Thursday, 5:50pm. Coffee with a skin on top, sitting on the table. You've rescheduled this meeting three times, and now the person is sitting across from you in a conference room that smells like someone else's lunch and an overworked air conditioner. You spent all morning preparing the honest words — drafting them, deleting them, drafting them again. And you hear your own mouth say: \"listen, let's do one more sprint, I believe in you.\" Your palms are wet. The person nods with the expression of someone who's just been pardoned from something they don't fully understand themselves. That evening, you run the runway numbers and see the price of that \"one more sprint\" as a figure that makes your lower back seize up.
\r\n\r\nYou just averaged down into a losing position. That same evening, the HR priest in the cardigan contentedly puts away his rosary beads — he's inside your head, he's always there, and he just won again.
\r\n\r\nEscalation of commitment isn't the sunk cost fallacy, though people confuse them daily. Sunk cost is a one-time mistake: you finish a bad meal because you already paid for it. Escalation of commitment is when the same mistake gets repeated in a series, with a fresh justification each time — and it was exactly this form that Barry Staw documented back in 1976, with 240 business school students in a role-play investment scenario (Organizational Behavior and Human Performance, vol. 16, pp. 27-44). The hardest people to get out were the ones personally responsible for the original, already-failing decision — not the ones who'd merely observed it. The more that's already invested, the harder it is to exit — even when exiting is obviously better than continuing. In trading, this kills accounts. At auction, it kills wallets. At 8,790 meters, it kills people in a single night, and we'll get to that night. In hiring, it kills the slowest of all: first the team, then the company, then you personally — leaving behind nothing but a clean conscience and a negative balance in the till.
\r\n\r\nNobody says to themselves: \"I'm a coward, so I'm giving it another quarter.\" The inner monologue sounds cleaner than that: \"he's/she's a good person,\" \"the context is tough right now,\" \"maybe I didn't explain the requirements clearly enough,\" \"one more sprint.\" This isn't a lie. It's an alibi — self-deception with a warm human face, one that lets you skip the painful thing today, at the expense of whoever pays more for it later.
\r\n\r\n\"He's a good person\" isn't an assessment of an employee. It's an epitaph you write in advance, so you never have to write an honest review.
\r\n\r\nThe HR priest of losing positions is a specific figure. He's in every company; sometimes, he's you. Cardigan, gentle voice. He doesn't say \"I'm scared to tell the truth.\" He gives a blessing: \"let's give the person another chance, we're not some faceless corporation.\" He performs last rites over a position that's still breathing, and offers you communion wine labeled \"humanity\" — so you don't notice it's the same ratchet, clicking one direction with every overdue month, making the exit more expensive. Every year, he walks a few more people straight into bankruptcy with a spotless LinkedIn reputation intact.
\r\n\r\nIn a neighboring industry, this exact conversation has been happening without metaphors for a long time. At a body shop after a crash, the mechanic inspects the car, and if the repair cost crosses a threshold — 60-85% of the car's value, depending on the insurance terms — a report gets filed: total loss. Not \"we'll patch it up a bit more,\" not \"the mechanic believes in this suspension.\" The formula was already sitting in the contract six months before the crash, back when nobody was crying over a bumper yet. The appraiser doesn't ask the part whether it hurts. He asks for the number someone else honestly calculated in advance, back when their hands were still dry.
\r\n\r\nThis analogy limps on purpose: a person isn't a bumper, and comparing a firing to scrapping a car body is a crude oversimplification, one that deserves an apology in advance. But a limping leg sometimes reaches an uncomfortable truth faster than a straight one: a threshold calculated with a cold head before the crash saves more people than one calculated through tears afterward. Not because people are car parts. Because a decision made in the moment of pain is always signed by the pain, not by the calculation.
\r\n\r\nThe shame here has a very specific architecture. Firing someone means admitting that you're the one who hired them. Admitting that means facing a mistake with your own name on the \"hiring manager\" line. It's cheaper to keep paying the salary. The ratchet clicks.
\r\n\r\nWhy does this hurt more right now than it did three years ago: back when money was nearly free and rounds closed in a week, a mediocre hire drowned unnoticed inside overall growth — runway was long enough that \"one more quarter\" cost you a line in a spreadsheet, not the structure of the company. In 2026, rounds take longer to close, and runway gets counted in months, not years: the exact same delay that used to be easy to overlook is now visible to the naked eye in the monthly burn rate. Cheap capital used to hide escalation of commitment inside overall growth. Expensive capital puts it out in the light every month, right alongside the office rent bill.
