Market Research: Why Your Customers Are Lying to You (And How to Find the Truth)

In the spring of 1985, the Coca-Cola Company had a problem that wasn't a problem at all. They were the most recognized brand on the planet. They'd been selling the same beverage for ninety-nine years. But Pepsi was gaining ground — slowly, steadily, one blind taste test at a time. The Pepsi Challenge, a brilliantly simple marketing campaign in which consumers sipped from unlabeled cups and overwhelmingly picked Pepsi, had been running for over a decade. And it was working. Coca-Cola's market share had been sliding since the early 1970s, from over sixty percent to just under twenty-four percent. Chairman and CEO Roberto Goizueta decided it was time to do something unprecedented: change the formula.

But Goizueta didn't act recklessly. He acted like a man who believed in data. Coca-Cola spent over four million dollars and conducted nearly 190,000 blind taste tests across the United States and Canada. The results were unambiguous. Consumers preferred the new, sweeter formula over both the original Coke and Pepsi by a margin of roughly 53 to 47. A second research firm was hired to repeat the study with another 100,000 participants. Same results. The data was overwhelming, consistent, and clear.

On April 23, 1985, Goizueta stepped before the press at Lincoln Center in New York City and introduced what he described as "smoother, rounder, yet bolder — a more harmonious flavor." The old formula was gone. New Coke had arrived.

Within weeks, the company was fielding 5,000 angry phone calls a day. By June, that number had swelled to 8,000 — a volume so crushing that Coca-Cola hired extra operators just to absorb the fury. A grassroots protest group called "Old Cola Drinkers of America" organized demonstrations where consumers poured bottles of New Coke into sewer drains. People began hoarding cases of the old formula the way survivalists hoard canned goods. One man in San Antonio drove his pickup truck to a bottling plant and bought $1,000 worth of original Coke before the last shipments ran dry.

Seventy-nine days after its launch, Coca-Cola surrendered. They brought back the original formula under the name "Coca-Cola Classic." Goizueta, standing at the same podium, faced a room of reporters who were no longer skeptical, they were gleeful. The reversal made front-page news in every major newspaper in the country.

Here is the part that matters for anyone building a product or launching a company: the research wasn't wrong. The 190,000 taste tests accurately measured what they set out to measure. People genuinely preferred the sweeter formula in a blind sip test. The catastrophic failure wasn't in the data. It was in the question. Coca-Cola's researchers had asked consumers which flavor they preferred. They never asked how consumers would feel if the new formula replaced the old one entirely. They measured a taste preference. They missed an identity.

That gap (between what market research measures and what actually drives behavior) is the subject of this post. And if you're building a company, it's the gap that will determine whether your product succeeds or quietly dies while your survey data tells you everything is fine.

The Say-Do Gap: Why People Can't Tell You the Truth

In 1977, psychologists Richard Nisbett and Timothy Wilson published a paper in Psychological Review that would become one of the most cited studies in the history of consciousness research. The title was precise and devastating: "Telling More Than We Can Know: Verbal Reports on Mental Processes."

Their central finding was this: people have little or no direct introspective access to their own higher-order cognitive processes. When asked to explain why they made a choice, they don't access the actual mechanism that produced the choice. They construct a plausible-sounding story after the fact: a story that feels true, that sounds rational, but that has no reliable connection to what actually happened in their brain.

One of their most elegant experiments involved four pairs of nylon stockings hanging on a rack. Participants were asked to evaluate which pair was the best quality, examining the knit, the weave, the sheerness, the elasticity. They handled each pair carefully. They deliberated. And then they chose.

The stockings were identical.

What drove their decisions was position. Twelve percent of participants chose the pair on the far left. Seventeen percent chose the second pair. Thirty-one percent chose the third. And forty percent chose the pair on the far right: a nearly four-to-one preference for the last item examined, driven entirely by the recency of handling it. This is a well-documented effect in consumer psychology: when evaluating options sequentially, people tend to prefer whatever they touched most recently.

