Market Research Methods for Startups on a Zero Budget

The market research methods that work for a startup with no budget are five: customer interviews, desk research on existing data, competitor and audience mapping, small-sample surveys, and fake-door tests. Run them in that order, and stop when the evidence stops changing your mind. Most founders do the opposite. They send a survey to friends first, get 40 polite answers, and call it validation.

This guide ranks the methods by what they cost, what they can prove, and where they mislead you. It ends with a one-week sequence you can run alone.

Why do so many founders research the wrong thing?

Sit with a handful of failed projects and a pattern shows up. The founder researched the solution ("would you use an app that does X?") instead of the problem ("what did you do the last time this went wrong?"). The first question invites compliments. The second invites facts.

The scale of the damage is documented. CB Insights' analysis of 431 VC-backed companies that shut down since 2023 found poor product-market fit cited in 43% of cases. That is a study of venture-backed companies, so a bootstrapped side project will not map one-to-one. The direction still holds: demand is the assumption people check last and should check first.

So before choosing a method, write down the one hypothesis that would kill the idea if it were false. Not three. One. Every method below is a way to attack that sentence.

Primary or secondary: which comes first?

Primary research means you collect the data yourself: interviews, surveys, tests. Secondary research means you read what already exists: reports, public statistics, competitor sites, forum threads.

Crunchbase's overview of market research methods makes the same recommendation most practitioners reach on their own: start with secondary research, because it is cheaper and tells you which primary questions are worth asking. Two evenings of desk research can rule out an idea that would have cost you three weeks of interviews.

The exception is when no data exists, which is common for a new niche. Then secondary research gives you adjacent markets, and you go to primary research sooner. Fine. Just know that you are guessing at the edges.

Which market research methods are worth your time?

Here is the short version, with the verdict first.

• Customer interviews: highest signal, zero cost, uncomfortable. Do these first.

• Desk research: cheap, fast, good at killing ideas. Do these in parallel.

• Competitor and audience mapping: tells you where attention already sits.

• Surveys and fake-door tests: useful only after you know what to ask and who to ask.

Focus groups are missing from that list on purpose. Eight strangers in a room produce group opinion, and group opinion is a poor predictor of what a single person pays for. Skip them unless you already have a customer base to recruit from.

How to run customer interviews that produce facts

Interviews are the method that most changes founders' minds, and the one most often done badly. Fifteen to twenty conversations with people who actually have the problem will teach you more than any survey of 300 strangers.

Three rules keep them honest. Ask about the past, not the future: "Walk me through the last time you did this" beats "Would you pay for this?" every time. Never describe your idea until the final five minutes, and if the person has not mentioned the problem unprompted by then, that is your answer. And ask what they currently do about it, including the ugly spreadsheet and the paid workaround. A person with a duct-tape solution has a problem. A person with no workaround often has a preference.

Finding people is the hard part. Go where they already complain: niche forums, subreddits, Slack and Discord communities, local meetups. In the Mountain West, that can mean outfitters, ranchers, or small manufacturers who will not answer a cold LinkedIn message but will talk to you at a trade event. Offer 20 minutes and a plain explanation of why you are asking. Do not pitch.

Record the calls, with permission. You will miss half of what is said while you are thinking about your next question, and a transcript lets you search for repeated words later. An AI notetaker handles that part cheaply.

The pattern to look for is repetition. If the same complaint, in nearly the same words, appears in 6 of 15 calls, you have a signal. If every person describes a different problem, you have not found a market yet. That is not a failure. That is a data point.

What can desk research actually tell you?

Desk research answers three questions: how big is the space, who already serves it, and how are people already talking about it. It cannot tell you whether your version will win. It can tell you whether the arena exists.

For market size, skip the paid reports. Work bottom-up instead: how many potential customers exist, what could each reasonably pay per year, and what share you could plausibly reach in the first two years. Public sources such as census data, industry association pages, and government statistics usually supply the first number. A rough bottom-up estimate you can defend beats a top-down figure copied from a paywalled report.