\r\n\r\nAt 11pm that same Thursday, you open a spreadsheet and start counting. Not \"one more sprint\" — specific weeks, specific numbers, the specific impact on everyone on the team who can see that the weak link is holding on while the strong ones carry the load. You see the crack — and you close the tab.
\r\n\r\nA trader who doesn't set a stop-loss loses money. A founder who doesn't set one on a person loses money, the team, and his own back — but he keeps the title \"I'm not a monster,\" the most expensive title in the company.
\r\n\r\nHere's where the system actually cracks — not at the moment of firing. It cracks far earlier, that Thursday at 5:50pm, when you say \"one more sprint\" and you both know it isn't true. The person across the table isn't stupid. She hears the hollowness of \"I believe in you\" exactly as clearly as you do. And she starts planning her own exit — just on her own timeline, not yours. And you keep paying salary for that parallel process.
\r\n\r\nAveraging Down on a Person is when you pour more months into a losing position for the exact same reason a trader pours more money into a falling stock: to avoid admitting the entry date.
\r\n\r\nIn trading, this is called \"averaging down.\" You buy more shares of something that's already falling, to lower your average price — and you convince yourself this is a \"long-term position,\" not a refusal to lock in the loss. The mechanism is identical. Its name: Averaging Down on a Person. That's the one thing worth taking away from this piece.
\r\n\r\nFriday, the city still waking up, 7:40am. You're driving to work and you're not thinking about the product — you're thinking about Dmytro. Or Olena. Or Andriy. The name has changed, but it's been the same thought for three quarters running. That's the actual signal — not \"a bad review,\" not \"missed deadlines.\" The signal is that you're thinking about them at 7:40 on a Friday instead of thinking about the market.
\r\n\r\nThere's a metric nobody tracks in the OKRs: how many hours a week a manager's head spends living inside \"the problem team member.\" If that number is more than two, the position is already costing more than what's written on the payroll sheet.
\r\n\r\nAnd there's one more character in this scene — the team lead who ends up carrying the load for two people. He doesn't complain directly. He just starts looking tired. Then he starts showing up to meetings with that expression you already recognize — the expression of someone who's run the numbers and is, for now, choosing to endure it. Every month of \"one more sprint\" isn't a neutral state. It's the active burnout of someone who's actually doing the work. The HR priest won't see this in the KPIs — he's too busy giving his blessing to a position that's already dead.
\r\n\r\n\r\n\r\nLet's name the mechanism precisely: Averaging Down on a Person runs in four phases, and each one makes the next more inevitable.
\r\n\r\n| Phase | \r\nWhat you tell yourself | \r\nWhat's actually happening | \r\nCost | \r\n
|---|---|---|---|
| 1. The first signal | \r\n\"This is just an adjustment period\" | \r\nA deviation from the hiring criteria has been logged | \r\nWeeks | \r\n
| 2. The first \"one more chance\" | \r\n\"I gave clearer feedback — it'll improve now\" | \r\nThe position is open, but you haven't closed it — the ego tax starts accruing | \r\nA month, plus the team lead's first sharp word | \r\n
| 3. The second top-up | \r\n\"Firing now is awkward (the quarter, the project, she's trying hard right now)\" | \r\nThe team sees the rules don't actually apply — the standard drops | \r\nThe team's standard, plus several more months of salary | \r\n
| 4. The till decides for you | \r\n\"Unfortunately, we have to make cuts\" | \r\nThe decision was made by a crisis, not by you | \r\nInternal reputation, plus everything above, ×2 | \r\n
The ratchet clicks between the phases, not inside them. Every click makes the next exit more expensive psychologically — and cheaper in the sense of \"I've already put in so much, it barely matters anymore.\" By phase four, you typically don't get yourself out. The till gets you out.
\r\n\r\nIf this looks like a disease of the inexperienced — here's an industry where performance gets measured more precisely than almost anywhere else on the planet. Staw and Ha Hoang, in 1995, tested this mechanism on the NBA (Administrative Science Quarterly, vol. 40): players drafted higher got more playing time and stayed in the league longer — even after statistically controlling for actual performance, injuries, and position on the court. Coaches with million-dollar budgets and a stat table for every single shot were doing exactly what you're doing in that conference room: defending the entry date. A high draft pick is a public decision with the GM's name on it. A player like that doesn't get cut from the rotation. A player like that gets averaged down on.
