Sixty-six percent of CX practitioners believe their customer experience is improving. Seventeen percent of consumers agree. That is a 49-point gap between the people running the program and the people the program is supposed to serve.

The source of the finding is what makes it difficult to dismiss. This is not an academic study or a consulting firm's provocation. It is the 2026 State of Customer Experience Report published by Medallia, drawn from 552 practitioners, 1,522 consumers, and over 600 anonymized enterprise programs running on the platform that produced the data. The vendor's own research, conducted on the vendor's own installed base, documenting a chasm between what its customers believe and what their customers experience. You could think of it as a hospital commissioning a study on patient outcomes and discovering that two-thirds of its doctors believe patients are getting healthier while the patients themselves largely disagree. At some point, someone has to ask whether the diagnostic equipment is working.

This is not, incidentally, the first time this particular finding has surfaced. In 2005, Bain and Company partnered with Satmetrix (then the home of NPS) to study the gap between executive self-perception and customer reality. They found that 80 percent of senior leaders believed their company delivered a superior customer experience. Eight percent of their customers agreed. A 72-point gap. Two decades later, after hundreds of billions of dollars invested in customer experience programs, measurement platforms, voice-of-the-customer technology, and more recently the bolting of generative AI onto all of the above, the gap has narrowed from 72 points to 49. At this rate of improvement, practitioner perception and customer reality should converge sometime around 2075. Mark your calendars.

What makes the persistence of this gap genuinely interesting is not the number itself but what it reveals about the incentive structures surrounding it. Twenty years of additional measurement infrastructure have not materially closed the perception gap. They have, however, generated a multi-billion-dollar industry selling the tools that produce the perception.

Why the gap is structural

The mechanism is not complicated, which is part of why it has been so easy to overlook. Survey-based CX programs hear from a fraction of the customer base. In enterprise B2B, response rates typically sit between five and fifteen percent, depending on channel and context. The Medallia report itself documents a further eleven percent year-over-year decline in response rates. The data source is shrinking. Research has consistently shown that non-respondents carry significantly lower NPS than respondents. Work by Bain and Company, among others, has documented this pattern across multiple industries. The customers most likely to be dissatisfied are the ones least likely to tell you about it through a survey. They do not fill in the form. They just leave.

So the practitioner who looks at the survey data and sees improvement may be entirely correct about what that data shows. The scores, among the people who responded, may genuinely be going up. The mistake is in what the practitioner believes the data represents. They are looking at a portrait of the six or fifteen percent of customers who engage with the feedback infrastructure and reading it as a portrait of the whole. It is not. It is a self-selected sample biased toward satisfaction, presented as though it were the population.

You could think of it as polling only the people who show up to town hall meetings and concluding the whole town is engaged. The people who show up are, almost by definition, the ones who care enough to show up. The ones who do not attend are the ones whose views you most need and least have. And when you report the poll results to the mayor, you are not lying. You are accurately describing the room. You are just wrong about who was not in it.

Medallia's report, to its credit, acknowledges several of the symptoms. Seventy-five percent of practitioners in the survey agree that surveys alone are insufficient, yet surveys remain the primary data source. One in three departments receiving CX insights takes zero action on them. One in three practitioners reports no demonstrable relationship between their NPS score and financial performance. A Forrester analyst presenting at Medallia's own webinar confirmed that nearly 90 percent of programs still use surveys, but fewer than a quarter of leaders rate their surveys as effective or highly effective. The practitioners know. They are telling the vendor's own survey that the vendor's approach has a problem.

The Gartner Magic Quadrant for Voice of the Customer Platforms, published independently in March 2026, arrives at a structurally identical conclusion: most enterprise CX programs remain largely survey-driven despite significant vendor AI investment, and the critical capability gap — connecting CX signals to financial outcomes — remains unsolved across all category leaders. 

The vendor's diagnosis, and why it makes things worse

Medallia's 2026 positioning is instructive. Having published research demonstrating that survey-based tracking is, in their own words, "failing to capture the true reality of today's complex journeys," Medallia's prescription is not to move beyond surveys. It is to reposition the CX function as a "Center of Impact," to unify "every signal across every channel, every journey, every role," and to layer AI services on top of the existing architecture. The framing has shifted from "listen to your customers" to "become a business driver." The underlying data infrastructure has not shifted at all.

This is worth pausing on, because the irony is precise. Medallia has published peer-reviewed-quality evidence that the measurement approach its platform is built on does not capture customer reality. Its response to this finding is not to change the measurement approach but to change the organizational framing around it. Stop calling CX a cost center. Start calling it a Center of Impact. The data feeding the center is still the same five-to-fifteen percent of accounts who responded to the survey, but the PowerPoint slides describing what the center does have been meaningfully updated.

