River
Y CombinatorBacked by Y Combinator
FREE TEMPLATE

Employee Engagement Survey Template

Four documents and five sheets, including a question bank where every question carries the decision it would change and the person who owns that decision.

Free download  ·  No account needed

Three gates, all closed before launch

What a question list has no way to decide

Marloe Logistics, fictional 240-person warehouse software company, 28 teams across 6 departments. FY26 round, threshold set 10 February.

Gate one. Does anybody act on the answer

Every question is paired with the decision it would change and the person who owns that decision, and the pairing happens before the question is asked. Sixty-two were drafted and twenty-four survived. Those twenty-four resolve to nineteen decisions held by nine people, and one of the nine owns six of them, which is a queue worth seeing in February rather than in July.

Eleven keep last round’s wording exactly, so eleven carry a trend line and thirteen begin one.

Gate two. Will the report you promised exist

The threshold is five responses rather than five people, and the map is computed from the org chart before any invitation goes out. Ten cuts were asked for: five exist, two exist in part, three do not exist at all.

Cut requestedCellsHeadcount
forecast
Reportable
at close
Verdict
Department, location, level, tenure, manager or IC191919Exists
Team282016In part
Gender322In part
Team crossed with level, tenure or gender22084Does not exist

Headcount said twenty teams would get their own report. Responses said sixteen, and two of the four shortfalls were a single answer away.

Gate three. What happened to the last set of promises

Fourteen commitments were made after last round. Five were delivered, four are still live, five were dropped, and none of the five drops was mentioned to the people who had been promised it. Split this year’s participation by that status and one company figure separates into three.

GroupTeamsPeopleLast roundThis roundChange
Commitment delivered or live911378%86%+8
Commitment dropped in silence55684%52%-32
No commitment made147183%73%-10
Company2824081%74%-7

The five teams that fell to 52 percent answered at 84 the year before, six points clear of the teams that have since climbed to 86. Not one of these three gates can be closed by a question list, because each of them needs your org chart, your decision-makers, or your own record of what you promised.

An engagement survey template is a list of questions, and a list of questions is a list of things somebody would like to know. This pack builds the bank backwards. Every question carries the decision it would change and the named person who owns that decision, filled in before it is asked, and anything with neither gets cut. A 62-question draft came down to 24: nineteen cut because nobody could act on the answer, eleven duplicate constructs, eight about matters already decided.

The second mechanism is arithmetic rather than judgement. The segment map is computed from the org chart before launch, at a minimum number of responses rather than headcount, so you know which cuts will exist before you promise one. Ten cuts were requested at Marloe Logistics. Five exist. Team exists for 16 of 28 cells. Team crossed with tenure has 86 populated cells and one that clears the threshold, and the four managers who asked for it were told in March.

Third, last round's commitments get reopened first, because that is what decides who answers this time. Fourteen were made, five delivered, four live, five dropped, and none of the drops was ever communicated. Teams whose commitment landed came back at 86 percent. Teams whose commitment was dropped came back at 52. Read the comments properly with comment synthesis; the ratings behind any performance question belong in a calibration session.

The three numbers a question list cannot produce

Questions with the decision and owner they were paired with before launch, the segment cuts that survive a stated threshold, and participation split by what happened to each team's last commitment.

Question Bank

Illustrative. Marloe Logistics, fictional 240-person warehouse software company. FY26 round. 62 drafted, 24 shipped, 38 cut.

QuestionDecision it would changeOwnerVerdict
I understand how my pay is decidedWhether band midpoints get published to employeesVP FinanceShipped
My workload over the last three months has been sustainableWhether Implementation caps go-lives at three per consultantVP CustomerShipped
I know what I need to do to reach the next levelWhether levelling criteria get published per job familyVP PeopleShipped
In a normal week I have enough uninterrupted time for focused workWhether two no-meeting mornings a week get protectedCOOShipped
I trust senior leadershipNothing. No named person can act on this within two quartersnoneCut
Rate your work-life balance from 1 to 10Same construct as the workload question, which names a periodnoneCut
Should we return to five days a week in the officeBoard settled three days in October. Not revisited before FY28noneCut
Do you feel your compensation is competitiveAnswered by a market benchmark, not by opinionnoneCut
Cut reasonCountNote
No owner, or no decision the answer would change19The largest group, every year, at every company
Duplicate construct11Kept the version that points at a decision
Already decided and not up for revision8The expensive ones. See the note below
Shipped2419 decisions, 9 named owners, one owner holds 6

Asking about something already settled does not gather information. It manufactures a mandate: whatever comes back, some group now has a percentage to point at, and the company must either act against a decision it already made or explain why it asked. Eleven of the 24 keep verbatim wording from last round, so eleven carry a trend line and thirteen do not.

Reportable Segment Map

Computed from the org chart before launch. Threshold 5 responses, not 5 people. 28 teams, 240 employees.

Cut requestedCellsForecastAt closePeople coveredVerdict
Department666240Exists
Location333240Exists
Level444240Exists
Tenure band444240Exists
Manager or individual contributor222240Exists
Team282016191In part
Gender322236In part
Team and level715334Does not exist
Team and tenure862116Does not exist
Team and gender63100Does not exist

Twenty teams had five or more people and were forecast reportable. Sixteen were reportable at close. Marketing returned 2 of 6, Sales Engineering 4 of 6, People and Workplace 3 of 5, Data and Analytics 4 of 6. Two of the four were a single response short, and all seven borderline managers had been given their exact number five days before launch. Eight further teams covering 26 people can never be reportable at any threshold; they get a facilitated conversation instead of a survey.

