Market research survey design, step by step

Start with the decision

Write down the business question (should we launch this product, at what price, to whom?) and the decision each answer will inform. Then sketch the analysis: which tables and comparisons will settle the question. Write the questions last, and only the ones those tables need.

Define who you need to hear from

• The target population: who exactly the result should describe, such as Australian adults who bought coffee beans in the past month.
• The sample source: your own customer list, a panel provider, or people intercepted on site.
• Quotas on the variables that matter, such as age band or region, so the sample's mix matches the population's.
• Enough responses in every group you will compare; the sample size guide shows how many.

Screen people in, and out

Put screening questions first. Ask about category use and, where it matters, exclude people who work in market research, advertising or the industry being studied. Do not reveal which answer qualifies, or people will guess it. End the survey politely for those screened out, and keep their records apart from completes.

Order the questionnaire

• Go from general to specific.
• Ask unaided questions before aided ones: “Which brands of coffee can you think of?” comes before any list of brands.
• Ask about behaviour before attitudes.
• Put demographic and sensitive questions last.
• Randomise the order of options, and of concepts, so position does not decide the result.

Testing concepts and prices

In a monadic test each person sees one concept; in a sequential monadic test each sees several, in rotated order. Monadic is cleaner, because nobody compares, and needs more people. For each concept, ask about appeal, believability, how different it is, and purchase intent at a stated price.

For price, the Van Westendorp Price Sensitivity Meter asks four questions: at what price the product would be so cheap you would doubt its quality, a bargain, getting expensive, and too expensive to consider. Plotting the four sets of answers as cumulative curves shows where they cross, which suggests an acceptable price range. It records what people say rather than what they pay, so treat it as one input alongside other evidence.

Protect data quality

• Soft-launch: open the survey to a small part of the sample first and check the data before the rest.
• Look for speeders, who finish far faster than anyone could read, and straight-liners, who give the same answer down a whole grid.
• Check open-text answers for gibberish and copied text, and look for repeat submissions.
• Use attention checks sparingly; too many annoy honest respondents.

Weighting and analysis

If the sample's mix differs from the population's on something that matters, weighting can bring it back into line: each response counts for more or less according to its group. Weighting reduces bias but widens the margin of error, so keep the weights moderate and always report the unweighted number of responses.

Report every percentage with its base, test differences between groups before calling them real, and keep the analysis plan you wrote at the start in front of you.