Their marquee 2026 customer showcase, Santander, is revealing in ways Medallia may not have intended. The Santander team explicitly operated under "no CX capital investment, no new tools, no added resources." The transformation was achieved by working better with what they already had. The success story, in other words, is that generating meaningful value from the Medallia platform requires an organization to invest heavily in change management, data integration, team restructuring, and behavioral discipline, none of which is in the platform. The technology is a foundation, not an engine. And the celebrated outcome — reducing projected attrition to 21 percent of forecast — was achieved by detecting signals of dissatisfaction and intervening with closed-loop follow-up. This is reactive signal detection, not prediction. The accounts that never complain are the ones the platform cannot see.

Why management is not more alarmed

The more interesting question is why leadership teams are not more concerned about this. Several explanations coexist, and none of them are particularly flattering.

The most charitable interpretation is ignorance. Many executives receive CX data as a finished product, a score on a slide, a trend line in a quarterly review. The statistical limitations of that score, the response rate it rests on, the profile of who is missing, are rarely presented alongside it. This is partly because CX teams have limited incentive to volunteer the information. Going to your leadership team and explaining that the data you have been presenting for the past three years is too weak to support the conclusions drawn from it is not, generally speaking, a career-enhancing move. The data continues to be presented, the limitations continue to be footnoted at best, and the leadership team continues to operate on a foundation that the people closest to it know is unreliable. This is not malice. It is institutional self-preservation, which in practice produces roughly the same result.

A second explanation is selective credulity. Sales teams, interestingly, have always doubted survey data instinctively. They look at the NPS score for an account they know well and say, "That does not look right. The wrong people responded." And they are often correct, though not always for the reasons they think. But the skepticism flows in one direction. Negative survey data is treated as suspect: "The detractors are always the loudest, the sample is too small, the timing was wrong." Positive data is treated as confirmation: "See, they love us." This is textbook confirmation bias, and it operates at the organizational level with remarkable consistency. The survey data is simultaneously too unreliable to act on when it says something uncomfortable and perfectly reliable when it says something convenient.

A third explanation is structural. The incumbent vendors in this space — companies with multi-billion-dollar survey businesses — have a natural and entirely rational incentive to promote the idea that the data is fundamentally sound and that the problem lies elsewhere, in process, in organizational alignment, in the failure to "speak business language." One of the more effective distractions the industry has created is the narrative that "nobody is taking action on the data." This may well be true. One in three departments receiving CX insights takes zero action, according to Medallia's own numbers. But the root cause of that inaction — which is that the data is not credible enough to act on — is rarely discussed. Instead, the conversation pivots to process improvement, executive sponsorship, and cultural transformation. These are useful ideas, and they are irrelevant if built on a foundation of data that represents six percent of your customer base and systematically excludes the accounts most likely to leave.

The cumulative effect is a kind of institutional equilibrium. The vendor sells the platform. The CX team reports the data. Leadership receives the score. Nobody in the chain has a strong incentive to examine whether the score means what everyone has agreed it means. And so the 49-point gap persists, not because anyone is being dishonest, but because the architecture of the conversation has been designed, over two decades and several hundred billion dollars, to make the question very difficult to ask.

What the practitioner is actually wrong about

It is worth being precise about this, because the imprecise version — "CX programs do not work" — is both unfair and unhelpful.

The practitioner is not wrong that the data they are looking at shows what it shows. Survey scores among respondents may genuinely be trending in a positive direction. The practitioner is not incompetent. They are not making an analytical error. They are working with the instrument they have been given, and working with it competently.

What the practitioner is wrong about is the relationship between that instrument and the customer base it is supposed to represent. They are wrong about coverage. They are wrong about who is missing. And they are wrong about what the missing data means — which is not nothing, but something considerably more important than a gap to be footnoted and moved past. The eighty-five to ninety-four percent of the customer base that does not appear in the survey data is not absent. They are making decisions. They are forming views about whether to renew, expand, or leave. They are just doing it silently, outside the measurement infrastructure, and the infrastructure has no mechanism for hearing them.

The 49-point gap is what it looks like when an entire industry builds its confidence on one end of that silence and the customer base lives on the other. What made the gap 72 points in 2005 and 49 points in 2026 is not that the industry got materially better at understanding customers. It is that it got better at understanding customers who were already talking. The ones who were not talking in 2005 are still not talking. The measurement infrastructure has simply gotten more sophisticated at ignoring them.

If you are a CX leader presenting survey-based metrics to a board, the question the 49-point gap raises is not whether your data is accurate. It probably is, within its own frame. The question is whether the frame includes the customers whose decisions will determine your next renewal quarter, and, if you are honest about the arithmetic, what it means that it almost certainly does not.