Create your first survey: 5 steps to a high response rate and usable data

For your survey to produce usable data, you need to ask the right people. Make sure your audience knows something about what you're asking, and that they are the people whose answers you actually need. Sending an engagement survey to all employees when you want a leadership perspective or sending a customer survey to lapsed customers when you want active ones, produces noise — not signal.

Once you know who you're asking, the next decision is how to reach them. Most audiences can be reached through several channels — email, SMS, in-app, and embedded links. Pick the channels where the audience is easiest to reach and most likely to respond. Use multiple channels if that makes sense for the audience, but don't fragment the response data across so many channels that the analysis becomes unmanageable.

Before you write a single question, decide what the survey is for. What conclusions do you need to be able to draw? What decisions will the data support? Without this clarity, the questionnaire drifts — and a drifting questionnaire produces drifting data.

If multiple internal stakeholders will use the results, get them aligned on the purpose before the design starts. The handful of conversations that take place before the survey is built are worth more than the dozens that take place after the data arrives. When everyone agrees on the purpose, formulating the questions becomes easier — and the questionnaire stays short and focused.

When the survey is going out anyway, it is tempting to add a few extra questions outside the primary purpose. The advice is straightforward: don't. Every extra question reduces the response rate, increases drop-off mid-survey, and dilutes the data you actually need.

This is especially true when respondents perceive some of the questions as irrelevant. The best surveys are the ones that cut the bone — every question contributes to the survey's overall purpose. If a question doesn't, it shouldn't be in the questionnaire.

Why every extra question costs you data

Response rate typically falls 1–2 percentage points per additional question over a 10-question baseline. For a survey going to 2,000 people at a 35 % expected response rate, every extra question costs roughly 7–14 fewer respondents. By question 25, the effective response has dropped by a third — and the dropouts are not random: they are disproportionately the busiest, most senior respondents whose perspective you most needed.

When you ask a customer or user a question, you implicitly create the expectation that the thing you're asking about can be changed. Ask whether they're satisfied with your store's opening hours and an unhappy customer will reasonably expect that you can change the opening hours. Why else would you ask?

The mechanism is simple: every question is an implicit promise. Asking about something you can't or won't change creates frustration when nothing happens — and accelerates the loss of trust the survey was supposed to build. If you can't change it, don't ask about it. The exception is exploratory questions clearly framed as discovery rather than evaluation, but those should be rare in a first survey.

Good reporting decides whether your survey produces value. To act on the insight, you need to handle the data correctly and turn the numbers into knowledge the rest of the organisation can use. The reporting structure must be designed before launching — adding it after the data arrives means redoing analysis you should have set up correctly the first time.

Three questions to answer before you launch:

Who needs access to the data the survey generates? Define the access list before launch — and check whether the platform supports the segmentation those stakeholders will need.

Can you slice the data the way you'll need to? By segment, by region, by product line, by time period — pick the cuts now, before the questionnaire is locked.

How will the results be presented? Can the platform make the data understandable without you having to do all the analysis manually every quarter?

For most contexts, 8–15 questions are the sweet spot. Response rate typically falls 1–2 percentage points per additional question over a 10-question baseline, and the dropouts are disproportionately the busiest, most senior respondents whose perspective is often most valuable. If a question doesn't contribute to the survey's primary purpose, leave it out.

Pick respondents who actually know something about what you're asking — and who can give you an answer you'll act on. Sending the survey too broadly produces noise; sending it too narrowly produces unrepresentative data. Once the audience is defined, choose the channels where the audience is easiest to reach: email, SMS, in-app, embedded link. Multi-channel distribution typically lifts response rate by 10–20 percentage points.

No. Every question implicitly promises that the thing you're asking about can be changed. Asking about things you can't or won't change creates frustration when nothing happens — and accelerates the loss of trust the survey was supposed to build. If you can't act on the answer, leave the question out. The only exception is exploratory questions clearly framed as discovery rather than evaluation.

Yes — always. Pilot the survey with 5–15 representative respondents before launching. Pilots catch confusing wording, broken logic, missing answer options and the gap between what you meant to ask and what respondents actually hear. Treat the pilot as a real survey: collect actual responses, look at drop-off points, talk directly to 2–3 pilot respondents about the experience, then revise.

Before launch — always. The reporting structure shapes what data you actually need to collect: which segments to capture, which time periods to compare, which audiences give access to which slices. Designing reporting after the data arrives means redoing analysis you should have set up correctly the first time. Reporting designed before launch reduces post-launch analysis work by 50–70%.