Market research for startups: a checklist with cited sources

Start with the customer, not the market size

The most common mistake is opening with a giant number: the market is 50 billion dollars. That number is almost always irrelevant, because you will never serve all of it. Start instead with a specific person: who has this problem badly enough to pay, what do they do today to solve it, and how much does that cost them in time or money? Ten real conversations with that person tell you more than any report. Write down the exact words they use for the problem, because those words are your positioning and your ad copy later.

Size the market from the bottom up

Once you know the customer, size the market from the bottom up rather than the top down. Three numbers, multiplied:

1. How many of these specific customers exist in the places you can actually sell to.
2. What each would pay you per year, at a price you have some reason to believe.
3. What share you could realistically reach in the first year or two, given one channel and a small budget.

That produces a number you can defend line by line, and more usefully, a number you can act on. Top-down sizing does the opposite: it starts from an industry total and takes a percentage, so the answer is only ever as good as the percentage you invented. "We only need 1 percent of a 50 billion dollar market" is not a plan, because nothing in it explains how the first hundred customers hear about you.

Write each of the three numbers with the source next to it. The point is not formality. It is that in three months you will not remember which figures you looked up and which you assumed, and that distinction is the whole value of the exercise.

Where to get numbers you can actually cite

Most of what a first-time founder needs is free and public. These are the sources worth knowing, and all of them are checkable by anyone you later show the research to.

• County Business Patterns, from the US Census Bureau, counts establishments, employment and payroll by industry and county, annually since 1964. If your customer is a business, this is how you count how many exist and where.
• Census Business Builder wraps the same data in maps by business type, with demographics, income and consumer spending for an area. It is the fastest way to get a defensible local number.
• data.census.gov is the full catalog behind both, including the American Community Survey for population and income, and the Economic Census for industry revenue and business sizes.
• The Consumer Expenditure Surveys, from the Bureau of Labor Statistics, are the only federal household survey covering the complete range of what consumers spend, broken down by category and by household type. If your customer is a person, this is how you find out whether they already spend money in your category and how much.
• Google Trends shows direction rather than volume. It will not size a market, but it will tell you whether interest in a term is growing, flat, or was a spike two years ago.

Two habits make these worth more than the numbers themselves. Record the date you pulled a figure, because these series get revised. And prefer the primary source to the article quoting it: the article may have rounded, aged, or misread it, and you inherit the mistake with none of the context.

Map the competition honestly, including the boring ones

List the direct competitors, then keep going, because the one that actually beats you is rarely on the list.

The most dangerous competitor is the status quo: the spreadsheet, the group chat, the intern, the habit your customer already has and does not hate enough to change. It has no marketing budget and it wins constantly, because switching costs something and staying costs nothing.

For each real competitor, write down three things: what they charge, what they are genuinely good at, and the specific gap you would fill. Being honest about the second one is what makes the third one believable. A comparison where every rival is bad and you win on every row is not analysis, it is a pitch, and any customer who has used one of those products will stop reading.

Then apply the test that matters: can you name a gap that a customer would pay to close, in their words rather than yours? If you cannot, that is a finding rather than a failure, and it has cost you an afternoon instead of a year.

This is also where researching a business idea most often quietly changes it. Founders come in planning to beat an incumbent and leave having found an underserved slice that the incumbent does not want. That is a better outcome than the one you were looking for.

Check the boring blockers before you fall in love

Spend an hour on the unglamorous questions, because these are the ones that kill a business quietly rather than dramatically.

• Regulation and licensing. Does selling this require a license, a registration, or an inspection you had not budgeted for?
• Cost to acquire a customer. Through a channel you can actually use, not a hypothetical one. If you have no idea, that itself is the finding.
• Real unit costs. After payment fees, refunds, returns, support and delivery, not before.
• Who else has to say yes. In business sales especially, the person with the problem is often not the person with the budget.

A market can be large, real, growing, and still a bad business, if it costs more to win a customer than that customer is worth. That single comparison decides more outcomes than market size ever does, and it is the one founders leave until last.

The collection of founder post-mortems in CB Insights' analysis of why startups fail is worth an hour of your time here. Its recurring theme, in the founders' own words, is building something interesting to solve rather than something that served a market need. Almost every account reads like a problem that an afternoon of this kind of checking would have surfaced.

Label what you assumed, and keep it living

Research is not something you finish. It is a document that should get less wrong every week.

The discipline that makes it useful is small: mark every claim as either sourced or assumed. A sourced claim carries a link and a date. An assumed claim carries nothing, and that is fine, as long as it is labeled. What you must not do is let the two blend, because a document where everything sounds equally certain is a document nobody can act on, including you.

The labeling has a second use. Your list of unlabeled assumptions is your test queue, in priority order. The one that would hurt most if false is the next thing to check, which is exactly where validating a business idea picks up.

A plan that says plainly "we assumed this and have not checked it" is more trustworthy than one that states everything with the same flat confidence. Experienced readers know which claims at this stage cannot possibly be settled yet, and seeing them marked honestly is what makes them believe the rest.