The engagement score: 7.4 out of 10, stable year over year. The resignations: up 30 percent, concentrated in exactly the teams the company can least afford to lose.
I've watched leadership teams sit with this contradiction, and the conversation always reaches the same fork. Either the survey is broken — wrong questions, bad timing, survey fatigue — or the leavers are anomalies: better offers, personal reasons, a regrettable coincidence.
Usually, neither is true. Both numbers are accurate. They're just measuring different layers of the company, and only one of those layers predicts what happens next.
What a survey actually captures
An engagement survey is an instrument pointed at stated experience: what people are willing and able to report about how work feels, in that week, on that form.
The answers pass through filters before they reach the spreadsheet. The mood of the moment — a survey filled the week after bonuses reads differently than one filled mid-reorganization. The question of safety — "anonymous" is a property of the tool, but felt anonymity is a property of trust, and people answer according to the second one. The audience effect — many respondents answer with one eye on what leadership wants to hear, or on what answer might plausibly improve their situation. And hope, or its absence — sometimes a high score means things are good, and sometimes it means people have stopped believing that honesty changes anything.
That last one deserves its own line, because it's the cruelest property of the instrument: disengagement can raise scores. The employee who has mentally left doesn't fight on the form anymore. The most dangerous quarter in a team's life — the one where people stop raising problems — often photographs as calm.
None of this makes surveys dishonest. It makes them a particular kind of signal: filtered, lagging, and aggregated around the question "how does this feel?" — which is a real question, just not the one most companies think they're asking.
What no questionnaire can see
Meanwhile, the events that actually decide a company's trajectory are structural, and they're invisible to any instrument made of opinions:
A decision bottleneck forming around one overloaded calendar, adding two weeks of latency to everything important. An ownership gap between two departments, generating friction every Monday and a workaround every Friday. A rework loop where "final" deliverables quietly come back twice. The slow shift in a meeting where one voice stopped contributing three months ago and nobody registered the silence.
People don't report these things on surveys — not because they're hiding them, but because from the inside, structure is like weather. You don't describe the air pressure; you just feel tired.
And there is the timing problem. Sentiment is a lagging indicator of structure. By the time frustration is strong enough, conscious enough, and safe enough to show up in an annual score, the structural cause has typically been operating for months. The survey doesn't see the disease early. It sees the antibodies, late.
The thermometer problem
Here's the fairest way I can put it: a survey is a thermometer. It can tell you the patient has a fever. It cannot tell you where the infection is.
A thermometer is genuinely useful — no good doctor would throw it away. The malpractice begins when the thermometer is the only instrument, because then every treatment gets aimed at the temperature itself. The score dips, so the company responds at the level of feelings: a motivation workshop, a values refresh, a team event, pizza.
If the underlying problem is structural — an ownership gap, a decision bottleneck, a rework loop — those interventions don't merely fail. They make things worse, because the team watches leadership respond to a structural disease with a social gesture and correctly concludes that either nobody understands the problem, or nobody intends to fix it. Next year's scores will capture that conclusion. The cycle is self-sealing.
What belongs next to the thermometer
The answer to the survey illusion isn't a better questionnaire. It's a second family of instruments, pointed at the structural layer the questionnaire can't reach: how long decisions take from raised to resolved; where work waits, and for whom; how often finished work comes back; who is consulted on a call versus who actually decides it; what share of senior hours goes to producing versus aligning.
Used together, the two layers finally make sense of each other. Sentiment tells you that something hurts and roughly where to look. Structural measurement tells you what to operate on. A company that holds both stops arguing about whether the survey is right and starts locating problems precisely enough to fix them.
One predictable objection: doesn't measuring at this level shade into surveillance? It's the right question to ask, and the answer lives in the target. Everything described above measures flows and structures — decisions, queues, rework, ownership — in aggregate. None of it requires monitoring an individual's screen, keystrokes, or private behavior. And in practice, the effect runs opposite to the fear: when problems can be located in structure, they stop being pinned on people. The team that "lacks commitment" turns out to sit behind a bottleneck. Structural measurement is, among other things, a defense attorney for the workforce.
The annual survey isn't wrong. It's lonely — a single blurry instrument pointed at the most consequential layer of the company, asked to carry diagnostic weight it was never built for.
Keep the thermometer. Then give it colleagues.