Employee engagement survey

Promise anonymity only if the survey can keep it

The first thing a reader checks is whether you can tell it was them. If you can, the answers you get back are the answers they are willing to have attributed — which is to say, the answers you already had. Do not ask for a name, do not ask for an email, and say on the first screen that you are not asking.

The leak is almost never the name field; it is the grouping questions. A department, a length of service and a job level, put together, describe exactly one person in a team of six. Keep the bands wide — four tenure bands, five departments, no job titles — make them optional, offer a “rather not say”, and say why you are asking at all: so the answers can be read by team instead of as one long list.

Then set a floor before you read a single answer: no group with fewer than five responses in it is ever reported on its own. Put that sentence in the intro as well. It is what turns a promise into a rule, and people can tell the two apart.

Ask about the drivers, not the mood

Engagement cannot be measured directly; it is the visible result of a handful of conditions, each of which belongs to somebody. Does useful feedback arrive? Is good work noticed? Is the workload finishable? Is there a next step visible from here? Do the tools work? Five statements on one agreement grid cover all of that in under a minute, and every row lands on a named desk.

Write each row in the first person and about one thing only. “My manager gives me feedback I can use” can be answered; “My manager is supportive and communicates well” is two questions wearing one coat, and the person with a kind manager who explains nothing has no honest column to tick. Then keep the same five rows next year — the entire value of a grid is the comparison.

One recommendation question, and what it is for

Asking how likely someone is to recommend the place as somewhere to work gives you a single number that moves, and it earns its place — as a summary, never as a diagnosis. It tells you the direction; the grid tells you the reason. Publishing that number on its own, with no repairs attached to it, is how an organisation teaches its staff not to answer next time.

Read it beside your own open answers rather than beside a benchmark from an industry you cannot see inside. A score that fell four points in a quarter where nothing changed except one restructure has told you something specific. The same score set against a figure from another company has told you nothing you can act on.

One box to write in — and a rule about quoting it

One open question, asked narrowly, is answered far more often than three asked broadly. “What is the one thing that would make this a better place to work?” comes back as a sentence you can act on; “Any other comments?” comes back as silence or an essay, and neither is usable. Keep it optional — a required open box costs you the ratings of everybody who was merely content.

Then decide how you will report the answers before you read a word of them. A verbatim quote identifies its author: a turn of phrase, a complaint only one person has, a reference to a project three colleagues worked on. Group what comes back into themes, count each theme, and quote nothing that names a person or a project — including the complimentary ones.

The survey is the cheap half

This year's response rate was decided by what happened after last year's survey, not by how well this one is worded. Publish the results within two weeks, including the parts that flatter nobody. Pick two things. Name the person who owns each one and the date it is due. Then say plainly what you looked at and chose not to change, and why — that last part is the one that always gets skipped and the one people remember.

Twice a year is a rhythm people can sustain; monthly is a survey they learn to close. Between the long rounds, a three-question pulse is enough to catch a team falling off a cliff, and it does not spend the goodwill the long one needs.