Market research for startups should reduce the risk of one near-term decision. It cannot certify that an idea will work. The useful output is a compact evidence trail, explicit unknowns, and a test that asks real people to do something costly enough to matter.
Frame a decision-sized research question
Is this a big market? is usually too broad for an early startup. Ask which customer has the problem most often, what workaround already consumes time or money, which buying trigger creates urgency, or which claim deserves the next week of product work.
Anchor the question to an artifact when possible: a landing page, demo, repository, deck, waitlist, or rough product. Research becomes more concrete when the audience can react to something real.
Use an evidence ladder
Higher rungs are not automatically perfect evidence, but lower rungs cannot substitute for them. A trend report may size a category; it cannot prove that a particular founder will adopt your workflow.
- Context: market definitions, regulations, category language, and credible reports explain where to look.
- Observed behavior: workflows, job posts, existing tools, budgets, and workarounds show that people already spend effort.
- Direct problem evidence: interviews and detailed complaints reveal frequency, consequence, and current alternatives.
- Commitment: a qualified person gives time, data, access, a referral, or agrees to a next step.
- Demand: repeated use or payment demonstrates stronger behavior than stated interest.
Market analysis: size what you can reach
Founders searching for market analysis usually want a number. A top-down total addressable market, such as a category's global revenue from an analyst report, can show that a space exists. It cannot tell you which accounts to contact this month, and a small percentage of a large number is not a plan.
Size bottom-up instead: count the accounts that fit a narrow target, then remove the ones you cannot reach or serve yet. The result is smaller, easier to check, and tied to a next action.
- Name the unit you sell to: a company, team, location, or person. Official counts often measure a different unit.
- Find a dated public count for the closest category, and record the dataset, year, and definition.
- Apply only filters you can source, such as size band or region. Label any filter you estimate.
- Remove what your product cannot serve today, and state the reason.
- Turn the remaining count into a list-building or outreach plan that a test can check.
Match each question to a source
Research tools and reports help when each one answers a specific question. Before paying for a tool, write down which row below it serves. Competitive analysis is one row of market research; the SaaS competitor analysis guide covers it in depth with a copyable matrix.
| Dimension | Useful sources | What they cannot tell you |
|---|---|---|
| How many accounts fit? | Government business counts, industry associations, public directories | Which accounts have the problem, or how many locations belong to one company |
| How do they work today? | Job posts, public process documentation, community threads, customer interviews | How often the problem occurs across the whole segment |
| What alternatives exist? | Competitor websites, documentation, pricing pages, review sites | Why buyers chose them; interviews and a competitor analysis are needed |
| What words do buyers use? | Search results, forum threads, sales calls, support tickets | Whether those people can be reached or will pay |
| Will anyone commit? | A demand test: replies, pilots, deposits, paid plans | Secondary research cannot replace this row |
Run a seven-step startup research loop
- Write one decision and the riskiest assumption behind it.
- Define the target customer narrowly enough that two researchers would find similar people.
- Collect source-linked context and one deep competitor analysis.
- List the strongest evidence for, evidence against, and evidence still missing.
- Choose one outreach or product test that can produce a stronger signal.
- Set acceptance criteria, channel boundaries, cost, and a stop condition before acting.
- Record the returned signal and decide: continue, revise, or stop.
Keep research separate from demand validation
Secondary research can reveal language, competitors, communities, and plausible pain. It cannot tell you whether the people you can actually reach will change behavior. Treat the research report as the map and the demand test as the trip.
In Soloop today, a research task returns a report with the source links it collected. X posts publish through a connected account. Gmail outreach sends one message per recipient. Reddit posts are prepared as drafts that you publish yourself. Google Ads campaigns launch only after you complete campaign setup and payment. Code changes run in a sandbox. Actions wait for your approval by default. Choose a test that respects the real channel boundary instead of writing a plan as if every channel were automated.
Use this one-page research output
- Decision: the choice this research must inform.
- Target: the specific customer and situation.
- Known: source-backed observations with dates.
- Inferred: interpretations that remain contestable.
- Unknown: missing evidence that could reverse the conclusion.
- Next test: exact action, target, channel, owner, cost, and stop condition.
- Signal: what happened and how it changes the decision.
Worked example: sizing one agency segment from public data
The founder, product, and decision in this example are invented for teaching. The industry counts are real: they come from the U.S. Census Bureau's County Business Patterns (CBP) 2023 release, read on September 30, 2026. They are not Soloop customer data or the output of a Soloop task.
Decision: a founder has a working prototype that tracks client approvals of ad creative. For the next four weeks, should outreach focus on U.S. advertising agencies with 10 to 49 employees, or on every agency regardless of size?
CBP 2023 lists 15,512 U.S. establishments in NAICS 541810, Advertising Agencies.
Census CBP datasetsOf those, 10,572 have fewer than 5 employees, 1,859 have 5 to 9, 1,300 have 10 to 19, and 1,000 have 20 to 49. The 10-to-49 band therefore contains 2,300 establishments.
CBP 2023 U.S. file (ZIP)An establishment is a single physical location, not necessarily a company. An agency with several offices counts more than once, so 2,300 is not a count of agencies.
Census CBP methodologyCBP covers establishments with paid employees. Agencies without employees are outside these counts, and businesses classified under another industry code, such as marketing consulting, are not included.
About CBPThe completed one-page output for this example
- Decision: agencies with 10 to 49 employees versus all agencies, for four weeks of outreach.
- Target: account leads at U.S. advertising agencies with 10 to 49 employees who manage client approval rounds.
- Known: 2,300 establishments sit in the 10-to-49 band (CBP 2023); the under-5 band is much larger, at 10,572.
- Inferred: agencies in the 10-to-49 band are more likely to have several account staff and a formal approval step. This is a hypothesis, not a Census finding.
- Unknown: how many distinct agencies the 2,300 locations represent, how many handle creative approvals by email today, and who would pay.
- Next test: build a list of 40 agencies in the band from public directories, then ask 10 account leads to walk through their last approval round. Stop if fewer than 3 describe repeated rework.
- Signal: not collected. This illustrative example ends at the plan.