♛
05.05.2026
The 72-Point Lie
Read Time: 7 Minutes
Hello
In 2005, Bain surveyed 362 companies. 80% of executives said their company delivered a superior customer experience. 8% of their actual customers agreed. The gap is 72 points. It is the most quoted statistic in customer experience consulting. It is also the most ignored.
I keep coming back to it. Twenty years later, the gap has not closed. Not closed, not narrowed in any meaningful way, not even contested. The global insights industry is at $153B a year. There are more research vendors, more dashboards, more brand trackers, more NPS programs, more persona documents, and now a full new layer of synthetic respondent platforms launching every quarter. Companies are spending more, measuring more, and reporting more. And the 72-point gap is still there.
I have spent enough time around customer insight functions to know why. The gap is not a measurement problem. The instruments are honest. The teams are diligent. The vendors are competent. The gap is structural. The instruments cannot see it because the instruments are inside the company that is trying to close it.
I want to walk through the mechanism. It is uncomfortable. I am writing it anyway.
The persona document is fan fiction
I have seen probably two hundred persona documents in the last decade. They all look the same.
The persona has a name. Alex, Jamie, Morgan. The photo is a stock image of a tastefully casual person in their mid-thirties. There is a quote across the top, two articulate sentences, conveniently aligned with the company’s product differentiator. There is a list of pain points, three to five items, each of which maps cleanly to something already on the roadmap.
Companies pay between $40,000 and $400,000 for these documents. They get printed and pinned to the wall. Product managers point to them in meetings to end arguments. “Alex wouldn’t want that,” someone says. The argument ends.
Here is what actually produced Alex.
Twelve interviews. Maybe fifteen. Conducted with customers who agreed to be interviewed. Customers who agree to be interviewed are, by definition, not the customers quietly considering churn. Customers thinking about churning do not opt into a sixty-minute conversation about how they feel about your brand. Advocates do.
Martin Orne published a paper in 1962 called “On the Social Psychology of the Psychological Experiment.” It introduced the concept of demand characteristics. He showed that respondents in any research setting unconsciously perform for the researcher. They give the answer they sense the researcher wants. Because that is how humans behave in an interaction where one party asks, and the other is being studied. The paper has been cited for sixty years. It is foundational behavioural science. Every researcher knows about it. Most insight functions still proceed as if it does not apply to them.
So your persona is what twelve cooperative people, performing for an interviewer who works for you, said when asked questions written by someone on your payroll, summarized by an analyst who has already read last year’s deck.
That is not the customer. That is a composite of what your team already believed, laundered through a process that looks rigorous because it costs money.
The customer who matters this quarter has no name. They have a Reddit comment.
The survey is a closed frame
Every survey has the same structural problem. Someone wrote the questions. That someone works for you. They wrote the question about onboarding because onboarding came up in the last QBR. They wrote the question about pricing because pricing keeps showing up in deal-loss notes. They did not write a question about the competitor that nobody in your building has heard of yet. They could not. Nobody in your building has heard of them.
Tversky and Kahneman published a paper in 1981 that has been cited over thirty thousand times. It is one of the most referenced papers in behavioural science. They demonstrated that logically identical problems, when framed differently, produce complete preference reversals. Not marginal shifts. Reversals. The frame of the question determines the shape of the answer.
Your survey is a frame. It is a frame your team built. The data that comes back is real, and honest, and useful within the frame. It is also the most unmet need among the things you knew to ask about. The thing that is actually losing you deals, the thing three of your best accounts are quietly thinking about right now, is not on the slide. It cannot be on the slide. You did not know to ask.
The standard cadence for enterprise brand tracking is twice a year. Question list locks six weeks before fieldwork. Results land eight weeks after. That is a four-month lag to confirm what you thought you already knew. The market does not run on that cycle. The market runs on whatever a procurement lead is typing into Perplexity right now.
NPS is structurally broken, and Bain knows it
NPS is the most widely used customer metric in enterprise software. It is also the most systematically misleading single number any insights team produces.
Rob Markey co-created NPS at Bain. He has acknowledged in published work that NPS surveys produce consistent positive bias. Promoters are the most likely to respond. Detractors are the least likely. Survey response rates run well under 15% in most channels, often half that. B2B averages around 13%.
That means your NPS score is the opinion of roughly one in seven of your customers, structurally weighted toward the customers who like you most. The rest are not a missing sample. They are a hidden population. They are doing something else. They are on Reddit. They are on Trustpilot. They are on G2. They are writing the actual story of your brand in real time, unprompted, with no awareness of your NPS program and no obligation to be diplomatic.
