Market Research Playbook

Market research proves, with evidence an investor can check, who has the problem, how many of them there are, what they use today and what they would pay. The seven steps below show how, with one example company getting ready for its seed pitch.

Many founders do market research twice: once, loosely, before building, and again in a hurry before an investor meeting. The second round is often where the gaps show. A market size built on one analyst headline, a survey with leading questions, or a competitor slide that leaves out the spreadsheet your customers use today will all be tested by an investor.

The method here is one you can run once, keep updating, and use for product and marketing decisions as well as the pitch.

Start from the decisions you need to make, not from the data you can find. Six questions cover most early-stage needs:

Questions 5 and 6 are best answered by testing with real offers, because what people say and what they do differ (Step 5 shows by how much).

Secondary research “uses existing data that has been collected by another entity”: government statistics, company filings, published studies. Primary research means “collecting new data for the purpose of answering a specific question”: your own interviews, surveys and tests^[1].

Use secondary research to count and map the market quickly and cheaply. Use primary research (Steps 3 to 5) to learn what the numbers cannot tell you: why customers buy, what they compare you with and what would make them switch. You need both. A market size with no customer evidence is a guess about demand; interviews with no market data cannot tell you whether the market is big enough.

Official statistics are free, documented and hard for an investor to dispute. For a US market, start here:

Check the data year before you quote a number. As of October 2026, the latest County Business Patterns data are for 2023, released in June 2025, and the Census Bureau has set no date for its 2024 data^[12]. Nonemployer Statistics now runs to 2024, released in September 2026^[13], and the 2022 Economic Census finished its scheduled releases in April 2026^[14].

Expect sources to disagree, and explain why. The Bureau of Labor Statistics’ QCEW counted 12.4 million establishments in the first quarter of 2026^[15]; its Handbook of Methods puts County Business Patterns at about 8 million: more than 6 million single-location businesses and 1.8 million establishments of multi-location ones. The programs draw on different records, state unemployment insurance files against the Census Bureau’s Business Register, and County Business Patterns leaves out most government employers, crop and animal production, rail transportation and the Postal Service^[16]. When two sources differ, say which you used and why. Investors may well check.

Fewer than most people think, if the group is well defined. Studies of similar people reach saturation, the point where new interviews add little, within about 9 to 17 interviews^[18]; in one study of 60 interviews it came within the first 12^[19]. Finding nearly every need takes more. Griffin and Hauser’s classic study, “The Voice of the Customer”, estimated that 20 to 30 interviews find 90% to 95% of customer needs, from 30 interviews about one product, coolers and other portable food carriers^[20]. The figure below puts these studies, and a 2024 review, on one scale.

In practice, that means:

• About 10 interviews per segment to find the main problems and patterns. Read patterns from the fifth, and stop early only when three in a row add nothing new (AAJ’s rule, the same one the Customer Interview Kit uses).
• 20 to 30 when you need a complete list of needs, for example before designing a product or a survey.
• More segments means more interviews. Saturation is per group, not overall.

Ask about what people did, not what they would do: the last time the problem came up, what they tried, what it cost, who was involved. Past behavior is evidence; opinions about a future product are not.

Use a survey after the interviews, to measure how common the things you heard are. Small wording choices change answers a lot. In Pew Research Center examples, support for military action fell by 25 points when one clause about casualties was added, and 58% chose an answer offered in a list against 35% who named it unprompted^[22].

Four rules from AAPOR and Pew Research Center guidance:

1. One idea per question. “Questions should be specific and ask only about one concept at a time”^[23].
2. Test it first. Pretest the questionnaire with a few people and fix what confuses them.
3. Don’t lead. Avoid wording that suggests the answer you want, and randomize answer order where you can.
4. Keep the measure stable. If you want to track change, ask the same question the same way each time.

Can you trust what people say they would buy or pay? Only partly. A study of 40 purchase-intention studies covering more than 65,000 consumers found an average correlation of 0.49 between what people said they would buy and what they bought. Intentions predicted better for existing products than new ones, for durable goods than everyday ones, and over short time horizons than long ones^[24]. New startup products are the hard case.

Price answers are inflated too. A 2020 meta-analysis of 77 studies of consumer goods found that people state a willingness to pay about 21% higher, on average, when the money is hypothetical than when it is real, and the gap is larger for higher-value products^[25]. Recent research suggests measuring willingness to pay against the option a buyer would otherwise choose: ask what price would make them equally happy to buy your product or that option, rather than asking for a price in a vacuum^[26].

So treat stated demand as an upper limit and look for behavior: pre-orders, paid pilots, letters of intent, or sign-ups on a page with a real price.

Start with what customers use today, which your interviews will reveal (Step 3), then research each alternative: its market share, strengths and weaknesses, barriers to entry and indirect competitors, as the SBA’s checklist suggests^[27]. My post on competitive analysis for startups sets out the full method, and the Competitor Research Grid is the worksheet.

Public company filings on SEC EDGAR are the richest free source on larger competitors: their segments, customers, pricing notes and stated risks. They also show how established companies size their own market.

What iRhythm’s IPO filing shows

iRhythm Technologies, which provides an ambulatory cardiac monitoring service, did both halves of this work in its 2016 IPO filing. It started from a CDC count of about 11 million US patients with a heart rhythm disorder, narrowed to an estimated 4.6 million diagnostic tests a year, valued them at what it said “we believe to be an existing $1.4 billion market opportunity”, and named Holter monitors and mobile cardiac telemetry monitors as the alternatives it competes with^[28].

The lesson for a seed pitch, marked on the filing above: start from a count you can source, narrow it to what is bought today, say which figures are your own belief, and name the established way of doing the job as the competition. The filing shows what iRhythm wrote, not how it made its choices.

Keep a single research file with three parts:

• A sources log. Every number, its source and year, and whether it is a public count or an assumption, as in Step 2, plus a confidence level.
• A segment view. For each segment: how many there are, what they need, what they use today and what they might pay. The Segment Scorecard is built for this.
• Decisions and open questions. What you decided, on what evidence, and what you still need to check.

Investors use this research to judge whether the company can get big. In a survey of 885 venture capitalists at 681 firms, 68% named the market as an important factor in their decisions, behind the team (95%), business model (83%) and product (74%)^[29]. A 2025 study of more than 8,000 deals seen by one early-stage investor found that the team best explained which startups raised at least $1 million, but market and product had more explanatory power for which raised more than $10 million, and for longer-term outcomes^[30].

They also check. A study of 722 funding requests to one US venture capital firm found that business plans were only weakly linked to funding decisions; the authors conclude that investors learn this information through other channels^[31]. A research file lets you answer their questions in minutes.

What changes by stage?

What mistakes should I avoid?

• Starting with a big analyst number. It sizes someone else’s definition of the market; count your own customers (Step 2).
• Asking “would you buy this?” Ask what people did last time (Steps 3 and 4).
• One survey, no interviews. You measure answers to questions you guessed; interview first (Step 3).
• Treating stated price as real. It runs high; get a paid commitment (Step 5).
• Ignoring the status quo. The outside lab, manual workarounds and doing nothing are competitors (Step 6).
• No sources log. You cannot defend a number you cannot trace (Steps 2 and 7).