Why a rising NPS does not prove customer experience improved

A rising NPS should prompt investigation

A rising Net Promoter Score, or NPS, should prompt a business to ask what changed before declaring that customer experience improved. Customers may be happier. But the increase may also reflect a different group of respondents or survey practices that favor higher ratings, so leaders using NPS to assess performance need to ask whether the measurement process can reveal evidence that challenges the result.

Understanding that process starts with the question NPS asks: “How likely are you to recommend this company, product or service to a friend or colleague?” Customers answer from 0 to 10. Ratings of 9 or 10 are promoters, 7 or 8 are passives, and 6 or lower are detractors. Subtracting the detractor percentage from the promoter percentage produces NPS; passives remain in the respondent population but have no separate term in the subtraction.

The calculation produces a score from −100 to 100. A positive result means promoters outnumber detractors, zero means their shares are equal, and a negative result means detractors outnumber promoters. Fred Reichheld of Bain & Company introduced NPS in the December 2003 Harvard Business Review contribution “The One Number You Need to Grow.” Bain is a consultancy with a commercial interest in advising businesses on growth and customer loyalty, so its role matters when considering the measure’s appeal.

The appeal comes partly from reducing a complex relationship to one question. Howard Lax, president of The LAX Group and director of experience management strategy at VistaXM, Inc., describes it this way: “The NPS provides the promise that all you need to do is ask one question to gauge insight into brand loyalty.” His commercial roles give him a stake in how businesses invest in customer experience management. After collecting ratings and calculating NPS, a business still needs customer explanations and separate records of referrals, renewals and spending to understand what the score means.

An improving score can still leave unsuccessful customers unheard

Before interpreting those ratings, examine who had a chance to provide them. A business that invites feedback only after a completed purchase selects people who finished the process successfully, leaving customers who abandoned it or could not proceed unable to describe their experience through the same measurement. A support survey sent only after resolution similarly excludes unresolved experiences.

These eligibility rules can shape the result even when every invited customer receives a neutral question. A higher score describes the selected experiences, while customers who never qualified remain outside the measurement. Changing the rules can also change the respondent population over time, so leaders need to inspect invitation criteria alongside the trend they are being asked to interpret.

Even eligible customers may remain unheard because receiving an invitation and answering it are separate events. A customer may receive the invitation, view the survey and decline to respond. Their experience may differ from those represented in completed ratings. Expanding eligibility gives customers a chance to be heard, while understanding participation requires examining what happens after the invitation.

Qualtrics’ 2026 Consumer Experience Trends research illustrates this broader risk of missing customer experiences. Qualtrics sells experience-management technology and benefits when businesses invest in collecting and analyzing feedback. Among 20,000 consumers across 14 countries, 30% said they tell no one after a bad experience, while 29% send feedback directly to the company. Direct feedback was down 7.5 points since 2021, and the share remaining silent reached the survey’s highest level; these findings concern consumer feedback broadly and cannot quantify NPS nonresponse.

Because customers may remain silent, a customer experience, or CX, team needs ways to find problems beyond survey comments. Cancellations, abandoned purchases and unresolved support cases can identify difficulties customers encounter while trying to do business. Comparing those records with survey findings helps reveal whose experiences may be missing. A lack of negative comments offers little reassurance when the customers facing difficulty may be absent.

Participation benchmarks also need careful interpretation. Survicate, a survey software provider that benefits from spending on survey tools, reported a median response rate of 9.98% in an analysis of 4,332 surveys from 460 companies. It counted survey starts or welcome-screen clicks against survey views. That denominator measures a different stage of participation from completed recommendation ratings divided by invitations sent.

The scope of Survicate’s analysis further limits the comparison. It covered several survey types and excluded link surveys distributed through channels such as email, so its median cannot serve as a general NPS completion benchmark. Nor can it be combined with Qualtrics’ consumer findings to calculate a missing-respondent share, since the datasets describe different behaviors and populations.

To make participation rates clear, follow customers through each collection stage. Record invitations, survey views, starts and completed recommendation ratings separately, then state the numerator and denominator for every reported rate. Before comparing teams or periods, identify the survey types and delivery channels included or excluded. Otherwise, similar percentages can describe different levels of participation.