Traffic and audience tools help here too. A platform like Similarweb estimates how much traffic the top players in a space actually get and where it comes from, which sizes a market by proxy. Treat the numbers as rough, since they are modeled estimates, not measurements. They are good enough to tell a $2 million niche from a $200 million one.

One more source people ignore: reviews of existing products. One-star and three-star reviews on competitors are free customer interviews. Read fifty of them and note the recurring complaints. That list is often a better product brief than anything you would write from scratch.

Where does your audience already spend attention?

Competitor mapping is not a list of company names. It is a map of where your future customers already look for answers, and what they accept as a solution today.

Build it in a simple table: five or six competitors or substitutes, what each charges, who each targets, and the single complaint customers repeat. Include the substitutes nobody calls competitors. For a scheduling tool, the real competitor may be a shared Google Calendar and a group text.

Audience research goes one level up. Instead of asking who sells to these people, ask which podcasts, newsletters, accounts, and communities they follow. That tells you both whether the audience is reachable and what it will cost you to reach it. If you cannot find three places where your customers gather, selling to them will be expensive no matter how good the product is.

Be skeptical of one thing here. A crowded market is not a warning sign by itself. It often means demand is proven. An empty market is the one that should worry you, because "nobody is doing this" is more often a sign of no buyers than of a clever gap.

When is a survey worth sending?

Late, and small. A survey is good at measuring how common something is, and bad at discovering what that something is. If you do not yet know the top five problems in your niche, a survey cannot tell you. It can only repeat your own assumptions back to you in percentages.

So run interviews first, then use what you heard to write closed questions: "Which of these have you dealt with in the last 90 days?" with options taken from real calls. Keep it under ten questions. Crunchbase cites SurveyMonkey data showing that 40-question questionnaires get a response rate about 10% lower than 10-question ones, and in a cold niche audience the gap is usually worse.

Who you send it to matters more than how many answer. Fifty responses from people who match your customer profile are worth more than 500 from a general panel. If you are tempted to send it to friends and family, do not. They will answer to be kind.

How do you test demand without building anything?

A fake-door test measures behavior instead of opinion. You describe the product on a simple page, add a clear call to action, and count what people do. The action can be joining a waitlist, booking a call, or paying a small deposit. The further down that ladder someone goes, the stronger the signal.

Set the bar before you launch. Write down, for example, that 30 visitors from a targeted community and at least 3 signups mean you continue, and anything under 1 signup means you change angle. Without a threshold written in advance, every result looks like a small win.

Two cautions. A free signup is cheap, so a waitlist alone proves interest, not demand. And traffic you buy from the wrong audience will give you clean-looking, meaningless numbers. Send targeted visitors only: the community you found during interviews, a niche newsletter, or a direct message list.

If you want a faster read on whether an idea is worth this effort at all, a structured check of the main assumptions can help you decide what to test first. It will not replace a conversation with a real customer. It will tell you which assumption is weakest.

Can AI do your market research for you?

Partly. AI tools are good at the desk-research layer: summarizing a competitor's pricing page, clustering review complaints, drafting interview questions, or producing a first-pass list of adjacent markets. They are poor at the parts that matter most, because they cannot talk to a customer and they will state an invented statistic with confidence.

Use them to speed up reading, not to replace evidence. Ask for sources, open every link, and discard any number you cannot trace. A research assistant that cannot show its work is a liability.

The rule is simple. If a claim will influence a decision about months of your time, a human source must back it up.

A seven-day sequence you can run alone

Here is how the pieces fit when you have a day job and no budget.

On days one and two, write your single killer hypothesis and do the desk research: market size estimate, five competitors, fifty competitor reviews. On day three, build the interview list of 20 people and send the first requests. On days four and five, run the first interviews and tag the repeated phrases. On day six, update the hypothesis based on what you heard, and draft a ten-question survey or a one-page fake door. On day seven, decide: continue, change angle, or stop.

Stopping is a legitimate result. A week spent killing an idea costs you a week. A year spent building it costs you a year.

What would you test first?

Look at your idea and finish this sentence: "This only works if ___ is true." Whatever fills that blank is your first research question, and probably your first interview. Which method would get you the answer fastest, and what result would make you stop?