Learn · customer-understanding

By Fieldwork Research Team · Published 2026-07-23

How to understand what customers really think

Most businesses have more customer data than they know what to do with. They have NPS scores, satisfaction ratings, support ticket summaries, app reviews, and survey responses. What they often don't have is a clear understanding of what customers actually think, as distinct from what customers say when asked a direct question in a structured context.

The gap between the two is where most product, marketing, and service decisions go wrong.

Why customers don't tell you what they think

Customers are not withholding. They are operating under a set of conditions that make honest, specific expression of their actual experience genuinely difficult.

They tell you what they think you want to hear. Social pressure shapes answers to direct questions. A customer who rates a product 7 out of 10 and says "it's pretty good" in a satisfaction survey may privately find it frustrating in specific ways they don't articulate because they don't want to be critical, don't want to seem difficult, or don't think their complaint will be taken seriously.

They can't always articulate what they experience. Much of the experience of using a product or service is pre-conscious: habits, frictions, small moments of confusion or delight that the customer processes but doesn't consciously attend to. Asked directly about their experience, customers produce a summary that reflects their overall impression rather than the specific moments that actually shaped it.

They answer the question as asked, not the question you needed to ask. If you ask "how would you rate your onboarding experience?" you get a rating of the onboarding experience. You don't get the information that would actually be useful: what the customer was trying to accomplish, where they hit friction, what they assumed that turned out to be wrong, and what would have made the experience faster or clearer.

They don't know what they don't know. A customer who never discovered that your product had a feature that would have transformed their workflow doesn't miss that feature. They don't know to miss it. A survey cannot surface a gap in the customer's knowledge about your product, but a well-designed interview can.

The conditions that produce honest answers

Getting to what customers really think requires creating conditions where honest, specific expression is possible. Those conditions are not complicated, but they are different from the conditions in most feedback collection contexts.

Low stakes. When a customer feels that their answer will have consequences for them personally, either in the form of social judgment or in terms of what the company does next, they moderate their response. The most honest answers come from conversations where the customer feels that their job is simply to share their experience, not to evaluate or decide anything.

Enough time. A survey that takes three minutes to complete cannot capture the specific details of a complex experience. An interview that allows a customer 45 minutes to tell a story, with follow-up questions that probe specific moments, captures orders of magnitude more useful information.

Questions that invite stories rather than evaluations. "Walk me through what happened when you first logged in" invites a story. "How would you rate your initial experience?" invites an evaluation. Stories contain specifics that are analysable. Evaluations contain impressions that are hard to act on.

Permission to be critical. Customers need explicit permission to share negative experiences. The framing matters. "We're trying to understand where things don't work as well as they could" invites more honest negative feedback than "we'd love to hear your thoughts on your experience."

An interviewer who is genuinely curious. Customers can tell the difference between someone who is going through motions and someone who actually wants to understand. Genuine curiosity produces more open, more detailed responses. This is true whether the interviewer is a human researcher or an AI designed to probe rather than accept surface-level answers.

The questions that surface real thinking

The specific questions that reliably surface what customers actually think are not the ones that feel most intuitive.

"Walk me through what happened" beats "how was your experience?" The narrative question produces a chronological account full of specific details. The evaluation question produces an impression. Details are analysable. Impressions are not.

"What were you hoping would happen?" beats "were you satisfied?" Surfacing the expectation before the outcome reveals the gap between what the customer anticipated and what they got. That gap is where the insight lives.

"What did you do next?" beats "why did you do that?" Asking why invites rationalisation. Asking what happened next reveals the actual sequence of events, from which the reasoning can often be inferred more reliably than from a direct why question.

"What would have to be different for you to...?" beats "would you recommend us?" The hypothetical surfaces what is preventing the desired behaviour. The recommendation question produces a number that tells you very little about what to do with it.

"Is there anything you expected me to ask but didn't?" at the end of every conversation. Customers often hold their most important insight in reserve, waiting for the right question that never comes. This question is the right question.

The difference between what customers say and what they do

One of the most reliable findings in consumer research is that what customers say they do and what they actually do are often different. Not because customers lie, but because self-reported behaviour is reconstructed from memory and shaped by how the question is asked.

