Exit Codes Are Not the Reason

The form everyone fills out the same way

A med-surg night unit loses fourteen nurses in five months. I have watched the version of this scene that comes next more than once. Someone in HR pulls the exit forms, lays them in a row, and reads them out: pay, pay, personal reasons, better opportunity, pay. So the unit gets a market raise drafted against the whole headcount, and the year's retention budget is committed before anyone has talked to a single nurse who walked. The form did its job. It produced a number a leader could act on. The problem is that the number is fiction, and the literature has been saying so, in plain language, since before most of those exit forms were printed.

I want to be specific about what happens in the room, because the abstraction hides the failure. An exit interview, in most places, is run on someone's last day or first day gone, by HR, with a dropdown menu. The leaver has a final paycheck to collect, a reference to protect, and former coworkers still inside. The rational thing to write is the safe thing. "More pay" burns no bridges. "My charge nurse staffed me with seven patients on a five-patient grid every third shift and lied about it upward" burns several. So the form fills with pay, the dashboard rolls up pay, and the board hears that the unit has a wage problem. Nobody in that chain is lying exactly. They are each doing the locally sensible thing, and the sum of locally sensible things is a wrong answer with a budget attached.

What the canon actually found

The cleanest demolition of the pay story is old and still unbeaten. Leigh Branham mined the Saratoga Institute's database^[1] of roughly 19,700 exit and post-exit surveys across seventeen industries and found a gap so wide it is almost comic: about 89 percent of managers believe their people leave for money, while 80 to 90 percent of the people who actually left name a push factor instead. A manager. A promise made in recruiting and broken in week two. No growth. Feeling invisible. Overwork. Lost trust in leadership. Branham's contribution was not just the catalog of seven hidden reasons; it was the insistence that a quit is a process, not an event. There is a first flicker of doubt, a period of weighing and quiet attempts to repair, a decision, then a search. Each stage shows on the floor if anyone is looking: withdrawal, fewer suggestions, sudden rigidity about the schedule. Managers do not look, because the form at the end captures the last straw and never the first crack.

Branham's other move is the one I cite most, and it is the one that should embarrass anyone running same-day exit interviews: distrust the first-day form, and use third-party post-exit interviews weeks after departure, once the leaver has stopped protecting references and will tell you the truth. The Work Institute, which runs tens of thousands of these third-party exits a year^[2], reaches the same floor from the other side. Their interviewers code verbatim answers into a fifty-reason taxonomy instead of letting HR pick a dropdown, and they put preventable departures at about 75 percent, with career development, work-life balance, and manager behavior consistently beating pay as the true cause. Read those two together and the indictment is total: the instrument most organizations use to learn why people leave is built to produce the one answer that is reliably false.

Gallup gets at the same wound with a different tool, and here I half agree and half push back. Jim Harter's Q12 meta-analyses^[3] are the most validated engagement instrument in the field, eleven editions deep, alpha north of 0.9 at the unit level, and the unit-level comparisons are real: top-quartile teams run meaningfully lower turnover than bottom-quartile teams in the same company. Where I stop nodding is the temptation to treat the survey as the answer. It is not. A survey locates the sick unit. It cannot tell you what the charge nurse on that unit already knows. Gallup's own retention people say as much when they report that about 42 percent of turnover is preventable and routinely ignored^[4], and that roughly half of leavers say someone could have kept them but nobody asked in the ninety days before they left. The number that should keep an operator up is that last one. The cure was a conversation, and the conversation never happened.

If Branham diagnoses the corpse, Richard Finnegan^[5] is the one who insists you can intervene while the patient is alive. His stay interview is not a survey and not a performance review; it is the direct supervisor sitting down with each person, asking five researched questions in order, and then shutting up. Listen 80 percent of the time. Probe with "tell me more," "give me an example," "how important is that to you." The five questions are only doors; the probe is the method. His core finding lines up with everything above and I think it is the single most useful sentence in the literature: the top stay-or-leave variable is trust in the direct boss. Not pay, not perks, not the mission statement. The person standing over you on a Tuesday. That is also why I get impatient with retention programs that route everything through HR and an engagement platform. The lever Finnegan found does not live in HR. It lives in a hallway exchange between a worker and a supervisor that no system records.

The same shape in every sector

What convinced me this is structural and not a healthcare quirk is that the exact same shape shows up wherever I look, with different vocabulary on the exit form. The push factors change costume; the mechanism does not.

