Market research in a day, with sources you can click
Almost every guide to market research tells you to "identify your target market" and then stops, which is exactly where the difficulty starts. This one names the datasets, the table numbers and the searches. Five questions, an afternoon at a desk, a week of conversations, and a finished report for a real example business with a link under every figure — including the two figures that are guesses.
Market research has one job for a small business: to change a decision you are about to make. Not to produce a document. If you already know you're opening the shop, and no number would stop you, the research is theatre and the afternoon is better spent on the lease. The useful version starts the other way round — write down the decision first, then the answer that would change your mind, then go and find out which one is true.
There are two ways to get this wrong. The first is doing none, and pricing a service at what feels reasonable when six competitors within two miles publish their rates on a page anyone can read. The second is doing the kind that can't change anything: forty pages, a market sized in billions, and not one sentence about whether the person three streets away will pay you sixty dollars on Saturday.
The five questions that actually decide something
A textbook lists twenty. A small business needs five, and the fifth is worth more than the other four together. Notice what a good answer looks like in each case — it's rarely a single number, and it's never a forecast.
- 1 · How many, how often
- How many people or businesses within reach could plausibly buy, and how many times a year. A good answer is a range with its assumptions named out loud — "between 8,000 and 20,000 paid jobs a year in this county, if one in five households has a bike and one in five of those pays someone to fix it".
- 2 · Who already sells to them
- A count and a list of names. If the honest answer is nobody, treat that as bad news until proven otherwise — an empty market is usually one that doesn't pay, not one nobody noticed. Count the substitute (doing it themselves, or not bothering) as a competitor, because it usually wins.
- 3 · What the market actually pays
- The published range, from competitors' own price lists, never from asking people what they'd pay. A low, a high, what's included at each end — plus the ones who publish nothing, because a business that quotes by phone has a price that moves.
- 4 · Which way it's moving
- Three to five years of direction — more search interest or less, more businesses in the category or fewer. Direction is knowable. The size of the market in 2031 is not, and nobody has ever made a decision with it.
- 5 · What one customer costs to get
- The only question that decides whether the business works, and the only one you can't answer at a desk. You buy this answer: a small ad test, a week of outreach, a stall at a market. If it costs $75 to get a customer worth $180 and they come back twice a year, you have a business. If it costs $300, you don't — whatever the other four answers said.
What a textbook adds — total addressable market, a five-forces diagram, a segmentation matrix, a PESTLE — are tools for allocating capital across markets, genuinely useful if your job is deciding whether a company should enter Brazil. If your job is deciding whether to charge $60 or $75 in Denver, they produce paragraphs rather than decisions. The last section on this page says which to skip and what to do instead.
One afternoon at a desk: where the free data actually lives
Everything below is free, public, and reachable without an account. The order matters — the first step produces the code that every later step needs. Allow about three hours the first time and forty minutes the second. Have a blank document open and paste the link next to each number as you go, because in six weeks you will not remember where any of it came from.
Get your NAICS code — it's the key to the rest
Every US government dataset about businesses is filed by NAICS, a six-digit industry code. Without yours you can't query any of them; with it, four of the steps below take two minutes each. Search by keyword at census.gov/naics — put in what you do, not what you call yourself, and read the neighbouring codes as well as the one you land on.
Our example, a mobile bike repair, sits in
811490(Other Personal and Household Goods Repair and Maintenance — bicycle repair without retailing new bikes), while most bike shops sit in459110(Sporting Goods Retailers). Two codes, two different competitor sets, and you want both. Note the 2022 revision renumbered every retail code —451110became459110— so older tables and newer tables use different numbers for the same shops.Sixty seconds on the place itself: QuickFacts
census.gov/quickfacts — type a county, city or ZIP and get population, households, median household income, the owner-occupier rate, and the number of employer establishments on one screen. It's the fastest sanity check there is, and the household count is the denominator for almost every estimate you'll make afterwards.
Denver County, for the example below: 740,613 people, 335,428 households, median household income $94,718, 48.8% owner-occupied, 27,422 employer establishments. Sixty seconds, five numbers, one link.
Census Business Builder — the whole picture on one map
cbb.census.gov/cbb is the single most underused tool on this list. Choose your industry (that NAICS code) and draw a geography — a county, a city, or a radius around a point — and it returns customer demographics, consumer spending for your category, the number of businesses already operating in it, and workforce data, on one map, with a downloadable report at the end.
Use the Small Business Edition. Draw the radius you'd actually serve rather than the county, because the county is a political boundary and your van isn't. Export the PDF: it's the closest thing to a free market report that exists, and it has the Census Bureau's name on it rather than a content marketer's.