A customer who says they check their email "a few times a day" may check it 40 times. A customer who says price "doesn't really matter" to them may be highly price-sensitive in practice. A customer who says they "always read the reviews" may buy impulsively and construct a narrative about review-reading after the fact.

This gap is one of the strongest arguments for combining what customers tell you with data about what they actually do. Analytics tell you what happened. Interviews tell you what customers think happened, and why they think they did what they did. Both perspectives are useful. Neither alone is sufficient.

The most important implication: when customers tell you they would do something, treat that as a hypothesis to test, not a prediction to act on. When customers tell you they have done something, probe the specific instance rather than the general claim.

What this looks like in practice

A consumer app with strong acquisition and poor day-30 retention runs a survey asking users who haven't returned why they stopped using the app. The most common responses are "I forgot about it" (38%), "I didn't need it anymore" (27%), and "I found an alternative" (18%). The team considers a push notification strategy to address the forgetting, a re-engagement campaign for the "didn't need it" segment, and a competitive analysis for the switchers.

Before committing to these initiatives, a researcher runs 12 interviews with users who installed the app and didn't return after the first week.

The interviews reveal something the survey categories couldn't capture. Almost every participant describes a similar experience in their first session: they opened the app, looked at the main screen, and couldn't immediately see how to accomplish what they'd downloaded the app to do. Most spent a few minutes exploring, found something that seemed relevant, tried it, and got a result that wasn't quite right. They didn't fail catastrophically. The app worked, technically. But it didn't work in the way they expected, and the gap between what they expected and what happened was just wide enough to make reopening the app feel like a low-priority task.

None of this appears in a survey response. It doesn't map to "forgot about it" or "didn't need it" or "found an alternative." It is a first-session orientation problem: users don't know what to try first and the result of their first attempt doesn't clearly demonstrate value.

The fix is a structured first-session experience that routes users to a single high-value action based on what they said they wanted to accomplish when they signed up. That route, and the result of following it, is what should happen in the first two minutes of every new user's experience.

Day-30 retention improves by 11 percentage points in the two months after the change. The push notification strategy, which the team had been planning to implement, is deprioritised because the retention problem is not a memory problem. It was always an orientation problem that the survey couldn't name.

Frequently asked questions

What is the best way to collect customer feedback?

The best method depends on what you need to know. For understanding what customers think about a specific experience in depth, qualitative interviews are the most powerful tool. For measuring sentiment at scale, surveys and NPS are appropriate. For understanding what customers actually do, behavioural analytics provide the most reliable data. Most businesses need all three, used in combination, because they answer different questions.

How do you get customers to give you honest feedback?

Create conditions where honesty is safe and expected. Frame conversations around understanding what could be better rather than evaluating what exists. Give customers enough time to tell a story rather than rate an experience. Ask questions that invite specific descriptions rather than general impressions. And listen to the answer without visibly reacting, so that customers don't moderate what they say based on how you're responding.

How is this different from reading reviews?

Reviews are customer expressions in a public, evaluative context. They are shaped by the norms of the platform, the customer's emotional state at the moment of writing, and the desire to communicate to other potential customers rather than to the business. They capture extreme experiences (very positive or very negative) more than typical ones. Direct conversations capture the full range of experience and can probe specific moments that reviews mention only in passing.

Can AI help understand what customers really think?

AI-conducted interviews, when designed with the right questions and the ability to probe vague or surface-level answers, can surface more honest and specific customer thinking than surveys at a scale that human moderation cannot match. The AI doesn't bring social expectations to the conversation in the same way a human interviewer does, which some customers find makes it easier to be direct. The analysis still requires human interpretation.

How many customers do you need to speak with to understand what they really think?

For a well-scoped question, 8 to 15 conversations will typically surface the primary patterns in what customers think and experience. The goal is not a representative sample in the statistical sense but sufficient diversity and volume to distinguish genuine patterns from individual perspectives. More conversations add confidence in the patterns identified. They rarely change what those patterns are.

What do you do with customer insights once you have them?

Map each insight to a specific decision or action. Customer insights that don't connect to a decision are interesting observations, not actionable research. The test of a good customer insight is whether it changes something: a product decision, a marketing message, a service process, a pricing structure. If the insight doesn't clearly point to a change, either the research question was too abstract or the insight needs further development before it can be acted on.

Related on Fieldwork

Last updated: 2026-07-23