Primary market research: the version that survives contact with reality

What to take away

Primary market research collects first-hand evidence from people or observed situations for one decision. Interviews, focus groups, diaries, surveys and usability tests all qualify.

Direct collection is not automatically better than evidence you already own. Its value depends on whether the question, the participants, the method and the analysis fit the decision in front of you.

This is planning guidance. It is not legal or ethics advice. Name the qualified reviewer for your sector before recruitment starts.

Start with the decision, not the method

Write a one-page charter before you choose anything. Name the decision, its owner, the deadline, and the options on the table. Name who is affected and what a wrong answer costs.

Then set the action for each outcome: positive, negative, mixed, inconclusive. A study framed as "learn about our customers" invites broad questions and selective reading.

The decision charter is the document every later choice hangs from. If no result would change what you do, stop and spend the budget elsewhere.

Audit what you already know

Check customer records, support tickets, complaints, and lost deals. Also check prior studies and official statistics. The U.S. Census Bureau and the Bureau of Labor Statistics publish population, employment and spending data you can use as a frame or a benchmark.

For each source, record the population and definitions. Also record the method and date. Then record the geography and reuse rights. Two sources can disagree because they measured different things, not because one is wrong.

New collection should target a gap the proposed method can close. Collecting what a free dataset already answers wastes the participant's time and your budget.

Turn assumptions into answerable questions

List the claims your plan depends on: who buys, why they leave, what they will pay, which channel works. For each, state the current evidence and your confidence in it.

Rank by what happens if the claim is wrong. Convert the top few into neutral research questions a participant can actually answer.

Avoid questions built to approve a plan you have already chosen. If the wording contains the answer, rewrite it before it reaches a screener.

Define the population and the sampling frame

The target population is everyone the finding should describe. The sampling frame is the list you can actually reach. These are rarely the same, and the gap between them caps your claims.

Write inclusion and exclusion criteria before recruitment. Decide how you will treat former customers, non-customers, approvers and people your current offer excludes.

A finding cannot extend to people the frame never reached. Say so in the report rather than letting a reader assume otherwise.

Choose the sample that matches the claim

Probability samples can support population estimates when the design and the response are sound. Nonprobability samples produce valuable discovery, but size alone does not make them representative. A thousand convenience respondents are still a convenience sample.

Document the frame and the selection procedure.
Document the recruitment channels and the sample-size reasoning.
Document the quotas and the expected response.
Document the weighting. Report coverage, nonresponse, self-selection, attrition and subgroup limits.

Recruit without hidden pressure

Use truthful invitations and proportionate screening. Keep screening data to what the decision needs, and explain deletion practices without publishing criteria that let people game the screener.

Pay fair incentives and offer an accessible way to decline. Do not let managers recruit their own reports into sensitive studies; the reporting line is the pressure.

Keep voluntary research visibly separate from any service a person needs. A participant who thinks declining affects their account is not consenting.

Make participation informed and protect people

Before participation, explain several things. Here is what to cover:

Consent is a process, not a signature. Check understanding, take questions, and honor a change of mind.

The U.S. Department of Health and Human Services' Belmont Report sets out respect for persons, beneficence and justice. Its regulatory context is narrower than commercial research, so treat it as an ethics prompt and apply current law, sector duties and qualified review to the actual study.

Minimize identifiers, separate contact details from responses, restrict access and set a deletion date. Set pause and escalation authority before fieldwork, and keep a duty of care after collection when the study creates a foreseeable risk.

Where the study touches personal data, the W3C Privacy Principles give designers shared language for privacy and warn against shifting that work onto individuals. Use it to interrogate your data flow, not as a substitute for counsel.

Design neutral instruments and pilot them

Use plain language, one concept per question, balanced answer options and a recall period that fits. Avoid leading premises, double questions, forced certainty and scales with unclear anchors.

In qualitative guides, start broad before probing. In tasks, avoid hinting the expected path. Translate meaning, not words, and test every language version.

Pilot the whole path:

Record defects and corrections, approve a controlled version, and log every material change made after fieldwork begins.

Control fieldwork and analyze honestly

Train researchers, monitor early sessions, verify eligibility and preserve original records. For surveys, look for fraud, impossible timing, patterned answers and contradictions using several signals together. Never discard data only because it conflicts with the hoped-for result.

In qualitative work, tie coding to the research questions, keep traceable notes, and separate participant language from your interpretation. Seek disconfirming cases. There is no universal interview count that guarantees coverage; scope, diversity and decision risk set it.

For quantitative work, predefine outcomes, exclusions, missing-data handling, weighting and subgroup analyses. Report base sizes, question wording, field dates, sample source and uncertainty. Statistical significance does not establish business importance or causality.

When methods disagree, compare populations, definitions, periods and incentives before averaging. A conflict often exposes a hidden segment or a measurement problem worth more than the majority figure.

Manage vendors, panels and tools

Define purpose, population, recruitment sources, and responsibilities in the contract. Define accessibility, incentives, fraud controls, and quality evidence in the contract. Define subcontractors, data locations, ownership, and permitted reuse in the contract. Define retention, deletion, and incident notice in the contract.

Ask how panel members were recruited and how often they participate. Ask what leaving costs: data export format, licence termination, historical access. The GOV.UK technology selection guidance frames those as buying questions about change over time, data control, security and ownership cost.

Panel pricing varies by market and vendor. A study in New York, Chicago, Austin or Canada carries different recruitment costs and incidence rates. Get a written quote per market rather than a planning average.

Transcription, translation, coding and summarization tools can assist narrow tasks. Confirm retention terms, data location, redaction and human review. Synthetic responses are not customer evidence and must never be reported as observations from real people.

Report for the decision and close the loop

Structure the report by decision question. For each, show the answer, the evidence, the conflicting evidence, and the population. Also show the method, confidence, limitations, and the implication.

Include minority experiences that expose serious barriers. Record who decided, which evidence mattered, and why any threshold was overridden.

After you act, compare the result with the prediction and update the assumption register. Refresh the evidence when product, price, channel, technology or regulation changes, and retire studies that no longer fit the question.

For consequential designs, the GAO evaluation design guide explains how evaluation questions, evidence needs and design choices fit together. It is written for federal program evaluation, so use its design discipline as a check, not as proof that a marketing result is causal.

Primary research decision record

Common questions

What is primary market research?

It is first-hand evidence collected from people or observed situations for a defined decision. Interviews, observation, diaries, surveys and usability tests all qualify when the question, sample and analysis fit.

Is primary research always better than secondary research?

No. Existing data from the Census Bureau, BLS or your own records is often cheaper and sufficient. Collect new evidence only when a material gap remains and your method can answer it.

How many interviews are enough?

There is no universal number. Coverage depends on the study's scope, the diversity of the population and the risk attached to the decision. Report what your sample could and could not support.

When should a study stop?

Pause when participant safety, rights, data integrity or the decision purpose can no longer be protected. Set that authority before fieldwork, not during it.