Why "talk to your customers" is bad advice

I say this as someone whose primary professional skill, for years, was talking to customers. As a consumer researcher and later in customer success, the work was not having conversations. The work was knowing which conversations to have, with whom, and what to do with what came back.

"Just talk to your customers" collapses three separate questions into one instruction. Each has a wrong answer that produces confident, wrong conclusions. In a year-1 value creation window, confident wrong conclusions are more expensive than no conclusions at all.

Which customers are representative

The customers who answer the call are not necessarily the ones whose behaviour explains the P&L. They are the ones with time, opinions, and a relationship that makes them want to talk.

That is a self-selecting sample with a known bias. Survey methodology has documented this for decades: respondents with affinity for the brand systematically overrepresent themselves, while the silent majority who pay quietly, churn quietly, or buy from a competitor systematically underrepresent themselves (Groves, 2006). No statistical adjustment after the fact can fully correct for unknown non-response bias.

The discipline upstream of any conversation is defining the economic ICP. Which segment generates the margin the VCP depends on. Which sub-segment has the highest LTV-to-CAC ratio. Which micro-segment is at risk. Talking to a customer outside that frame produces information, not signal.

A portco CEO who has had thirty customer conversations and is confident he understands his market has usually had thirty conversations with the wrong distribution of customers.

Which of their stated needs is valid

Customers report needs they cannot reliably introspect on.

Nisbett and Wilson (1977), now cited over 14,000 times, established that people have little direct introspective access to the cognitive processes driving their decisions. When asked to explain their preferences, they generate plausible causal stories after the fact and cannot distinguish those stories from genuine introspection. The choice blindness paradigm has since extended this. In the original 2005 study, 87 percent of participants failed to detect the manipulation and produced fluent introspective justifications for choices they had not actually made (Johansson, Hall, Sikström, and Olsson, 2005).

This is not a flaw in the customer. It is a property of human cognition.

The B2B implication is sharper than the psychology alone. Stated-preference research consistently finds respondents overstate willingness to pay relative to revealed behaviour. Earlier meta-analyses across mixed contexts found a median hypothetical-to-real ratio of 1.35, with substantial variance across studies (Murphy, Allen, Stevens, and Weatherhead, 2005). The most recent meta-analysis specifically for consumer goods, drawing on 77 studies and roughly 45,000 observations, reports an average hypothetical bias of 21 percent (Schmidt and Bijmolt, 2020). What the customer says they want is data. It is not the answer.

The job is to separate three layers: what the customer says, what the customer means, and what the customer would actually do under economic friction. Stated and revealed preference are different variables (Samuelson, 1938). Most year-1 strategy decisions made off customer interviews use stated preference because revealed preference is harder to extract.

The diagnostic question. Not "what do you want?" but "walk me through the last time you tried to solve this. What did you do, what didn't work, where did you give up?"

The first conversation produces wishlist. The second produces mechanism. This is the jobs-to-be-done interview methodology, calibrated around the moment of struggle rather than stated preference (Christensen, Hall, Dillon, and Duncan, 2016). The milkshake study is the canonical example: customers asked what they wanted produced flavour feedback. Customers asked to describe the situation in which they bought it revealed they were hiring it for the morning commute, in competition with bagels and bananas.

How to translate the need into an offer

The line most commonly cited here is attributed to Henry Ford: if he had asked customers what they wanted, they would have said faster horses. There is no evidence Ford ever said this. The earliest attribution traces to a 2001 letter in Marketing Week, decades after his death (Vlaskovits, 2011). The line is apocryphal. The point is not.

The customer can describe the constraint accurately. They cannot reliably describe the solution.

A faster horse is one articulation of the underlying constraint: personal mobility without the cost, care, and infrastructure of keeping a live animal. A car is the translation of that constraint into a product. The customer is the authority on the constraint. The operator is the authority on the translation.

The translation step is where most "we listened to our customers" initiatives fail. The portco hears a stated solution, builds the stated solution, and ships something the customer asked for and does not buy. The customer was answering a different question than the one being asked, and no one in the room had the methodology to notice.

What this means for an operator

"Talk to your customers" is correct as a directional instinct and dangerous as a methodology.

The operating question for a year-1 VCP is not whether the portco is doing customer conversations. It is whether those conversations are sampling the right ICP, distinguishing stated from revealed preference, and translating constraint into offer rather than transcribing wishlist into roadmap.

When those three are missing, the initiative produces a deck, a few quotes, and a confident misreading of the market. When they are present, the same input produces a positioning shift that moves the EBITDA bridge.

The advice is not wrong. The advice is incomplete. The incompleteness is what costs the portco the year.