County Business Patterns — how many competitors have employees
County Business Patterns counts establishments, employment and payroll by county and NAICS code, back to 1986. The most recent reference year as this was written is 2023, released in June 2025 — annual data runs about eighteen months behind, which is fine for a trend and useless for this quarter. Query it on data.census.gov by typing your NAICS code and your county into the search box, or pull it from the CBP API if you'd rather have a CSV.
One caveat that changes the answer completely for a local service business: CBP counts only establishments with paid employees. Every one-person operation — the sole trader, the person with a van, most of your actual competition — is in a separate dataset, Nonemployer Statistics. The Census Bureau's own summary of that dataset puts it flatly: the majority of all business establishments in the United States are nonemployers. In a category like repair, cleaning or trades, they are most of your competition. Look at both datasets or you'll conclude there are four competitors when there are forty.
The five ACS tables worth knowing by number
The American Community Survey is the annual survey that fills in everything the decennial census doesn't. On data.census.gov, type the table ID straight into the search box, then set Geography to your county or place and pick the year. Five tables cover most small-business questions:
DP05age, sex and race — who lives there.DP04housing — owned or rented, when built, how many with a garage, which matters more than you'd think for home services.S1901household income in bands, so you can count the households above the income where your price stops hurting.S0801commuting — how people get to work, including the share who cycle or walk.S2401occupation, which is how you find out whether the neighbourhood is full of your B2B customers.Use the 5-year estimates for anything smaller than a big city; the 1-year estimates only cover areas above 65,000 people and are noisier. If the table won't load, it's usually because the geography is too small for that estimate.
BLS Consumer Expenditure Survey — turn households into dollars
A household count isn't a market. The Consumer Expenditure Survey is what turns one into the other: it publishes what an average American household actually spends, by category, broken out by income band, household size, age and region. The 2024 data, released December 2025, puts average annual expenditure at $78,535 per consumer unit, and the demographic tables break that down far enough to find your category.
Multiply the households in your area by the average spend for your category, in the income band you're targeting. The answer is too big — it assumes everybody spends the average — but it is a ceiling with a government publication under every input, which is more than a market-sizing slide usually has.
BLS QCEW — more current than CBP, and what a hire costs
The Quarterly Census of Employment and Wages covers roughly 95% of US jobs, by county and NAICS, quarterly, with about a five-month lag — much fresher than CBP. It gives establishment counts, employment and average weekly wage. The wage figure is the one to write down: it's what you'll have to pay the first person you hire in your industry, in your county, and it's usually a shock. Use the Data Viewer or the Industry Finder from that page.
Google Trends — the direction, not the volume
trends.google.com/trends/explore. Type the phrase a customer would search — "bike repair near me", not "bicycle maintenance services" — set the geography to your metro area rather than the country, and the range to five years. You're looking for two things: the shape of the year (when your season starts and how deep the trough is) and whether the line is drifting up or down.
Compare against a second term you know is stable, which is the only way to tell a real decline from a change in how people search. And do the free version of keyword research while you're there: type your term into Google itself and read the autocomplete suggestions and the "People also ask" box. Those are real queries, and they're phrased in your customers' words rather than yours.
Local licensing and registry records — the list nobody reads
This is the most specific data you will find and almost nobody uses it. Three places, all free. Your state's business registry — Colorado's, for the example — lets you search by name and shows the formation date, status and registered agent of every company in the state, so you can see who entered your category and when, and who has quietly gone delinquent. Your city's licence directory — Denver's — lists which trades need a licence and often who holds one. And your state or city open-data portal, which frequently publishes the whole licence list as a downloadable table: Active Business Licenses, Denver is exactly that.
Sort a licence list by issue date and you have something no report will sell you: how many businesses entered your category in your city last year, by name and address. That is the local version of "is this market growing", and it's a spreadsheet away.
The other half: ten conversations and one price
Everything above is secondary research — someone else collected it, and it tells you the size and shape of the thing. None of it can tell you whether a person will pay you. That's primary research, and for a small business it's ten conversations and one test, not a survey platform.
Finding ten people in a week
Go where they already are rather than trying to summon them: the local subreddit and the neighbourhood Facebook group, the club or class or league built around the thing, the queue at the place that sells the adjacent product, and two steps out from your own contacts ("who do you know who…"). For B2B, five booked calls beats fifty emailed replies, and the list you build by hand from Google Maps is better than any list you buy. Offer something for the time — a free diagnostic, a coffee, the results — and be honest that you're deciding whether to start.