Your NPS score is what the responders thought when you asked them. The non-responders are saying something else when you don’t ask. Only one of those determines whether you hit your renewal number, and it is not the one in the deck.
The synthetic persona is the most dangerous instrument in your stack
73% of market researchers reported using synthetic responses in 2025. One in three deployed them within the previous thirty days. The vendor list grows every quarter. Synthetic Users. Yabble. Evidenza. Aaru. Simile. Ditto. Lakmoos. Toluna HarmonAIze. Qualtrics shipped its own version. Plus, every team that built one with ChatGPT, Claude, or Gemini and called it a research panel.
I want to be precise about what these are.
A synthetic persona is a large language model, trained on public internet text, generating a statistically plausible answer to your question in the voice of a fictional person, based on the aggregate of what people have said about vaguely similar topics in the past. It is auto-complete with a name tag.
Carnegie Mellon researchers studying synthetic responses found what they called the surrogate effect. Synthetic answers cluster more tightly around means than real human data. Real humans are inconsistent, contradictory, and sometimes incoherent. That mess is not noise. It is the actual texture of human decision-making. Synthetic responses sand it smooth. The headline number from a synthetic study can look clean while the underlying structure of the data is wrong.
Nielsen Norman Group ran synthetic users against three of their own real studies and concluded the synthetic responses were “too shallow to be useful,” sycophantic, and biased toward telling the team what the team wanted to hear. Their 2025 follow-up softened the position, suggesting synthetic outputs may have directional value in exploratory research. Even at that softened level, the case for routing their output into GTM decisions does not hold up. ACM Interactions ran a separate critique that called the practice “epistemic freeloading,” appropriating the language of research while abandoning its standards. One documented case in that critique: a design team spent three weeks generating synthetic personas, then one hour with real rural patients revealed the depth the synthetic version could not produce.
Your team is not buying a panel. They are buying the most expensive confirmation engine in the history of customer research. And they are presenting their output to the board as the voice of the customer.
The deck is air cover, not intelligence
I will say the thing nobody says in the readout. A meaningful portion of enterprise customer research is not commissioned to learn. It is commissioned to provide organizational permission to do what leadership has already decided.
A leader has a direction in mind. The direction is politically sensitive. Maybe it is a price increase. Maybe it is a product line shutdown. Maybe it is a GTM. The org needs a reason to move. The research gets briefed. The brief shapes the questions, not corruptly, just institutionally, toward the kind of finding the initiative needs. The readout surfaces the predetermined conclusion with a margin of error and a confidence interval.
It works. The initiative gets approved. Everyone nods. Nobody gets fired.
Most brand-side teams do not incorporate social listening data into routine decisions. Agencies use it somewhat more often. The cleanest data, the unprompted public language of actual customers writing without a researcher in the room, is the data most systematically excluded from the decisions it should inform. The data that gets into the board deck is the data that was designed to get into the board deck.
This is not cynicism.
It is organizational physics.
Large organizations move on narrative. Customer research is the most politically defensible source of narrative available, so it gets shaped to serve that function. Every time. The cost is not visible in the quarter the decision gets made. It accumulates in the quarters after.
What the gap looks like when you go look at it
Tangerine has won J.D. Power #1 in client satisfaction among midsize Canadian banks for fourteen consecutive years. Forbes named them #1 Bank in Canada in both 2025 and 2026. The award is on their homepage. It is in every board deck. It funds the next year’s marketing budget.
In the same window, Trustpilot is full of “constant crashes,” “issues with e-transfers,” and “horrible customer service.” Reddit is full of “things just keep getting worse.”
Both data sets are real. The award is built on a structured study with sound methodology, capturing thousands of verified customers. Those customers are real people. They are also the customers who were reachable, willing to participate, and in a state of mind to answer survey questions. The customers writing on Reddit at 11pm on a Thursday were not in a state of mind to answer survey questions. They were in a state of mind to tell the truth.
The award was earned by a version of the operation that no longer describes the whole experience. It now functions as institutional cover for not fixing what the Reddit threads are naming.
The most expensive lie in marketing is the truth you stopped updating.
WHOOP has the same problem in a different shape. Their brand customer is Cristiano Ronaldo. He had been a private user for years before becoming a global ambassador and investor in May 2024. That is the customer in the brand deck. Articulate, aligned, aspirational.
When WHOOP 5.0 launched in May 2025, the actual customer base, the 2.5 million paying members, started writing a different story. WHOOP had previously published a blog post promising free hardware upgrades to members with six or more months of tenure. The 5.0 launch came with $49 to $79 upgrade fees. Bloomberg covered it. The Verge covered it. TechCrunch covered it. The reaction was fast enough and loud enough that WHOOP reversed within days. Free upgrades for members with twelve or more months left on their subscription. Refunds for those who had already paid.
