How to Run an Anonymous Survey for a Client, Step by Step

The mechanics of running an anonymous survey for somebody else's employees, in the order the decisions actually come up. Assumes you have done the diagnostic work and know roughly what you are asking.

1. Agree the cohorts before you write the questions

The single most common way one of these engagements fails is producing findings nobody can act on because everything is reported company-wide.

Decide with the client, up front:

• Which breakdowns matter. Department and tenure carry most engagements. Location if sites differ meaningfully.
• What the minimum reportable group is. Whatever you pick, hold it. A group under the floor should be sealed, not shown with a caution.
• What happens when a team is too small. There will be one. Agreeing now that its slice stays hidden avoids a conversation later where someone asks you to just show it this once.

This has to happen before the survey goes out. A demographic tag added after answers exist does not retroactively organise them.

2. Set the workspace up so the client owns it

Create the client's workspace, then invite an owner from their side before anything else. Someone in HR, People or Legal who will still be there after you have gone.

Do it first for two reasons. It is a five-minute task that becomes a scramble on send day. And a workspace only you can open is one the client loses at the end, along with the results they paid for, which is the kind of thing nobody notices until it is a problem.

3. Write fewer questions than you want to

Twelve is a lot. Twenty is too many and the last eight get straight-lined.

Start from a template built for the situation - climate assessment for an opening read, post-complaint pulse after an incident, layoff aftermath for survivors - and cut. The questions that earn their place are the ones whose answers change what you recommend. Ask psychological safety and everyday respect separately; plenty of places are polite and still unsafe.

Leave one open text field. It is where the finding you did not think to ask about shows up.

4. Send the kickoff from the client, and say the mechanism

The invitation should come from the client's leadership, not from you. It should say three things:

1. Who is running it and why. Name your firm. Do not let people discover it from the link.
2. What happens to the answers. Not "responses are anonymous" - everybody says that. Say where the answer is encrypted and who is unable to read it. A mechanism is checkable; a promise is not.
3. What will happen with the findings. Even "we will share a summary in four weeks" beats silence. People who have answered surveys into a void do not answer the next one.

Attach the employee handout: how this protects you. One page, written for them rather than for the buyer.

5. Watch participation, not results

Do not read partial results and start forming a view. Early responders skew, and you will anchor.

Watch the participation rate by cohort instead. A department at 15% while the rest are at 60% is itself the finding, and it is the one worth a conversation with the client mid-flight.

6. Present what can be verified

When you write it up, hand over the signed results document alongside your deck, so the client's own people can confirm the numbers are the ones the survey produced. One line in your methodology section covers it. It costs you nothing and it is the difference between findings and assertions.

7. Hand it over, on purpose

At the end, either the client takes over the subscription and keeps running their programme, or the workspace closes. Decide which with them rather than letting it drift, because a live workspace nobody owns is the state where data sits around for years.

If they take it over, nothing moves: same workspace, same results, and your access continues until they end it.

FAQ

How long should the survey stay open? Ten working days is usually right. Long enough for holidays and shift patterns, short enough to still feel current. Extending it rarely rescues a bad participation rate; a message from a manager sometimes does.

Should I let the client see results before I do? No. Agree that up front. You need to be able to frame a difficult finding rather than have it read raw in a leadership channel first.

What if a cohort is too small to report? It stays sealed. Fold it into a wider grouping for the write-up if you must, and say in the report that you did.

Can the client see who responded? No, and neither can you. Participation is a count, not a list. If your tool can tell you who has not answered yet by name, its anonymity claim is doing less than you think.

See how an engagement runs.