How to plan and run a survey you can trust

Pick two or three things you need to learn this cycle, then choose questions that speak directly to those goals. A survey that tries to cover everything at once is hard to act on afterward, because you can't tell which question moved which result.

Set the cadence up front. Quarterly pulse surveys work well with 10 to 15 questions and a narrow focus; a deeper annual survey can run 20 to 25 questions with more open-text room. Anything longer than that starts costing you completion quality more than it adds insight.

Avoid launching during a known busy stretch, close of quarter, right after a reorganization, or the week before a holiday. Scores and response rates both drop for reasons that have nothing to do with what you're trying to measure.

Anonymous means even you can't trace an answer back to a person. Confidential means a small group, usually HR, can see individual answers but the person's manager and coworkers cannot. Decide which one you're actually offering before the survey opens, and say so in plain language in the instructions.

True anonymity is hard to guarantee on a team of four. Anyone can work out who wrote a comment about “my manager” when there's only one manager and four possible respondents. Where full anonymity isn't realistic, say so rather than promising something your team size can't support — an overstated privacy promise that people don't believe produces worse data than an accurate, more modest one.

Report results at a level where no individual can be identified by elimination. Many organizations suppress or roll up any group smaller than five respondents into a broader category rather than reporting it on its own.

• Strip your own opinion out of the question stem. “How satisfied are you with our competitive benefits package” tells the respondent what to think before they've answered — describe the thing, not your judgment of the thing.
• Don't ask two things in one question. “My manager communicates clearly and treats me fairly” forces a respondent who agrees with one half and not the other to guess how to answer. Split it into two questions instead.
• Give time-bound questions an explicit window. “Have you received recent feedback” is ambiguous; “in the last three months” removes the guesswork and keeps answers comparable across people and survey cycles.
• Label every point on your scale, not just the two ends. An unlabeled midpoint gets interpreted differently by different respondents, which quietly adds noise to your results.

Response rate is completed responses divided by people invited, multiplied by 100. Decide up front whether partial completions count as responses, and keep that definition consistent between survey cycles so your rate is actually comparable over time.

eNPS is the share of promoters (scored 9 or 10 on a 0-10 recommend-to-work-here question) minus the share of detractors (scored 0 through 6), using percentages of total respondents. Passives, scored 7 or 8, count toward the total but drop out of the subtraction.

Use the response rate calculator and the eNPS calculator on this site to do the arithmetic. Both run in your browser, so the numbers you enter stay on your own screen.

• The most useful comparisons are usually this team against the org average, this quarter against last quarter, and tenure under one year against tenure over three years. A single organization-wide number is the least informative thing a survey produces.
• Be careful comparing very different jobs against each other. A warehouse team and a sales team can have equally healthy cultures and still answer the same question differently because of the nature of the work, not the quality of the workplace.
• Never break results down to a group small enough that one person's answer could be identified by elimination. When in doubt, roll a small group into a larger, related one instead of reporting it alone.

• Share results with the people who answered, not just with leadership. Even a short summary of what came back and what will change builds trust for the next survey; silence after a survey is one of the most common reasons response rates drop in later cycles.
• Pick at least one thing to visibly act on before your next survey cycle. You don't need to fix everything the results surface, but acting on nothing tells people their answers didn't matter.
• Keep a consistent set of core questions across cycles so you can track direction over time, rather than treating each survey as a one-off snapshot.