The reversal is the evidence. The word “membership” had been wired into the customer base as a procedural noun that meant tenure entitled you to something. When the policy looked like a subscription, the cognitive dissonance was immediate, visceral, and public. No amount of Cristiano Ronaldo content could absorb it. Cristiano Ronaldo is not the customer who churns. He is the customer in the deck.
Gong has the third version
Gong calls itself a “Revenue AI Operating System.” That is the noun Gong chose, announced at Celebrate in October 2025 and extended through Mission Andromeda in February 2026. The noun customers use, on Capterra and Reddit and G2, is “coaching tool,” or “call recorder,” or “that expensive thing for large sales teams.” An operating system requires a hundred-thousand-dollar enterprise procurement process and a dedicated RevOps owner to justify. A coaching tool does not. The buyer’s brain has not made the leap Gong’s framing assumed it would make. Every expansion motion lands in a buyer who was never cognitively prepared to receive it.
The map the analyst drew has Gong in the upper right quadrant of “Revenue Intelligence Platforms.” The map the market drew has Gong filed under “expensive call recorder.” Only one of those determines whether the upsell closes.
Your company has the same problem. You just do not know which noun the market is actually using for you yet. Your insights stack cannot tell you. It was not built to.
Seven layers, all in the same direction
Every standard research methodology distorts. That is not the problem. The problem is that they all distort in the same direction.
Framing effects.
The question constrains the answer.
Demand characteristics.
Respondents perform for researchers.
Non-response bias.
The most frustrated customers do not respond, the advocates do.
Social desirability bias.
In interviews, people report the behaviour they aspire to, not the behaviour they actually exhibit.
Confirmation bias in analysis.
Contradictory findings get categorized as outliers, and confirming findings get categorized as insights.
Survivorship bias in the sample.
You only interview customers you still have.
Recency and saliency bias.
Customers report vivid recent experiences, while chronic friction remains invisible because it never produces a single memorable moment.
Each one individually distorts a little. Together, they compound every layer in one direction. Toward the company looking better than it is. Toward the gap being smaller than it is. Toward the problems being less urgent than they feel on Reddit at 11pm on a Thursday.
Your Qualtrics is honest.
Your NPS is honest.
Your brand tracker is honest.
They are looking at a reflection.
Nobody told them.
What is actually happening
The market is not quiet. It is talking constantly. It is talking right now, on forums and review sites and community threads and analyst notes and earnings call transcripts and language model search results, completely outside your research stack, with no awareness of your quarterly cadence and no obligation to perform for an interviewer.
That language, the unprompted, uncurated language written when nobody in your building is asking, is the only data that tells you what the market actually believes about you. What it says when nobody is asking.
Two positionings exist at all times. The intended one, the one in your decks and your brand guidelines and your offsite output. The operating one, the one that shows up in deal-loss notes, in analyst categorization, in search results, in review patterns, in the noun the market reaches for without prompting. Your stack is built to measure the first one. It is structurally incapable of measuring the second.
That is what the 72-point gap actually is. It is the distance between the company’s intended positioning and the market’s operating positioning. It has persisted for twenty years because the budget allocated to close the gap and the instruments meant to detect it are governed by the same organizational immune system. The system selects for confirmation. It cannot help itself. That is what it does.
Where this leaves you
You have a stack. You have a brand tracker, an NPS program, a persona doc, a competitive map, and probably a Qualtrics dashboard with a live widget. People on your team believe in it. The line items renew every year.
And right now, while you read this, a customer is writing a review you will never see, in language your team has never coded, naming a problem your roadmap has never named, that will cost you three accounts next quarter.
I built Monopoly because I got tired of watching this happen and having nothing to point at. Monopoly reads the language the market actually uses. It maps the noun the market reaches for when nobody is asking. It surfaces the gap between your intended positioning and your operating positioning, in the only register that determines whether the deal closes.
It is a correction layer for the stack you already have, so when you do invest in surveys or panels or interviews, you are asking questions the market is actually answering. It uses 75+ public data sources, AI Visibility analysis, perception gap measurement, and real human behavioural evidence. It is built for $100M+ revenue companies because that is where the cost of the gap is largest, and the cost of the correction is smallest.
The choice is the one every leader has been making for twenty years without knowing it. Keep buying research that comforts you. Or buy intelligence that corrects you.
One feels safer. The other moves the number.
Monopoly 2.0 is here.
Take a look monopoly.ceo
DIGEST
