Engagement Survey

For a long time the only window a company had into its own condition was the exit interview, the last conversation held with someone who has already decided to leave. By then the person usually has the next job lined up, and they answer carefully. They will keep running into former colleagues, a reference call may be coming, and what comes out of that meeting is usually one sentence about a new opportunity.

So companies found out late, every time. A founder believes the mood is fine, four people leave one team over six months, and the problem finally shows up as turnover rate. The people who left don't come back, and the ones who stayed guess among themselves about why.

At twenty people a founder can fill that gap by talking to everyone. At forty-five it stops working. Meeting each person monthly means booking two conversations a day, and what those conversations produce lives in one person's memory, which can't be compared against last quarter.

An engagement survey is the document that takes over that job. It asks everyone the same questions before anyone quits, and it stacks the answers quarter over quarter or half year over half year. The point is to see a decline as movement in a number rather than as a resignation letter.

One round holds four blocks: the core items about engagement itself, items by area, free-text questions, and the attributes used to slice the results.

1. Core items — five to eight of them, and not a word changes between rounds. Short statements such as "I know where the company is heading" and "I can see how my work contributes to that direction," answered on a five-point scale. The average of this block is the headline number, and it is the one value carried forward and compared across rounds.
2. Items by area — this is where a decline gets located. Split the questions into relationship with the manager, growth opportunities, compensation, collaboration, and workload, with three to five items each. Common additions are psychological safety, whether people can raise mistakes or disagreements without paying for it, autonomy, whether people decide how the work gets done, and transparency, whether the information needed to make a call is actually shared.
3. Free-text questions — two or three is plenty. Scores say where things are bad and stop there. When compensation comes back at 2.6, whether the problem is the amount or the opacity of how amounts are set only surfaces in this box.
4. Attributes — team, tenure, and job family are usually enough. Slicing the answers is what reveals the team hidden behind the average. Because combining all three can identify a single person, any cell with fewer than five respondents gets rolled up to the level above before anyone looks at it.

Keep the questionnaire under 30 items and under 15 minutes. Past that, the later items collect answers picked at random, which is worse than not asking. Twice a year is the common cadence; teams move to quarterly only when they have the capacity to act on what comes back, since a round that changes nothing costs more than it returns.

Everyone writes it, anonymously. Designing the items and compiling the results belongs to whoever runs people operations, and at a company of around forty that is often the founder. Anonymity is usually protected by running the survey in an outside tool rather than an internal one, and free-text answers are published grouped by theme rather than quoted verbatim, because phrasing gives writers away more often than people expect.

Three groups read it, each with a different view.

1. Leadership — the company-wide average, the scores by area, and the change since the last round. What becomes a priority next quarter gets decided here.
2. Managers — their own team's results. A team with fewer than five respondents receives the parent organization's numbers instead of its own.
3. Everyone — the results along with what the company has decided to do about them. This last step sets the response rate for the next round. One experience of being asked and seeing nothing change drops participation from the eighties into the fifties, and from then on mostly the frustrated answer, which makes the score read worse than the reality.

"The overall average held at 3.9, but growth opportunities dropped to 3.1 on its own."

The finding came from the sliced items rather than from the headline number.

"That team had four respondents, so we're not showing results at the team level."

The reporting boundaries were set in advance to keep responses anonymous.

"There are things from the last round we still haven't answered, so let's close those out before we send another one."

Said by someone who knows that asking again without acting is what kills the response rate.

Pagedeck went from 28 people to 45 in the six months after its Series A. Grace, the CEO, ran the first survey right after that. The overall average was 4.0, but respondents with less than six months of tenure sat at 3.2, and several free-text answers described not knowing who to ask and searching for answers alone at their desk.

Grace shared that result as it stood at the all-hands and assigned one person to own onboarding. The next round brought that group up to 3.8. What she watched more closely than the score was participation: 76 percent in the first round, 91 percent in the second.

The most common mistake is reading a high average as a healthy organization. The company-wide number is what gets printed in large type on the summary page, so that's where attention goes. Averages hide the bad places. A company at 4.1 overall with one team at 2.8 is ordinary, and if that team has six people it barely moves an average across 45. Leave it six months and the visible signals arrive first — burnout, or quiet quitting — and the company learns about it after someone resigns, which is where it started.

The other mistake is treating the act of surveying as an improvement in itself. Asking does signal interest, so the idea is plausible. In practice, asking and then doing nothing leaves things worse than before the survey went out. From an employee's side, the company has just produced evidence that speaking up changed nothing. When there aren't even two or three items a team can realistically act on, postponing the round is the better call.