Better reporting makes missing experiences visible, but businesses also need to invite them deliberately. Shelley Parnell, Dogtopia’s chief experience officer, describes including families whose dogs attend alongside parents whose dogs do not pass its introductory Meet & Greet process: “The biggest mistake is only listening to happy customers. Dogtopia intentionally seeks feedback from parents whose dogs don’t pass the Meet & Greet. Those are some of our most difficult conversations, and some of the most valuable.” The practice gives Dogtopia access to unsuccessful introductions and a way to examine how it handles them.

Inclusion can still fail if staff actively protect the score. Staff may pressure customers to give a 9 or 10, or exclude dissatisfied customers from invitations. Either practice can raise reported NPS while hiding problems management needs to address. Compensation links and rigid targets can encourage this behavior because teams have a reason to influence collection and defend favorable results.

Pressure can also affect how customers answer once a survey reaches them. CMSWire contributor Martha Brooke examines these mechanisms in “Your NPS Strategy Is Becoming ‘Feedback Theater’,” published March 6, 2026. She identifies social desirability bias, in which people give answers they believe others want, along with leading questions and non-anonymous surveys as sources of inflated ratings. Her recommendations include systematic comment tagging, correlations that measure statistical relationships, and crosstabs that compare responses across groups.

These analyses can guide investigations into loyalty and churn, the loss of customers, but statistical relationships alone do not establish causes. They also cannot repair every collection problem: broader eligibility cannot remove pressure to give a high rating, and neutral wording cannot recover excluded experiences. Once a business has examined inclusion, participation and efforts to protect the score, it can turn to what the completed ratings reveal.

The same NPS can describe very different customer populations

Even well-collected ratings lose information when reduced to a single subtraction. Consider two hypothetical surveys, each with 100 respondents and an NPS of 20. Example A has 60% promoters, 0% passives and 40% detractors, producing 60 − 40 = 20. Example B has 30% promoters, 60% passives and 10% detractors, producing 30 − 10 = 20.

The shared result hides a substantial difference between the populations. Example A combines a large promoter group with a large detractor group, while Example B has fewer customers at either end and a majority of passives. A manager seeing only the headline would miss the difference between 40% detractors and 10%. That difference should shape which experiences the business investigates and how it interprets the customer relationship.

To preserve that information, report NPS alongside promoter, passive and detractor shares. Add response counts to show how many people the result represents, then use comments to explain the experiences behind the ratings. Relevant segment findings can reveal whether different customer groups encounter different problems. The headline remains easy to communicate, while readers have enough detail to question it.

For example, a hypothetical subscription business could compare new subscribers with long-term customers to investigate concerns by tenure. Examining each group’s comments alongside its ratings could reveal an onboarding problem among newer customers or a product concern among established ones. Combining the groups too early could hide either pattern, so segmentation helps identify where an investigation should begin.

The same grouping choices matter when interpreting change over time. A different balance of customer groups can move the aggregate score even when experiences within those groups remain similar. Before treating a small movement as meaningful, examine response counts and changes in the surveyed population. Reporting the distribution improves interpretation, but separate behavioral evidence is needed to establish what respondents subsequently did.

Behavioral evidence must be allowed to challenge a score recovery

Even a carefully sampled and fully reported NPS measures stated willingness to recommend. Steve Fenton, an eight-time Microsoft MVP and Octonaut at Octopus Deploy, describes how his employer limits the score’s role: “We use NPS as a signal, but don’t act on it directly. We track the trend and identify specific events that cause a drop.” Octopus Deploy is a software vendor with a commercial interest in customer loyalty, so this is an account of its own customer measurement.

In that account, Fenton described a pricing change that temporarily unsettled NPS, followed by work on high-impact product features and steady score improvement throughout 2026. A decline following a pricing change calls for investigation into the customer response before management decides whether to reverse it. The later recovery describes an improving recommendation trend, but renewal and revenue records are needed to establish whether customer behavior improved alongside it.

A hypothetical subscription software company could make that comparison by investigating onboarding difficulties, product limitations and unresolved support requests behind lower ratings. Comparing those findings with renewal data would help assess whether the concerns are associated with customers leaving. A cancellation investigation would also examine renewal behavior and customers’ reasons for departure. Retention, churn, purchasing, referrals and spending each reveal actions that recommendation ratings alone cannot establish.

Reichheld’s later work provides another way to examine that financial behavior. In 2021, he and two Bain colleagues introduced Earned Growth Rate, an accounting-based measure covering revenue growth from returning customers and the customers they refer. It measures the economic contribution of repeat business and referrals directly, extending the measurement question from what customers say to the revenue associated with what they do.