But here's the finding that matters for market research: when asked to explain their choice, participants cited the knit, the weave, the sheerness, the elasticity, properties that were identical across all four pairs. They gave detailed, confident, specific explanations for a preference that was caused entirely by spatial position. When the researchers asked directly whether the position on the rack might have influenced their choice, virtually no one said yes. Only one participant in the entire sample acknowledged that position could have played a role.

They weren't lying. They genuinely believed their explanations. The brain had made a decision through one mechanism (position effect) and then produced a completely different narrative (product quality) to explain it. Nisbett and Wilson's conclusion was stark: when people report on their mental processes, they aren't introspecting. They're theorizing. They're generating the most plausible-sounding explanation they can find, and they have no way to distinguish that explanation from the actual cause.

This is the foundation of what researchers now call the say-do gap: the systematic divergence between what people report about their preferences, intentions, and motivations, and what they actually do. The gap isn't occasional. It isn't limited to unusual circumstances. It's structural. And the data on its magnitude is sobering.

Sixty-five percent of consumers say they want to buy from purpose-driven, sustainable brands. Twenty-six percent actually spend their money that way: a thirty-nine percentage point gap between stated intention and revealed behavior. In ready-to-eat cereal purchases, the gap between purchase intent and actual purchasing is nearly fifty percent. Across categories, people overstate their likelihood of buying a product by a factor of five. They claim to behave twice as virtuously in recycling and dental hygiene as they actually do.

If you're building a company and your market research consists of asking potential customers what they want, what they'd pay, or whether they'd use your product, you are collecting the business equivalent of those stocking evaluations. Confident, specific, detailed answers that have no reliable connection to what your customers will actually do.

Revealed Preference: The Economist's Solution to a Psychology Problem

In 1938 (almost four decades before Nisbett and Wilson published their findings) economist Paul Samuelson had already proposed a theoretical framework for dealing with the unreliability of stated preferences. He called it revealed preference theory, and its premise was elegant: stop asking people what they prefer. Watch what they choose.

Samuelson's argument was that traditional economic theory depended on utility functions, mathematical representations of how much satisfaction a consumer derives from different goods. The problem was that utility is internal, subjective, and unmeasurable. You can't observe someone's satisfaction. You can only observe their behavior. So Samuelson proposed building economic theory entirely on observable choices. If a consumer buys product A when product B is available at the same price, then the consumer has revealed a preference for A. No surveys needed. No introspection required. Just behavior.

The distinction between stated preference and revealed preference has become one of the most important concepts in behavioral economics, and one of the most consistently ignored in startup market research.

Stated preference is what happens in a survey, a focus group, a customer interview where you ask "Would you use this?" or "How much would you pay?" Revealed preference is what happens at the cash register, in the app store, on the landing page where someone either enters their credit card number or clicks away. The two frequently disagree.

Sony learned this the hard way in the 1980s when they conducted a focus group for a new Walkman model. The product came in two colors: the traditional black and a bright yellow "sport" edition. Participants were enthusiastic about the yellow model. They loved the color, praised the sporty aesthetic, said it felt fresh and different. The moderator was pleased. The data was clear. People wanted yellow.

On their way out of the session, each participant was offered a free Walkman as a thank-you gift, their choice of black or yellow. Every single person took the black one.

This is the say-do gap in its purest form. In the social context of a focus group, participants said what felt interesting, novel, socially appealing. In the private context of an actual choice, they reverted to what they would really use. The stated preference was yellow. The revealed preference was black. And if Sony had built their production run on the focus group data alone, they'd have warehouses full of unsold yellow Walkmans.

The Herman Miller Aeron chair tells the opposite story. When focus groups were shown the chair's unconventional mesh design in the 1990s, they hated it. They called it ugly. One participant reportedly called it "the chair of death." Traditional market research would have killed the product. Herman Miller launched it anyway, based on ergonomic performance data and the behavior of people who actually sat in the chair for extended periods. The Aeron became one of the most successful office chairs in history, a fixture of every startup and design studio in America, and a permanent entry in the Museum of Modern Art's collection.