What to ask, and the one question to never ask
Ask about the past, never about the future. "When did you last pay someone to do this?" "What did it cost?" "How did you find them?" "What was annoying about it?" "What did you do the time you didn't hire anyone?" Every one of those has a factual answer the person actually knows.
Never ask "would you use this?" or "what would you pay?". Nobody can predict their own behaviour, and everybody is kind to a person who is clearly hoping for a yes. You will collect ten yeses and no customers. Rob Fitzpatrick's The Mom Test is a hundred pages on this one point and it's the best afternoon of reading available on the subject.
Free primary research you can do tonight: read the reviews
Open Google Maps, find the five nearest competitors and read their three-star reviews from the last two years — not the one-stars, which are usually a single bad day, but the threes, written by people who were basically satisfied and still had a specific complaint. Tally the reasons in a spreadsheet. A complaint that repeats at three different competitors is a description of the gap you're walking into, written by your future customers, for free. The full teardown is on its own page →
Then ask for money — this is the only real answer
Conversations are cheap and so is agreement. A deposit is not. The strongest research a founder can do is to put a price on a page and see whether a stranger pays it: a pre-order, a $30 booking deposit, a paid pilot at a discount, a slot on a Saturday. One person paying tells you more than thirty saying they would, and it also answers question five — what it costs to get a customer — because you had to go and get that person from somewhere.
A finished market report, with its sources
Here is what an afternoon of the above produces. Northline Bike Repair is the example business used across this site: one founder, a van, mobile bike repair in Denver, deciding in week one whether to start and what to charge. The report is five rows, because there were five questions. Two of the figures are guesses, and they're labelled as guesses — which is the part most market reports leave out.
| Question | What we found | Where it came from |
|---|---|---|
| 1 · How many, how often | 335,428 households in Denver County; median household income $94,718; 48.8% own their home. Assume one in five households keeps a bike worth repairing, and one in five of those pays someone rather than doing it themselves or ignoring it: roughly 13,400 paid repairs a year in the county, about $800,000 at a $60 basic ticket. The two one-in-fives are guesses. Everything after them is arithmetic. | Census QuickFacts, Denver County |
| 2 · Who already sells to them | 27,422 employer establishments in the county in total; the bike shops sit in NAICS 459110 and the repair-only outfits in 811490. But CBP counts employers only — the one-person mobile operators, which are the direct competition, are in Nonemployer Statistics. The count that decided it was hand-made: searching "bike repair" across five Denver ZIP codes on Google Maps returned 31 listings, of which 4 were mobile and 2 of those had no reviews in the past year. | County Business Patterns, Nonemployer Statistics, Google Maps (counted by hand) |
| 3 · What the market pays | Six nearest shops publish tune-up prices: $65–$110 basic, $130–$190 full overhaul. Two publish nothing and quote by phone — which means their price moves. The two existing mobile operators add a $20–$35 call-out fee on top and neither takes a card on the doorstep. | Competitors' own pricing pages, plus the Services tab on their Google Business Profiles |
| 4 · Which way it's moving | "bike repair near me", Denver metro, five years: broadly flat year on year with a hard season — the March-to-September index runs about three times the January floor. Denver's cycle-commuting share (ACS S0801) has not recovered to its 2019 level. Read together: this is a leisure and errand market now, not a commuter one, which changes who to talk to and when to advertise. | Google Trends, ACS table S0801 on data.census.gov |
| 5 · What one customer costs | A $300 Facebook test over nine days: 214 visits to the booking page, 4 paid bookings, $1,486 collected. About $75 to buy a customer worth roughly $370 on the first visit — two of the four were two-bike households — against a typical single job of about $180. Counted from the Stripe charges, not from what the ad platform reported. | The founder's own test — the only line here that had to be paid for |
And then the part that makes it research rather than a document — what it changed. Northline was going to charge $50 for a basic tune-up; the published range said $65–$110, so it launched at $60 with a $25 call-out and stopped being the cheapest thing in the city by accident. The map showed four of the six shops clustered downtown and nothing serving the north-east, so the van starts there. The season told it to spend its small ad budget in March rather than spreading it over twelve months. And the honest verdict on the whole thing was yes, but not a big one: roughly 13,400 paid repairs a year in the county, and Northline needs about ninety a month, which is 8% of the county — hard but not absurd, and a very different conversation from "the US bicycle market is worth billions".