How much authority NPS should retain alongside those measures has been debated by CMSWire contributors in coverage since 2023. In “Stop Celebrating NPS, Start Fixing the Experience,” published April 22, 2026, Joseph G. DeRosa keeps NPS as a signal while criticizing celebrations of NPS and Customer Satisfaction Score, or CSAT, movements without corresponding customer change. He recommends defining success from the customer’s perspective, combining surveys with behavioral evidence and following through on feedback in real time.

DeRosa’s emphasis on customer outcomes leaves room to retain NPS; contributor Eric Karofsky gives conflicting behavioral evidence greater authority. In “NPS Is Lying to You,” published March 24, 2026, Karofsky proposes examining resolution, whether customers accomplished their intent, and live sentiment, their expressed reactions, in interactions handled by artificial intelligence, or AI. Such interactions can reveal behavior and immediate reactions that a later survey misses. He retains NPS where boards expect it while prioritizing more direct evidence in operating decisions.

Contributor Brian Riback goes further and rejects NPS. In “Net Promoter Score Isn’t a Growth Strategy, It’s a Comfort Metric,” published Feb. 26, 2026, as the first of a two-part series, he cites independent research that, in his account, fails to connect NPS reliably with revenue growth or share of wallet. Share of wallet is the proportion of a customer’s relevant spending captured by a business. Riback does not use NPS and favors repeat purchase, retention and share-of-wallet data.

Benchmarks need comparable populations and methods

Behavioral evidence helps a business judge its own outcomes; external benchmarks show how its score compares with a reference population. Survicate’s 2025 NPS Benchmarks by Industry analyzed more than 5.4 million responses across 2,187 surveys from 599 companies. The surveys came from Survicate customers and ran between January and October 2025. Those details define the population and collection period behind the comparison.

Within Survicate’s dataset, median NPS was 49 for business-to-consumer, or B2C, businesses and 38 for business-to-business, or B2B, businesses. Retail and ecommerce had a median of 55, compared with 30 for software. Between 2024 and 2025, medians declined in 10 of the 11 industries tracked; retail and ecommerce fell from 68 to 55, while the cross-industry median remained 42. These figures provide reference points, but the comparisons alone cannot explain industry differences or establish universal targets.

Before adopting such a reference point, compare the audience, customer population, geography, timing and collection channel or method. Establish whether the survey concerns a particular interaction or the broader relationship, because a survey after a successful purchase asks customers to assess a different experience from a relationship-wide survey. Within software businesses, account administrators and individual users can also report valid but different experiences. A benchmark becomes useful when those differences are understood.

Comparisons across markets require another check: how people use the rating scale. In “Net Promoter Score: Top 3 Impactful Trends,” published Sept. 25, 2023, CMSWire contributor Olga Potaptseva addresses cultural differences in use of the 0-to-10 scale. She also discusses machine learning to predict scores, benchmarking beyond industry peers and using NPS within a balanced measurement system. Before using a benchmark median to set a target, examine its research method and relevance to the customers served.

Build a measurement process around the decision

The need for comparable results starts with survey design. Decide what the business needs to understand, then choose relationship NPS for the overall customer relationship or transactional NPS for a particular interaction. Keep those result sets separate. A support call can affect a relationship, but its rating concerns a different subject from the customer’s assessment of the business as a whole.

Once the subject is clear, choose a survey tool and delivery method suited to the customers being asked. Email, text or an in-product invitation can carry the standard recommendation question and rating scale. Follow it with “What is the main reason for your score?” A brief survey with neutral wording lets customers explain their ratings without encouraging a favorable answer.

The invitation plan then determines whose explanations the business can receive. Define recipients and timing, include perspectives beyond completed purchases and resolved support cases, and maintain a consistent schedule while limiting repeated invitations. Record invitations and responses at the participation stages described earlier. Consider how respondents may differ from nonparticipants so the reporting keeps missing experiences visible.

With collection rules established, investigate the results through comments, customer groups and statistical analysis. Read explanations alongside ratings, group recurring concerns and compare relevant segments. Check response counts and population changes before interpreting small movements. This sequence helps distinguish a recurring experience problem from a concern concentrated in one group or a shift in who answered.

Lax recommends key driver analysis to examine statistical relationships between experience ratings and willingness to recommend. A business might collect assessments of product reliability, ease of use and support responsiveness, then test how strongly each relates to recommendation ratings. The analysis helps identify, quantify and prioritize areas for investigation. A frequently mentioned concern may differ from the factor with the strongest statistical association.

