The Man Who Invented NPS Just Told Us to Stop Trusting It

In 2025, on the main stage at Medallia Experience, Fred Reichheld—the Bain & Company partner who introduced the Net Promoter Score in 2003—said something few people in the customer experience industry expected to hear.

“I’m sick of surveys. I don’t fill them out anymore.”

He said surveys had been “abused so horribly” and called the Net Promoter branding itself a profound mischaracterization. Medallia described a collective gasp from the audience.

It was not a funeral for NPS. Reichheld was not arguing that customer loyalty, advocacy, or experience had stopped mattering.

He was warning that an instrument had become a substitute for the thing it was supposed to help us understand.

That distinction matters—especially in banking.

Why one number conquered customer experience

It is worth remembering why NPS won.

Before 2003, customer experience measurement was dominated by long satisfaction surveys that were expensive to run and difficult to explain to a board. Reichheld offered an irresistible alternative: one question, one score, and an apparent connection to growth.

Executives got a metric that fit inside a quarterly deck. Consultants got a common language. Software companies got a category. An entire operating system of surveys, benchmarks, targets, incentives, and Chief Customer Officer mandates grew around the promise that one number could summarize the customer relationship.

But Reichheld’s original argument was more nuanced than the way many companies eventually implemented it. His research attempted to connect survey answers with actual purchasing, referral behavior, and company growth. The recommendation question was supposed to be useful because it related to economics—not because asking the question was itself a strategy.

Over time, too many organizations detached the score from the behavior. NPS moved from signal to scoreboard, and then from scoreboard to target.

That is where the trouble started.

What the evidence has—and has not—shown

In 2021, Gartner predicted that more than 75% of organizations would abandon NPS as a measure of success for customer service and support by 2025. A later TELUS Digital and Statista survey, reported by CMSWire, found that only 23% of U.S. enterprise CX leaders were using NPS to measure performance.

NPS did not disappear. It was quietly demoted into a larger measurement stack.

Independent research has also challenged its most famous promise. A longitudinal study published in 2024 examined NPS and revenue across airlines, supermarkets, and insurance companies over periods ranging from five to eleven years. Its conclusion was direct: NPS was not an indicator of future revenue growth.

That does not make NPS meaningless. It means NPS is not a reliable stand-alone growth instrument.

The larger problem is conceptual: organizations keep asking one attitudinal measure to explain an entire relationship.

The missing half of loyalty

Marketing science identified the problem long before NPS existed.

In 1994, researchers Alan Dick and Kunal Basu described customer loyalty as a relationship between two dimensions: relative attitude and repeat patronage.

In plain language:

• Attitudinal loyalty is how customers think and feel: preference, trust, satisfaction, commitment, and willingness to recommend.
• Behavioral loyalty is what customers actually do: return, transact, consolidate, purchase, and allocate their wallet.

The two can reinforce each other. But they are not interchangeable. Situational constraints, switching costs, convenience, pricing, habit, and available alternatives can all pull attitude and behavior apart.

Many NPS implementations flattened that two-dimensional relationship into one visible score.

The evidence for the gap is not subtle.

Bain has long reported that 60% to 80% of customers who defected to a competitor described themselves as satisfied or very satisfied shortly before leaving. Satisfaction did not guarantee staying—which, ironically, was one of the problems NPS was originally designed to address.

A 2026 study made the behavioral gap even more concrete. Researchers examined 532 repeat customers across face-to-face service industries in Japan. Relative satisfaction explained just 0.8% of the variation in share of wallet.

That is not proof that satisfaction never matters. It is evidence that a strong experience evaluation can coexist with very different spending behavior.

To be fair to the other side

It would be dishonest to present this as a settled, one-sided case.

Attitude and behavior are not strangers. Banking research linking customer surveys to actual CRM purchase records has found that attitudinal loyalty can predict recommendations, cross-buying, and total spend—and that the strength of those relationships can vary by how the customer was acquired.