What the report did not answer: whether a person will choose a van coming to their driveway over walking the bike three blocks to a shop they already know. No dataset holds that. Only the $300 test touched it, and four bookings is a hint rather than a proof — which is why the next thing on the list was another $300, not another afternoon of research.
Research that isn't worth your afternoon
The TAM slide. A line that reads "the US bicycle repair market is worth $4.2bn, growing at 3.1% a year" is decoration. Nobody has ever made a decision with it, it usually traces back to a press release for a report nobody has read, and it tells a founder in Denver nothing about Tuesday. If you're raising money and an investor asks, build the number from the bottom up out of your own five answers — households in reach, times the share who'd plausibly buy, times your price. You'll get a smaller number and a much better meeting.
The bought industry report. A syndicated report is $2,000 to $5,000 and is written for a company deciding whether to enter your industry, not for you deciding whether to enter your street. It will not contain your county. Spend the money on an ad test instead. And if you genuinely need one, try your public library card first — many US library systems include free remote access to business databases that resell this material — and your local SBDC or SCORE mentor, who will often pull reports for you for nothing.
The survey of people who like you. Fifty responses from your friends, your mailing list and your old colleagues is fifty people being encouraging. It looks like data because it has percentages. Ten strangers describing what they last paid for is worth more than five hundred of these.
Five-year forecasts, and focus groups. The direction of the last three years is knowable and worth an hour; the size of the market in 2031 is a made-up number in a confident font, and a decision that depends on it is already wrong. Focus groups are expensive, slow and owned by the loudest person in the room — ten one-to-one conversations cost nothing and tell you more.
Common questions
How long should market research take for a small business?
One afternoon for the desk research — three hours the first time, forty minutes once you know the tools — and about a week for ten customer conversations you can fit around everything else. Then a small paid test, a few hundred dollars, to answer the one question a desk can't. Anything that takes a month is either a very unusual business or procrastination with a spreadsheet.
Where can I get free market data for my area?
Census QuickFacts for the shape of the place; Census Business Builder for demographics, consumer spending and business counts on one map by industry and radius; County Business Patterns for employer establishments by county and NAICS code, plus Nonemployer Statistics for the one-person businesses it misses; the American Community Survey on data.census.gov for income, housing, commuting and occupation; the BLS Consumer Expenditure Survey for what households actually spend by category; BLS QCEW for current employment and wages; Google Trends for direction; and your city or state's business licence records for who entered your category and when. All free, no account needed.
What is the difference between primary and secondary research?
Secondary research is data somebody else already collected — census tables, government surveys, competitors' published prices, reviews. It's free, fast and tells you the size and shape of a market. Primary research is what you go and find out yourself: conversations, a price test, a stall at a market. It's the only kind that can tell you whether a specific person will pay you. Do the secondary first, because it makes your primary questions much better.
Do I need to know my total addressable market?
Only if you're raising venture capital, and even then build it from the bottom up — people in reach, times the share who would plausibly buy, times your price — rather than quoting an industry figure. For a business that is going to be funded by customers, TAM is a decoration. What you need is the number of sales you must make each month and whether that is a plausible share of the people near you.
How do I find out how many competitors I have?
Three sources, and you need all three. County Business Patterns on data.census.gov gives establishments by NAICS code and county, but only those with paid employees. Nonemployer Statistics covers the sole traders, who in most local service categories are the majority. Then count by hand: search your category on Google Maps across the ZIP codes you'd serve, and cross-check against your city's business licence list, which often names every licensed operator with an address.
How many customer interviews are enough?
About ten, if they're strangers rather than friends. The repeating pattern usually appears by the sixth and the eleventh rarely tells you anything new. If your ten don't agree with each other, you're looking at two different customer types and the finding is that you have to choose one before you can price, position or advertise anything.
Should I buy an industry report?
Almost never. A syndicated report costs $2,000–$5,000, is written for a large company considering market entry, and will not mention your county. Try your public library card first — many library systems include free remote access to business databases — and your local SBDC or SCORE chapter will often pull material for you at no cost. Otherwise, spend the money on a test that produces a real customer.
Can AI do market research for me?
It can do the desk half well: find the right dataset, run the queries, read fifty competitor pages and reviews faster than you can, and hand back a report with sources you can check. That is what Velofound's research does, and the sources are there precisely so you can verify rather than trust. What it cannot do is talk to ten of your customers or persuade a stranger to pay a deposit — and those remain the two things that actually settle the question.
Ask it to research your market.
Describe your business and Velofound will research the market and the competition for you, with the sources under every figure — next to a website that's already live and taking payment. Free to start.
Start free →