The focus groups were accurately reporting their aesthetic reaction. They were inaccurately predicting their purchasing behavior. Those are two entirely different things, and market research that conflates them will lead you astray every time.

The Mom Test: How to Ask Questions That Even Your Mother Can't Lie About

If asking people what they want doesn't work, what does?

Rob Fitzpatrick, a programmer and startup founder who'd watched too many companies build products based on false validation, wrote a slim, blunt book in 2013 called The Mom Test. The title comes from a simple observation: if you ask your mom whether your business idea is good, she'll say yes. Not because it is. Because she loves you and doesn't want to hurt your feelings. And the problem, Fitzpatrick argued, is that almost every customer conversation functions the same way. People are polite. They want to be helpful. They can see you're excited. So they tell you what you want to hear.

The book's central insight is that the quality of your market research is determined entirely by the quality of your questions, and that the questions most founders ask are engineered, without realizing it, to produce false positives.

"Would you buy this?" is the worst question in market research. It costs nothing to say yes. It requires no commitment, no sacrifice, no trade-off. It activates the respondent's desire to be agreeable rather than their actual evaluation of whether the product solves a problem they care about enough to pay for. Fitzpatrick's rule is absolute: never ask a question that your mother could lie about.

Instead, Fitzpatrick proposes three rules for questions that produce reliable signal:

Rule 1: Talk about their life, not your idea. Don't describe your product and ask if they'd use it. Ask about the problem your product is supposed to solve. "Tell me about the last time you tried to do X" is infinitely more useful than "Would you use a product that does X?" The first question retrieves a real memory from their real life. The second asks them to simulate a future behavior, which, as Nisbett and Wilson demonstrated, they are structurally incapable of predicting accurately.

Rule 2: Ask about specifics in the past, not generics about the future. "Would you pay for this?" tells you nothing. "How are you solving this problem right now?" and "What does that solution cost you, in money, in time, in frustration?" tells you everything. If the person has never bothered to look for a solution, the problem isn't painful enough to support a business. If they've cobbled together a workaround involving spreadsheets, email reminders, and three separate apps, you've just found a hair-on-fire problem that they've already demonstrated willingness to invest effort in solving.

Rule 3: Talk less and listen more. The moment you start pitching, you've stopped researching. Fitzpatrick recommends withholding any product demo for at least fifteen minutes in a customer conversation, because the instant you show the product, the conversation shifts from "tell me about your problems" to "tell me what you think of this design." The first conversation reveals demand. The second collects opinions. And as we've established, opinions are the least reliable data source available.

The common thread across all three rules is the same insight that runs through Nisbett and Wilson's research: people cannot accurately predict their own future behavior, but they can accurately report their own past behavior. The Mom Test doesn't solve the say-do gap by asking better hypothetical questions. It solves it by refusing to ask hypothetical questions at all.

Watch the Mop: How Observation Beats Interrogation

In the late 1990s, Procter & Gamble was sitting on a problem that their surveys couldn't see. Their line of floor-cleaning products, various mop-and-bucket detergent systems, had generated stable revenue for years. Customer satisfaction surveys showed acceptable numbers. Nobody was complaining. If you'd relied on stated preferences alone, you'd have concluded that the floor-cleaning category was mature and settled.

Then P&G did something unusual. Instead of asking customers about floor cleaning, they went to customers' homes and watched them clean floors. This was one of the company's first forays into ethnographic research: the practice of observing behavior in natural settings rather than collecting self-reported data in artificial ones.

What the researchers observed was revealing in ways that no survey would have captured. First: the floors were already clean when the researchers arrived. Even though people knew the researchers were coming to watch them clean, they'd tidied up beforehand: a behavior so instinctive that it happened despite being counterproductive to the research. That alone told the team something about the emotional dimension of cleaning that a survey question ("How often do you clean your floors?") would have missed entirely.