Subtraction check: Operations and Finance returned 20 responses, of which Finance (6) and IT and Security (5) clear the threshold individually. Publishing both alongside the department figure exposes the 9-response remainder across three teams of two to six people. Department published, team detail withheld.

Participation Tracker

The same survey, split by what happened to each team's commitment from last round. 178 of 240 responded.

GroupTeamsPeopleLast roundThis roundChange
Commitment delivered or live911378%86%+8
Commitment dropped, never communicated55684%52%-32
No commitment made147183%73%-10
Company2824081%74%-7

The five teams that stopped answering had the highest participation in the building the year before. They answered, they were promised a specific thing, and they watched nothing happen for eleven months. The reasons were all defensible: a go-live cap became unaffordable in a forecast review nobody from Implementation attended, a knowledge base rewrite lost its writer in May, two contract writers lost their budget to base pay corrections. None of the five was ever communicated.

Where the variance actually livesRangeSpread
Company, all 178 responses61%one number
Six departments54 to 6814 points
Sixteen reportable teams33 to 9360 points

The team at 33 sits inside a department reading 63, because the same department contains a team at 82. A department average is not a summary of its team results. It is where team results go to disappear, and the arithmetic guarantees it.

What is in the pack

01

Question Bank

Every question with the decision it would change, the named owner, the action if unfavourable, and whether the wording is verbatim enough to carry a trend line. Cut questions stay in with their reason.

02

Reportable Segment Map

Every cut anybody asked for, with populated cells, cells clearing the threshold, and people covered. Computed before launch so you promise only what will exist.

03

Participation Tracker

Each team against the number of responses it needs to get a report at all, with last round's commitment status beside it, which is what makes the participation number diagnostic.

04

Prior Round Commitments

Every promise from last round with its owner, its status, and the artifact an employee could actually see. Three status values: delivered, live, dropped, carried forward into an action planning pack that rechecks the actual item a delivered commitment was supposed to move.

05

Result by Segment

Company, department and team, with the team range printed beside the department range, and below-threshold segments shown as below threshold rather than as a percentage.

06

Survey Design, Communication Plan and Action Planning Guide

The record of every pre-launch decision, five drafted messages including the borderline-team note, and the plan that turns nineteen pre-agreed decisions into commitments with artifacts.

07

Two rules every prompt reads first

No question ships without an owner and a decision. A threshold is set once, in responses, and never lowered afterwards to make a number appear.

How it works

  1. 1

    Send last round and the org chart

    Previous results, whatever was promised afterwards including the promises that died, and headcount by team, level, location and tenure.

  2. 2

    Get the commitment register and the participation split

    Every promise with a real status, and last round's participation split three ways by what happened to each team's commitment.

  3. 3

    Cut the bank and fix the threshold

    Each question paired with a decision and an owner, anything ownerless cut, and the segment map computed before a single invitation goes out.

  4. 4

    Field it and report where the variance is

    Borderline teams told the response count they need, then a result by segment with the team range beside the department range.

Frequently asked questions

Why would you cut a question like "I trust senior leadership"?

Because at 41 percent favourable nobody can name what changes. It is a real construct and there is no owner and no decision, so asking it spends the employee's two minutes and their goodwill on nothing. Replace it with a question about something the company controls, like whether decisions get explained, which has an owner.

Most of my teams are under ten people. Is this useless?

It is the case the segment map exists for. Eight of Marloe's 28 teams can never be reportable and they are told so before launch rather than after. Those 26 people get a facilitated conversation run by somebody outside their reporting line, which is a better instrument for five people than a survey.

Do I need an external benchmark?

Only if your wording matches the published instrument exactly, which is the whole basis of any comparison. NIOSH kept half its quality of work life items verbatim from a 1977 survey specifically so answers could be compared across 25 years (NIOSH). Reworded questions get a figure and no trend line.

Will this tell me which managers are the problem?

No, and it refuses to. Results are reported by team, never by named manager, and never sorted by score, because a ranked list is read as a league table within seconds and the bottom three spend a month explaining rather than fixing. Whether a manager is the cause is a real question and a favourable percentage is not its answer.

How do I get participation back up?

Publish what happened to every commitment from last round, including the drops, with the actual reason for each. Teams whose commitments landed came back at 86 percent; the five dropped in silence came back at 52. One of those five lost its budget to base pay corrections in the merit cycle.

What happens to the free-text comments?

They are counted here, 611 of them from 178 respondents across four open questions, and read separately. Skimming several hundred comments for quotes that support the scores you already have is advocacy rather than analysis, so that job gets its own method in comment synthesis.

Can I run a survey during a union organising campaign?

Check it before launch rather than after. Soliciting employee grievances during a campaign can be treated as an implied promise of benefits, while a practice predating the campaign may continue unchanged (NLRB). An unchanged annual cadence is worth protecting for that reason alone. Take the specific facts to counsel.

Find out how many of your questions have an owner

Send last round's results, whatever was promised afterwards, and your headcount by team. The first things back are the questions nobody can act on, the segment cuts that will not exist, and what your participation number is going to be.

Cut my question list