A CTO I know messaged me last month. His third platform lead in two years had resigned the day before.

He asked me to help him rewrite the job spec.

I asked him a question instead. What did the first two say on the way out?

Long pause. "Different reasons. One went to a startup, one wanted more money. Bad luck, mostly."

Three people. One seat. Two years. He wanted a better job advert. He needed a postmortem.

One Exit Is Noise. Three Is a Spec.

You would never accept this reasoning in production.

If a service fell over three times in two years, nobody on your team would shrug and say "different reasons every time." You would pull the logs. You would line up all three incidents and hunt for the shared cause hiding under three different-looking symptoms. Then you would fix the thing underneath and write it up.

Point the same failure at a human being and the rigour evaporates. Every departure becomes unique, personal and unconnected. Bad fit. Bad timing. Better offer. Life stuff.

Some of it is true. All of it is convenient.

Here is why we reach for coincidence. The alternative means the role itself is broken. And we designed the role. So the fault comes home to us.

Empty office chair pushed back from a developer desk with a dark monitor and abandoned coffee mugs

The Numbers Are Not on Our Side

Gallup ran a nationally representative study of 717 people who voluntarily left an employer in the previous 12 months. 42% of them said their manager or their organisation was in a position to stop them leaving. Not "would have needed a miracle". In a position to stop it.

It gets worse. 45% said no manager or leader had talked to them about their job satisfaction, their performance or their future at any point in the three months before they quit. Of the people whose manager did open a conversation, only 17% were asked what it would take for them to stay.

We are not losing these people to the market. We are losing them to silence.

The people meant to break the silence are running on empty too. Gallup's State of the Global Workplace research found manager engagement dropped from 27% to 22% in a single year, the steepest fall of any worker group.

My own survey work found 99.5% of respondents had experienced one or more types of bad boss. Nearly everybody. The manager layer is where retention lives or dies, and the manager layer is exhausted.

The Seat, Not the Person

Here is the pattern I see over and over in engineering orgs. The role gets written as a wish list, never as a workable job.

Signs the seat is broken rather than the hire:

Responsibility with no decision rights. They own uptime. They own nothing else. Every fix needs budget approval from someone who has never carried a pager.

Three jobs in a trench coat. Platform lead, on-call rota manager, vendor wrangler, and the person who fills in security questionnaires. Written as one headcount because one headcount was what finance approved.

Bus factor of one, by design. The role exists because a single person understands the legacy system. Nobody else gets time to learn it, so nobody does. Holidays get cancelled. Eventually so does the person.

Two bosses, two mandates. Reports into engineering, judged by product. Or the reverse. Both directions hear "yes" and neither ends up satisfied.

No mandate to fix causes. They are paid to absorb pain rather than remove it. Firefighting is a job description in most orgs, and firefighters burn.

None of these show up in an exit interview. What shows up is "I got a better offer." True, and useless.

Whiteboard org chart with one box rewritten several times, faint ghost outlines of earlier names still visible

"We Hired Wrong" Is the Comfortable Answer

Every leadership team I have worked with reaches for the hiring excuse first. The bar slipped. The interview loop missed something. The recruiter oversold us.

Sometimes true. Once.

Three times in a row and the hiring process is no longer the variable. You changed the person and got the same result. In any other system you would call it a control experiment, and you would go looking at what stayed the same.

What stayed the same is the role. The scope. The reporting line. The authority. The workload. The unwritten expectation of weekend recovery work.

There is a second comfortable answer, and it is worse. "The market is hot." The market is always hot for the people you most want to keep. Blaming the market is a way of saying you have no lever, which frees you from pulling one.

The Contagion Nobody Budgets For

The cost is never one seat.

Turnover spreads. Research published in the Academy of Management Journal found coworkers' job search behaviour and their attachment to the organisation predict individual quitting, over and above someone's own satisfaction and their own pay. Your colleagues leaving changes your odds of leaving.

So a respected engineer resigning is not one event. It is a broadcast to everybody who trusted their judgement. When the seat beside me empties for the third time, I stop reading it as bad luck. I start reading job adverts.

The Bill You Already Paid Three Times

One engineering-focused breakdown, citing SHRM figures, puts replacement cost at six to nine months of the departing person's salary once you count recruitment, lost output and ramp-up time. Specialist technical roles run higher.

Do the arithmetic on my CTO friend. Three replacements in two years at six months of salary each is eighteen months of payroll for one seat. He spent a year and a half of budget avoiding a week of root cause analysis.

I noticed a Reddit thread going big last week full of people furious about being charged twice for the same thing. Companies do it to themselves every hiring cycle and file it under recruitment.

One figure staggering under an enormous stack of servers and laptops while colleagues walk out through a door

How to Debug a Role

Treat the seat like an incident. Same discipline you would apply to a database outage.

1. Postmortem the seat, not the person. Blameless, written down, shared with the team. What did the role demand, and what did it withhold?

2. Read the last three exits side by side. One at a time they look unrelated. Stacked up, the shared cause tends to jump out inside ten minutes.

3. Ask the survivors first. Stay conversations beat exit interviews, because people still in the building have something to gain from honesty. Ask it straight: what would start you looking?

4. Count the decisions. List everything the role is accountable for. Then mark which ones the person owns outright. A short second list means you have designed a scapegoat.

5. Split it or give it teeth. Two headcount, or one headcount with real authority and a budget. Pick one. Reposting the same advert a third time is a decision to pay the same bill again.

6. Fix it before you post. Every week the role sits open, the remaining team absorbs the work. They notice what you chose to leave broken.

The Question Worth Asking

My CTO friend went back through the two earlier exits. Both people had asked for the same thing, eight months apart: authority to retire the legacy billing service. Both were told to wait for next quarter. Both left within a quarter.

The third one never asked. She had watched what happened to the previous two.

So here is your question. Think of the role with the highest churn in your organisation right now. When the next person quits for a different-sounding reason, will you believe them?