What the Survey Cannot Ask

The survey window opens on a Monday and runs for three weeks. Two reminders go out. A month later the results arrive as a deck, split by function and by team, each score against last year and against a benchmark.

There is a heatmap. Managers in the red are asked to hold a session, produce an action plan, and log it.

None of this is careless. In most organisations it is the most rigorous thing anyone does about the conditions people work in.

Where this sits

The last essay was about the values on the wall, and why they lose to the numbers on the screen within sight of them. It ended by pointing at the instruments.

Every large organisation measures engagement, sentiment, performance and productivity. Not one of them describes the working arrangement that produced the numbers.

This is the same failure, one step closer in. Not what the organisation says it believes, but the layer that checks whether the belief is holding.

Sixteen years of looking

The best known survey is Gallup's. Twelve statements about your own experience of work, answered on an agreement scale, run across more than 140 countries and reported every year.

Its 2026 report, published on 8 April, put global engagement at 20 per cent. The longer series matters more: 12 per cent in 2009, a peak of 23 per cent in 2022, falling for two years since.

Sixteen years of continuous measurement, at enormous scale, and the number has moved eight points and is now going backwards.

Gallup's figures are almost certainly right about what they ask. So what would have to be true about the survey for that flat line to be the expected result rather than a disappointing one?

It asks the wrong thing of the only witness there is

Every item is a self-report, and that is not the flaw. Conditions inside a person's working week are not observable from outside without watching them, so the person inside it is the only instrument there is.

The flaw is what the survey asks of them. It collects how that week landed, and stops.

Two people on the same team, same manager, same project, same deadline, will not answer identically. Nothing in the survey can tell you whether the difference is in the working arrangement or in the two of them.

Gallup is candid about the limit. Its guidance to managers running a results session is never to ask how someone scored an item, and to ask instead how the items made them feel.

It looks once a year

Drift has no day on which it happens. A target raised by three per cent because last year's was met. A step added to a process after something went wrong once.

The survey looks at that once a year, and asks people to summarise twelve months from memory. Gordon Bower's work at Stanford, published in 1981, found that we recall a greater share of the experiences that match the mood we are in while recalling them.

Ask someone in a bad fortnight what the year was like, and they will describe a bad year.

So the same team returns one number in the week a release lands and another in the week after a reorganisation is announced. Nothing structural moved in between.

Pulse tools and always-on listening raise the sampling rate. They do not change what is being sampled, and frequency was never the constraint.

Almost nobody acts on it

Gallup published a figure in January 2023 that deserves to be better known. Eight per cent of employees strongly agree that their organisation takes action on surveys.

The same article says that surveying and then doing nothing makes engagement fall and turnover rise. That is the maker of the survey saying that its survey, used the way most people use it, makes things worse.

The reason is where the remedy gets sent. Gallup's own position is that a manager accounts for up to 70 per cent of their team's engagement, so the action plan goes to the manager.

A team scores low on having what it needs to do the work. That plan now belongs to someone who does not set the budget, cannot change the staffing, and did not agree the delivery date.

They will hold the session and produce a plan made of what is within reach. Next year the score will move a little, or not, and nobody will be able to say why.

And even then, it stops at the feeling

Suppose all three were fixed. Full response rate, measured continuously, acted on every time by someone with real authority.

It would still not tell you what the working arrangement is. It would tell you, precisely, how a population feels about it, which is a consequence of that arrangement and not the thing itself.

A measurement is structural when what it records would still be true if nobody had been asked.

How many people can put work on one person without seeing what they are already carrying. Whether the date was fixed before the work was scoped or after. Which approval can be routed around, and by whom.

None of those is an opinion. Each is a fact about the working arrangement, each is knowable, and none appears on any engagement survey, because a survey asks people.

The test, which takes ten minutes

Take your last engagement report. Find the items that moved.

For each one, ask what changed in the working arrangement between the two surveys. Not what was launched, communicated or trained. What changed about who can commit whom, and by when.

If nobody can name it, there are two readings. Either nothing structural changed and the number moved anyway, or something did and the survey missed it.

Opinion: The Survey Is Not Broken. It Stops One Step Short.

Our position is that the engagement survey is well made, correctly calibrated, and stops at the consequence without ever reaching the cause.

This is not an argument for stopping. How people feel about their conditions is a real reading and the earliest warning most organisations have, and an organisation that stopped asking would know less, not more.

The missing step is the one after it. A feeling reported at work is produced by something, and the something is nearly always a specific feature of the working arrangement rather than a fact about the person reporting it.

It is an argument about what the number is evidence of. A falling score says something in the working arrangement has moved. It does not say what, or when, or which of last year's four changes did it.

So the response gets chosen from a behavioural menu, because that is the shape the output comes in. Communication, training, manager capability, recognition, wellbeing, and a values programme if the score is bad enough.

Human resources did not choose a survey that cannot reach structure. It was handed a brief written in the language of behaviour and then held to account for outcomes produced by a working arrangement it does not control.

The misassignment was made above it, and it has held because the alternative requires an executive team to accept that the working arrangement is theirs.

The provocation we would leave open is about scale. The values programme addresses everyone at once and reaches nobody in particular. The survey addresses everyone individually and returns an average.

Both are aimed at the population. Neither is aimed at the working arrangement, and one room is left where an organisation can still look directly at how somebody's work is put together. Two people, half an hour, every week or every fortnight, almost everywhere on earth.

That room is the one to one, and it fails in its own way. The next essay is about why it turns into a review of the week's tasks almost every time, for a reason neither person in it chose: tasks are the only thing both of them have language for.

Essay one, The End of Structured Thinking.

Essay two, The Most Measured, Least Informed Person in the Room.

Essay three, The Organisation Nobody Chose.

Gallup, State of the Global Workplace 2026 report, published 8 April 2026

Gallup, "So You Administered an Employee Engagement Survey. Now What?", 4 January 2023

Bower, G. H., "Mood and Memory", American Psychologist, vol. 36 no. 2, 1981, pp. 129 to 148

The contents of this article are for informational purposes only and do not constitute professional, legal, or financial advice.