For example, a hypothetical software business might find that difficulty completing core tasks has a stronger association with low recommendation ratings than opinions about interface appearance. That finding would justify investigating usability before cosmetic changes. But customer tenure, pricing and product-plan differences could influence both experience ratings and recommendation intent, creating an apparent relationship between them. Targeted changes followed by fresh feedback and business-outcome checks help test the explanation.

An investigation also needs an owner who can act on its findings. In a hypothetical case involving repeated billing errors, the billing team would examine why the errors recur. Where respondents have agreed to follow-up, acknowledge the concern, explain the next step and track the issue through resolution. A thank-you message completes the acknowledgement; the team still needs to resolve the problem.

Responsibility continues after the intervention because the business needs to know whether it worked. Remeasure under comparable conditions and assess whether improved feedback accompanies better customer outcomes. Comparable conditions make the change easier to interpret, while outcome checks establish whether the intervention addressed the intended problem. Ownership therefore extends from identifying the concern to evaluating what happened after action.

Choose measures that explain the outcome

Evaluating an intervention requires measures suited to the outcome being investigated. CSAT measures satisfaction with a particular experience, while Customer Effort Score, or CES, measures how easy a task was to complete. Recommendation intent, satisfaction and effort ask different questions; retention and resolution records describe different behaviors. The decision the business needs to make determines which evidence is relevant.

For example, a support team trying to reduce repeat contact needs to establish whether problems were resolved and why customers returned. It can connect those records with survey evidence, then test which measure best explains the outcome it wants to change. Lax recommends starting with the outcome to decide where NPS belongs. Incentives should follow the same logic by rewarding problem resolution.

Changing the metric without changing those incentives can preserve the original problem. CMSWire contributor Sue Duris makes this organizational distinction in “Why NPS Became Customer Experience’s Favorite Punching Bag,” published March 31, 2026. She defends NPS within a mature measurement system and locates failure in compensation links and treating a metric as a complete CX strategy. Her five-question test examines whether a CX measure drives decisions, and she argues that replacing NPS with CSAT or CES can reproduce the same failure.

More complex customer journeys may require several kinds of evidence to support those decisions. In “The Future of Customer Experience Metrics: Moving Beyond NPS,” published Feb. 3, 2025, CMSWire contributor Jill Grozalsky Roberson recommends supplementing NPS. She discusses CES, sentiment analysis to examine expressed reactions, customer lifetime value to assess relationship value, and first contact resolution to track problems solved in the first support interaction. She also considers Experience Quality Index, a measure of experience quality, and predictive analytics, which uses data to estimate future outcomes.

Trust presents a different measurement question again because a recommendation rating captures sentiment at the time of asking. In “How to Measure Brand Trust Without Relying on NPS,” published Nov. 8, 2024, Karofsky distinguishes short-term recommendation sentiment from longer-term trust, loyalty and resilience. He compares the federal government’s trust question, the HBR/Deloitte HX TrustID survey and the Trust Equation, a framework for assessing trustworthiness. Choosing among these approaches begins with deciding which aspect of trust the business needs to understand.

Lax makes the selection principle explicit: “While the Net Promoter Score metric has soared in popularity and offers a simple and straightforward way to gauge brand loyalty, it is not always the right CX metric to focus on,” he said. “Businesses that want to build better customer experiences should never blindly adopt NPS as their holy grail. The best CX metric is that which best explains the customer behaviors you are trying to encourage and the business outcomes you want to achieve.”

In conclusion

A rising NPS is a reason to investigate, not proof that customer experience improved. Before using it to assess performance, leaders need to know who received the survey, whose experiences are missing and whether collection practices encouraged favorable ratings. The score’s distribution and customer explanations then help identify where action may be needed.

The next test is whether that evidence agrees with customer behavior. If NPS rises while renewals fall or unresolved support cases increase, the disagreement deserves attention. Recommendation intent cannot establish whether customers stayed, completed their tasks or had their problems solved.

For executives, the decision is how much authority to give NPS. It can remain a useful signal, but targets and compensation should not reward teams for protecting a score at the expense of hearing difficult feedback. Assign responsibility for recurring problems, track them through resolution and assess whether interventions improve the outcomes that matter to customers and the business.

A measurement system earns its place by helping leaders make better decisions. Keep NPS where it contributes to that work, pair it with evidence that can challenge it and change the measures when they no longer answer the business question.