How to Conduct Market Research for New Business: 7 Steps

You have an idea. You think it's good. You've told a few friends and they nodded enthusiastically, which is the worst possible validation mechanism ever invented, because friends are polite and startup ideas are not.

So how do you actually conduct market research for a new business without burning three months and a few thousand dollars on a report nobody reads? I've done this for solo projects, small agencies, and one product that I killed after six weeks of research that cost me about $400 total. That kill saved me an estimated $15,000 in development. Not a hero story—just the math of doing the work before the build.

Here's what I've learned: market research for a new business is not a report. It's a decision-making process. The output isn't a PDF. The output is whether you build, pivot, or walk away.

How to conduct market research that actually changes your decision

The framework I use has five steps. Most guides bury it under jargon, so here it is stripped down:

1. Write the decision you're trying to make. "Should I build a $49/month tool for freelance designers?" is a decision. "Understand the market" is not.
2. List your assumptions—the ones that, if wrong, kill the business.
3. Pick the cheapest method that can falsify each assumption.
4. Collect data from primary and secondary sources.
5. Decide. Then keep collecting.

That first step is where 90% of founders skip ahead. I did too, the first time. I ran a survey with 47 questions, got 112 responses, and had absolutely no idea what to do with the data because I'd never defined what "yes, build it" looked like.

Primary vs. secondary research: which one first?

Secondary research is data someone else already collected: industry reports, government statistics, competitor pricing pages, Reddit threads. It's cheap and fast. Primary research is data you gather yourself: interviews, surveys, observation.

Do secondary first. It costs you a weekend. Primary research is expensive in time, and you don't want to waste interviews asking questions a public filing already answered.

How much does this actually cost?

Numbers nobody puts in the brochure. Based on what I've spent and what I've seen others spend:

Method	Realistic cost	Time	Best for
Desk research (reports, filings, forums)	$0-50	1-2 days	Sizing the market, spotting competitors
Customer interviews (10-20 calls)	$0-300 in incentives	2-3 weeks	Understanding pain points
Online survey (100-200 responses)	$150-800	1-2 weeks	Quantifying a pattern you already suspect
Focus group (professional)	$3,000-8,000	3-4 weeks	Testing messaging at scale—rarely worth it for a solo founder
Landing page + ad test	$200-1,000	1-2 weeks	Testing demand before you build

For most new businesses under $10,000 in starting capital, the top three rows are your entire playbook. Focus groups are for consumer brands with a marketing department.

What is the 3-3-3 rule for marketing?

The 3-3-3 rule is a lightweight planning heuristic: three audience segments, three core messages, and three marketing channels. The idea is that spreading yourself across ten segments and eight channels guarantees you do all of them badly.

It's not a formal law. It's a constraint tool. And it maps well onto research: when you finish your interviews, you should be able to name three segments. If you can't, your sample was too broad or too small.

The segments must share a question

Here's the version I actually use: for each of your three segments, you need to answer three questions. What job are they hiring your product to do? What do they use today? And what would make them switch?

If two of your segments give the same answer to all three questions, you don't have three segments. You have one segment and a marketing wish.

What are the 5 C's of marketing analysis?

The 5 C's are Company, Customers, Competitors, Collaborators, and Context. It's a classic framework from marketing strategy courses, and it's genuinely useful as a pre-research checklist—it forces you to notice what you're ignoring.

• Company: your actual capabilities, not your aspirations.
• Customers: who buys, why, and what they currently spend money on.
• Competitors: including the ones who aren't direct competitors—a spreadsheet is competition for a project management tool.
• Collaborators: partners, distributors, platforms you depend on.
• Context: regulation, economics, tech shifts.

The framework's weakness is the "Context" bucket, which tends to become a dumping ground. Keep it narrow: what changed in the last 18 months that makes your timing good or bad?

Can ChatGPT do market research?

Partly. And this is where I'll be blunt: it can do the desk research layer, and it cannot do the primary layer.

What it handles well: summarizing patterns across public sources, generating interview question drafts, clustering open-text survey responses, drafting competitor comparison tables you then verify by hand.

What it fails at: knowing your actual customers, telling you what people who aren't online think, and avoiding the confident fabrication of plausible-sounding claims. If you ask it to estimate your market size, it will produce a number. That number is a guess wearing a suit.

The honest workflow

Use an LLM to draft your interview script, then cut half the questions. Use it to summarize secondary sources you've already read. Use it to tag 200 survey comments by theme. Never let it substitute for twenty conversations with people who have the problem you're solving.

What business will boom in 2026?

Nobody knows. Anyone who tells you otherwise is selling a course. But you can watch structural conditions rather than forecasts. Aging demographics in most developed economies keep pushing demand toward health services, care logistics, and accessibility tooling. Regulatory pressure around data and AI is creating compliance and audit work. Energy retrofitting has decades of runway because buildings don't replace themselves.

The pattern I trust more than any trend list: businesses that solve a boring, expensive, recurring problem for another business tend to survive downturns. Consumer fads don't. Research the boring stuff.

The bias nobody warns you about

Your sample size is wrong in the way that matters. Most founders obsess over how many people answered. The real problem is who answered and whether they told you the truth.

Social desirability bias is the big one. Ask someone "would you pay $30 a month for this?" and they'll say yes about 40% more often than they'd say it with a credit card in hand. I learned this the hard way after a survey where 61% of respondents said they'd pay for a feature—and 4% actually did when I launched it.

The fix: stop asking about the future. Ask about the past. "When did you last pay for something to solve this?" not "Would you pay?" Past behavior is data. Future intentions are fiction.

Correlation is not causation

If your survey shows that 70% of your target customers use a Mac, that does not mean Mac users buy your product. It means your sample skews toward Mac users. Note it. Don't build a strategy on it.

Building your own research plan (a template, not a PDF)

You don't need a 40-page document. You need one page with these rows:

1. Decision: what will you do differently based on the answer?
2. Assumption: the belief you're testing. One per row.
3. Method: interview, survey, desk research, or landing page test.
4. Success threshold: what number means "go"?
5. Kill threshold: what number means "stop"?
6. Deadline: two to four weeks out.

That last column is the one people skip. Without a deadline, research becomes a hobby. Six weeks max for the first wave, then decide.

If you want a real-world example of what this looks like filled in: I built one for a subscription box targeting home bakers. Decision—launch or not. Assumption—people will pay $35/month for curated ingredients. Method—12 phone interviews plus a $300 ad test. Success threshold—4 interviews confirmed they'd bought a similar box before, and the ad test beat a 2.5% click-to-email rate. Kill threshold—fewer than 3 confirmed purchases or a rate below 1.2%. Result: the ad test hit 1.8%, below target. I killed it in week four.

That's market research. Not a report. A decision, with a receipt.