Market Research: How to Know What Customers Actually Want

Most market research is done backwards. A team decides what they want to build, then designs a survey that confirms it's a good idea, then presents the results as evidence. This is not research. It is expensive reassurance.

Real market research starts with a decision you need to make and works backwards to the minimum information required to make it confidently. That discipline — decision first, data second — is what separates research that changes behaviour from research that fills slide decks.

Start With the Decision

Before collecting a single data point, write down the decision your research will inform. Not "we want to understand our customers" — that is a project description, not a decision. A decision looks like this:

• Should we launch in the SMB segment or the enterprise segment first?
• Should the product be priced at £29/month or £49/month?
• Should we build the integration with Salesforce or the integration with HubSpot?

Once the decision is explicit, the research almost designs itself. You know exactly what you need to find out, you know what evidence would change your mind, and you know when you have enough information to stop.

Primary vs Secondary Research

Secondary research uses data someone else collected — industry reports, academic studies, government statistics, competitor filings, review site aggregates. It is fast, cheap, and often sufficient for understanding market size, growth rates, and broad trends. Start here before spending anything on primary research.

Primary research is data you collect yourself — interviews, surveys, usability tests, experiments. It is slow and expensive but directly answers your specific question about your specific customers. You commission primary research when secondary data doesn't exist, doesn't fit your context, or isn't granular enough to be actionable.

Qualitative vs Quantitative

These are not competing approaches — they answer different questions and are most powerful when combined.

Qualitative research (interviews, focus groups, ethnographic observation) is exploratory. It tells you why people behave the way they do, what language they use to describe their problems, and what tradeoffs they care about. You can't generalise from 10 interviews, but you can discover hypotheses you didn't know to test.

Quantitative research (surveys, analytics, A/B tests) is confirmatory. It tells you how many people hold a given view or exhibit a given behaviour, with statistical confidence. It validates or refutes hypotheses generated from qualitative work.

The classic mistake is running quantitative research before you understand the question well enough to write good survey items. A survey is only as good as the hypotheses embedded in its questions — if those hypotheses are wrong, the data is misleading regardless of sample size.

Jobs to Be Done

The most useful reframe in customer research comes from Clayton Christensen's Jobs to Be Done (JTBD) framework. Instead of asking "who is my customer?" (demographic segmentation), ask "what job is the customer hiring this product to do?"

A milkshake is hired to make a commute less boring. A drill is hired to hang a picture. A business school is hired to get access to a network, not to deliver education. The job reveals the real competition (the milkshake competes with bananas and bagels, not other milkshakes) and the real value proposition.

JTBD research focuses interviews on three moments:

1. The trigger — what happened that made you start looking for a solution?
2. The switch — why did you choose this over what you were doing before?
3. The anxiety — what nearly stopped you from switching?

These three questions generate more actionable insight than any satisfaction survey.

Survey Design Principles

When you do need a survey, four principles reduce the risk of garbage data:

One question per question. "How satisfied are you with our product's speed and reliability?" is two questions. Respondents who like speed but hate reliability have no good answer. Separate every compound question.

Avoid leading questions. "How much do you enjoy our new dashboard?" presupposes enjoyment. "How would you describe your experience with the new dashboard?" does not.

Use behavioural anchors, not attitude scales. "How likely are you to recommend us?" (1–10) is fine for tracking NPS trends. But "In the last 3 months, how many times have you recommended us to someone?" gives you revealed behaviour, which is more reliable than stated intention.

Test for price with Van Westendorp, not direct questions. Asking "how much would you pay for this?" produces anchored, socially acceptable answers. One approach is the Van Westendorp Price Sensitivity Meter, which asks four questions about acceptable and unacceptable price points, triangulating a realistic acceptable range.

The four questions map directly to price boundaries:

Plotting cumulative response curves for all four questions reveals the acceptable price range (PMC to PME) and the optimal price point (OPP) — where equal proportions say the price is too cheap versus too expensive. This gives you a defensible price range grounded in actual customer psychology, not guesswork.

Competitive Research

Understanding competitors is a subset of market research that is systematically underinvested. Most teams do a feature comparison matrix and stop. More useful sources:

• Review sites (G2, Trustpilot, App Store): read the 3-star reviews, not the 5-star ones. The 3-star reviews name specific frustrations from people who still like the product — these are unmet needs waiting to be solved.
• Job postings: a competitor hiring heavily for sales engineers in a new vertical signals where they're expanding. A sudden drop in hiring signals resource constraints.
• Pricing page changes: public SaaS pricing is tracked by tools like PricingSaaS. Moves signal strategy shifts.
• Churned customer interviews: if you can talk to customers who left a competitor to come to you, or who left you for a competitor, these conversations are the highest signal research you can do.

Common Traps

Confirmation bias is the dominant failure mode. Researchers design studies that will validate their hypothesis, interpret ambiguous findings favourably, and dismiss contradicting evidence. The antidote is to write down before the research what result would cause you to kill the project — and then actually kill it if that result comes back.

Customer proximity bias affects founders and product managers who talk to the same ten customers repeatedly. These customers are articulate, engaged, and atypical. Their feedback is valuable but systematically unrepresentative of the median customer.

Analysis paralysis is the opposite problem: commissioning more research to delay a decision. Research reduces uncertainty; it does not eliminate it. At some point you have enough information to make a good decision and additional research is just procrastination with data.