respondents surveyed in a defined period, not a permanent property of an industry. If your own survey covers a different period, you cannot know from the two NPS figures alone whether the difference comes from customer views or from the changed conditions under which people answered. Put the dates beside the scores, and prefer comparisons that cover similar periods. Bain’s account of benchmark work emphasizes like-for-like comparisons among customers considering real alternatives.
Finally, ask for the size and makeup of the response group. The illustrative 20-person calculation shows why a small denominator can make a point difference look more decisive than it is. A large response count does not fix a mismatch in audience or question, but a missing count makes the number hard to interpret. For a published benchmark, I would want at least the population, survey type, period, sample size, question context, and collection method. If those details are unavailable, the honest reading is “rough reference point,” not “we are 13 points behind.”
Choose the benchmark that fits the decision
For a strategic question such as “How do customers see us beside the brands they could choose instead?”, I would favor a competitive, double-blind study in the relevant market. The useful comparison is your score and competitors’ scores gathered under the same study conditions. The cost is having to obtain research that covers the right alternatives and respondents; an easy public figure may not do the job. This is the setting where a relative gap has a clear meaning for market position. Bain describes double-blind competitive comparison as the way to understand relative standing.
For the question “Are our customers’ overall views improving?”, I would first compare relationship surveys from comparable customer groups and periods using the same question and collection approach. This is a practical internal benchmark: last period’s result helps interpret this period’s result. It does not require claiming that the number equals the market’s view. If the customer mix or survey timing changed, put that change next to the trend rather than announcing a clean gain or loss. A stable method makes movement easier to read; changing the method may still be necessary, but it creates a break in the comparison.
For “Did the new support process improve the support experience?”, I would compare responses after that same type of interaction. An overall relationship benchmark is too broad to isolate a support call. The interaction score can point the team toward a problem or improvement near the touchpoint; it should be read with the actual comments and circumstances of the interaction. Qualtrics explains that transactional feedback addresses specific touchpoints. If only the support-call score moved, I would investigate support before describing the whole brand as having changed.
One organization may therefore need several baselines, with a different use for each. That is not an invitation to average them. Combining a relationship score, a support score, and a competitive score would erase the very differences in respondents and questions that make each useful. Label each result with its survey type and population wherever it appears. This small piece of context prevents a dashboard or presentation from turning distinct answers into a false single verdict.
“Good” depends on the comparison you can defend
There is no sound universal cutoff that turns an NPS into “good” for every company. A positive score says that promoters outnumber detractors among the people who answered; it does not establish that the company leads its market. A negative score says the reverse for those respondents; it does not, on its own, tell you whether rivals fare better or worse. The arithmetic is clear, while the business judgment depends on who was surveyed and what choice the number is meant to guide.
For example, consider an illustrative competitive study that asks the same kind of market respondents about three brands under matched conditions. Suppose the scores are +50, +55, and +75. The brand at +50 has many more promoters than detractors, yet it ranks behind the two alternatives in that study. Now imagine a separate market where the same +50 leads the relevant alternatives. The absolute number has not changed; the competitive meaning has. Bain uses this kind of within-market comparison to show why an attractive score can mislead when viewed without peers.
Published industry averages can still orient a first conversation, provided their limits stay visible. SurveyMonkey GetFeedback calls its benchmarking data a high-level reference rather than a scientific resource. That is an appropriate way to treat a number whose underlying respondents or question may not match yours. It might justify a closer look, but I would not attach a budget, performance rating, or public claim to the gap until I knew what the two surveys actually compared.
If the benchmark details are thin, the next useful step is not to search for a more flattering threshold. Preserve the number with its source and limitations, then establish a comparison suited to the decision: a repeatable relationship measure, a matched interaction measure, or a proper competitive study. The preferred path takes more care than copying an industry average, but it gives the score a job it can perform. An NPS benchmark is useful when it helps you interpret a defined group of people’s answers against a genuinely comparable group. Without that match, precision in the number only hides uncertainty in the comparison.