Trucking is the most honest example, because the failure has a clock on it. Stay Metrics surveyed tens of thousands of drivers^[6] as a neutral third party in the first weeks on the job and scored what actually breaks drivers inside ninety days. Home time less than promised tops the list at 93.4 percent. Then inexperience, low dispatcher satisfaction, work-family conflict, low recruiter satisfaction. Read that sequence and the cause is not the road; it is the distance between what a recruiter said and what week two delivered, plus a dispatcher who stopped returning calls. ATRI puts the industry's recent turnover near 48 percent^[7] in a soft freight market, down from the 90-percent-plus that large truckload carriers used to treat as weather, and replacement runs eight to roughly thirteen thousand dollars a seat^[8]. The fix for a broken home-time promise is in marketing and dispatch policy. It is never in the driver lounge, and a fuel-card perk does not touch it.

Hourly work tells the same story with the pay variable held down on purpose. EmployBridge runs the Voice of the American Workforce survey^[9] across tens of thousands of hourly workers a year, and in a stretch where wages actually rose, the top stated reason for leaving the last job became "not feeling appreciated," with overwork right behind it. Pay still wins the decision to take a job; appreciation and control over the schedule decide whether people stay. This is the part most retention dashboards get wrong. They are built to track the thing that gets you hired and blind to the things that decide whether you are still there in six months.

On the schedule point, Zeynep Ton and the Good Jobs Institute^[10] did the work I respect most, because she refuses to treat any of this as sentiment. Her Good Jobs Diagnostic scores the job itself, an 86-question instrument plus frontline interviews aimed at finding where the operating model manufactures its own turnover. Understaffing, in her frame, is not a labor-cost win; it is a design failure that drives the vicious cycle of low investment, poor execution, low sales, and lower investment still. I bring her up to push back on a lazy reading of everything above. Yes, the information failures are real and recoverable. But Ton is a useful reminder that sometimes the floor is telling the truth and the operating model is genuinely broken, and listening harder will not staff the shift. Those are different problems, and conflating them is how good diagnoses turn into wasted budgets.

The shocks that hide the same failure

Two situations bend the curve and deserve their own note, because in both the people who can see the next failure are the ones structurally least likely to say so. The first is the post-layoff survivor. Joel Brockner's survivor research^[11] established decades ago that the commitment of the people you keep hinges less on the cuts themselves than on whether the process looked fair: advance notice, clear criteria, dignity in the execution. When the outcome is bad, people audit the procedure. The modern sizing is brutal. Leadership IQ surveyed thousands of survivors^[12] and found 74 percent reporting lower productivity, 69 percent saying quality declined, and 87 percent less likely to recommend the employer. The layoff was meant to keep the strong performers; the second, voluntary wave is what takes them, and it forms in exactly the quiet you cannot see in a pulse survey tied to an employee ID. Nobody writes "I am job hunting because I no longer believe the rationale" next to their own name.

The second is union vulnerability, and here the cost of not knowing is measured by a statute. Phillip Wilson argues for getting "left of boom"^[13], meaning you detect and defuse before the petition, because once cards are signed the options collapse. NLRB election petitions roughly doubled^[14] from FY2021 to FY2024, the highest intake in a decade, and after the Cemex decision an employer has fourteen days to respond to a recognition demand. The signal that predicts a drive, the trust by shift, the grievance with no ticket, the new informal spokesperson, lives in the heads of shop-floor leads who have every reason to keep it from management. By the time it reaches the official channel, it is a petition, and the response window has already shrunk to two weeks.

Why it stays in their heads

Pull these threads together and they converge on one mechanism, which is the thing I actually believe. The information that predicts a departure rarely lives in the data systems. It lives in the heads of the people doing the work, and it stays there because every channel built to collect it punishes honesty. Tell your own manager you almost quit and you have effectively handed in early notice. Write the real reason on an exit form and you torch a reference. Raise the broken promise in a town hall and you mark yourself. So the keeper says nothing, the leaver writes "pay," the supervisor recodes it to a clean dropdown, the dropdown averages into a turnover percentage, and the percentage reaches the board as a color. Each step is locally rational. The sum is a decision-maker who is structurally the last to know the one thing the floor knew first.

The grim joke in all of it is that the truth is not missing. Every named expert above is, underneath the framework, describing the same recovery move: get a neutral party to ask a specific person about a specific recent moment, somewhere the answer cannot be used against them. Branham's third-party post-exit. Finnegan's listen-eighty-percent stay interview. The Work Institute's coded verbatim. Stay Metrics' day-seven neutral survey. None of it is exotic. The reason organizations do not have the answer is not that the answer is unknowable. It is that the answer is unasked, because the only channels they have built for asking are the ones the floor learned long ago to lie into.

Which points, finally, at the kind of instrument this calls for. Not another engagement survey, which locates the sick unit and stops there. Not another exit form, which collects the safe answer by design. Something closer to what the canon has been describing for twenty years and few have operationalized at scale: a neutral, confidential conversation, anchored to a real recent shift rather than a feeling on request, run for enough people that no answer traces to one person, and read back at the altitude where each person actually knows something. That is the instrument we are building at Latent Variables. The literature already told us where the truth sits. The open problem was only ever reaching it before the resignation letter does.