Accenture’s 2025 global banking study offers another important counterweight. It found that banks with the highest customer advocacy scores grew revenue 1.7 times faster, while advocates held an average of 17% more products with their primary bank.

So the argument is not that customer attitudes are irrelevant. At a portfolio or brand level, advocacy, trust, and experience can correlate with meaningful economic outcomes.

The problem appears when a macro relationship is treated as a micro diagnosis.

A bank can know that advocacy leaders tend to grow faster and still be unable to infer the condition of one customer’s relationship from a single survey response. Correlation across institutions or segments does not tell an operator what one customer will do next—or why.

That inconsistency is the point.

NPS can be a useful measure of expressed advocacy. It is not the entire customer relationship, and it should not be asked to behave like one.

CX still matters. Sentiment is only half the instrument.

None of this is an argument against customer experience.

Customer experience influences trust, effort, advocacy, retention, product use, and growth. The mistake is not measuring how customers feel. The mistake is assuming that the feeling automatically reveals the behavior.

Consider two banking customers.

One scores the bank a 9 out of 10 but keeps most of their deposits, investments, and daily financial activity somewhere else.

The other scores the bank a 5 but still holds a mortgage, receives a direct deposit, pays bills, and carries fifteen years of history with the institution.

An NPS dashboard can distinguish their expressed attitudes. It cannot tell you which relationship is economically central, which one is healthy, or what the disagreement between attitude and behavior means.

Sentiment is not the wrong half.

It is only half.

Behavior completes the instrument.

What behavioral loyalty looks like in banking

Financial institutions often approximate behavioral loyalty through Primary Financial Institution status. But “primary” is not a permanent customer attribute, and it cannot be established by a direct-deposit flag alone.

The more useful management question is behavioral:

How central is this institution in the financial activity we can actually observe?

Potential evidence includes recurring inbound flows, transaction frequency, bill payments, card usage, active product use, recurring automated activity, tenure, and meaningful relationship interactions.

Much of that evidence already exists inside a bank’s transaction, product, payment, digital, and relationship systems. But internal activity still reveals only what happens inside the institution’s walls. It does not expose every competing account, external balance, or reason money moves.

That limitation matters. The correct construct is observed behavioral primacy, not omniscience.

And the relationship can erode without formally ending. J.D. Power reported in 2026 that 20% of retail bank customers had moved money away from their primary bank during the previous three months. The checking account remained open. The relationship simply became less central.

That is soft switching: the customer does not announce a departure. Their behavior migrates before the account closes—if it ever closes at all.

A survey may not see it. A closure report will see it too late. A direct-deposit flag can miss it entirely.

Stop confusing measurement with understanding

Forrester’s 2026 CX outlook warns that budget pressure will pull 15% of CX teams into a “death spiral” of metrics obsession—producing more dashboards while becoming less able to explain what should change, why it matters, or how it creates value.

That is the real warning in Reichheld’s comment.

The problem is not one survey question. The problem is what organizations allow one number to stand in for.

If NPS becomes a target, it gets managed. If it becomes the customer strategy, it crowds out conflicting evidence. If it becomes the definition of relationship strength, customers whose feelings and behavior diverge become invisible.

Banks do not need to throw NPS away.

They need to put it back in its proper place.

NPS can tell us something important about what customers are willing to say. Behavioral evidence can tell us something different about the relationship they are actually running. Neither is complete ground truth. Together, they can reveal where the signals agree—and where the disagreement demands investigation.

In the next article, I will introduce a framework I have been developing through my work in banking: Relationship Health × Behavioral Primacy.

It creates four relationship states that a sentiment score alone cannot reveal:

• healthy but not primary;
• primary but unhealthy;
• weak on both dimensions; and
• strong on both.

The most important insight is not the classification itself.

It is this:

Primary does not mean healthy. Healthy does not mean primary. A bank needs to know the difference before it acts.