Then the actual cleaning began, and the researchers noticed something the customers themselves had never articulated in any survey or focus group: people spent more time cleaning their mop than cleaning their floor. The process was cumbersome, sweep first to get the loose dirt, then fill a bucket, then mop, then wring out the dirty mop, then mop again, then wring again, then dump the water, then rinse the mop. The friction wasn't in any single step. It was in the accumulated weight of all of them. And because the friction was distributed across the entire process rather than concentrated in one dramatic pain point, customers had never thought to complain about it. They'd adapted. They'd accepted. And if you'd asked them "Are you satisfied with your mop?" they'd have said yes, because the question presupposed that mopping was the activity, and the mop performed the activity adequately.

The researchers asked a different question, though they didn't ask it verbally. They asked it by watching: What if a mop could attract dirt to itself?

The observation that people always swept before mopping revealed that the mop's fundamental problem wasn't cleaning power, it was that mops push liquid and dirt around the floor rather than capturing it. This insight led to the development of a disposable-pad system that used electrostatic charge to attract dust and debris without water. No bucket. No wringing. No two-step process.

The Swiffer launched in July 1999 and generated $100 million in sales in its first four months. Within a year, P&G had sold over 11.1 million starter kits in the United States alone. The product created an entirely new category in home cleaning: a category that no customer had asked for, because no customer had identified the problem.

This is the difference between stated-preference research and revealed-preference research at its most consequential. P&G didn't ask people what they wanted in a floor-cleaning product. They watched people clean floors. The behavior revealed the problem. The problem revealed the product. And the product generated a hundred million dollars in a quarter: which is approximately a hundred million dollars more than another mop-and-bucket survey would have produced.

The Question Behind the Question

Coca-Cola's 190,000 taste tests weren't bad research. They were precise answers to the wrong question. The question they asked, "Which flavor do you prefer?", had a clear, measurable, replicable answer. The question they needed to ask ("How would you feel if the Coke you've been drinking your entire life was gone forever?") was harder to operationalize, harder to quantify, and infinitely more important to the business outcome. They measured a sensory preference and missed an emotional identity.

This is the pattern that repeats across every market research failure worth studying. Sony's focus group accurately measured aesthetic enthusiasm and missed purchasing behavior. P&G's surveys accurately measured mop satisfaction and missed the entire cleaning workflow. Every startup that asks "Would you use this?" and hears "Definitely!" is collecting accurate data about politeness and inaccurate data about demand.

The fix isn't to stop doing research. The fix is to stop doing research that asks people to predict their own behavior, because fifty years of cognitive science, from Nisbett and Wilson's stockings to Samuelson's revealed preference theory, has demonstrated that people cannot do this reliably. They aren't lying to you. They're lying to themselves. Not out of malice. Out of architecture. The brain generates explanations for its own behavior the way the heart generates a pulse: automatically, continuously, and without any requirement that the explanations be true.

Your job as a founder is to design research that routes around this limitation. Ask about the past, not the future. Watch behavior, not surveys. Measure commitment, not enthusiasm. And when someone tells you they would definitely buy your product, remember the stockings on the rack, four identical pairs, and forty percent of people chose the one on the right, and every single one of them could tell you exactly why it was the best quality.

They were wrong. They were confident. And they had no idea.

If you're ready to build a research system that captures what customers actually do rather than what they say they'll do, The Launch System walks you through the complete validation process (from initial customer discovery through minimum viable product testing) using behavioral signals instead of opinion surveys. It includes the exact interview scripts, observation frameworks, and commitment-tracking templates that separate real demand from polite enthusiasm. Because in the end, the only market research that matters is the kind that accounts for the fact that your customers, all of them, every single one, are unreliable narrators of their own lives. And so are you.