\r\n\r\nAnd the same mechanism with no budget ceiling at all looks like this: the Shoreham nuclear plant on Long Island, announced in 1966 with a $75 million budget, grew across 23 years of sequentially justified \"one more quarters\" to over $5 billion — and was shut down without ever generating a single unit of commercial power. Jerry Ross and Barry Staw dissected this case in the Academy of Management Journal (1993) as an anatomy of organizational escalation: not one individual decision looked insane. Only the sum did. Your \"problem team member\" is the exact same plant, just with a payroll sheet instead of a reactor.
\r\n\r\nStop-loss, as a discipline in trading, means one thing: the exit level gets set BEFORE entry, not in the moment of pain, when emotion is already rewriting the rules. Because in the moment of pain, you have exactly two options — lock in the loss, or convince yourself it's a \"long-term position.\" And emotion always picks the second one.
\r\n\r\nKill-criteria are a stop-loss written on hiring day. Observable, specific, not moral judgments: not \"bad attitude,\" but \"misses the agreed scope with no warning for two sprints running, plus three rounds of feedback with no change in trajectory.\" Writing them in advance is the only way to keep emotion from rewriting the standard at the exact moment when paying for the deviation has already started to hurt.
\r\n\r\nThere's a profession where kill-criteria aren't written out of love for process — where the cost of not having them lies buried under rocks on a mountainside. A mountain guide on an eight-thousander lives by one rule: turnaround time. Rob Hall, the most experienced commercial Everest guide of the nineties, preached it to his Adventure Consultants clients as law, beyond appeal: 2pm. Wherever you're standing at 2pm — even fifty meters from the summit — you turn around. The rule gets written down below, at base camp, over tea, when there's plenty of oxygen and thinking comes easy. Because up in the death zone, there's nothing left to think with: a brain at 8,700 meters with no oxygen will agree with anyone who suggests not turning back.
\r\n\r\nOn May 10, 1996, Hall reached the summit at 2:10pm — already past his own limit. And he didn't go down. He stayed to wait for a client — Doug Hansen, a postal worker from Seattle, whom Hall himself had turned back in '95, almost within reach of the summit, and whom Hall himself had convinced to come back the following year. Hansen crawled to the summit around four in the afternoon. A storm was already gathering below. Eight people died on the mountain that night — among them Hansen, Hall, and Scott Fischer, the leader of a parallel expedition, who'd reached the summit at 3:45pm, also nearly two hours past his own rule.
\r\n\r\nMichael Roberto dissected that night in the California Management Review (2002) as a textbook case of escalation of commitment: clients had paid up to $65,000, years of preparation were on the line, public expectations were riding on it, a journalist was embedded with the group, two companies were competing over reputation. The more that's invested, the harder it is to turn back. But the sharpest detail isn't even about the money. Hall didn't break his own rule for all his clients — he broke it for the one client with the most already invested: last year's turnaround, and his own promise. Doug Hansen was his \"one more sprint.\"
\r\n\r\nAviation knows this exact mechanism under the bureaucratic name plan continuation bias, though pilots call it, among themselves, \"get-there-itis.\" A NASA Ames review of nineteen airline crashes between 1991 and 2000 found this factor causal in nearly half of them. The closer the runway, the more it hurts to divert to an alternate airport — even though the math of the diversion hasn't changed by a single meter. That's why, in aviation, the decision to go around is written into procedure, and it doesn't get put to a vote of feelings in the cockpit.
\r\n\r\nA guide doesn't ask how badly you want the summit. He asks what time it is. \"I'm so close already\" isn't an argument on a mountain. It's a symptom.
\r\n\r\nYour conference room is the exact same mountain, just slower: instead of a storm, the till; instead of oxygen, runway; instead of the summit, \"it's about to click into place.\" The ratchet clicks at altitude too: every half hour after 2pm costs more than the one before it — and that's exactly why each one feels like the last one you'll need.
\r\n\r\nEvery company has a line item on its balance sheet that nobody writes down: the \"ego tax.\" It's the cost of holding a position that generates no return, purely to protect yourself from admitting you opened it. In normal accounting, this line lives quietly inside \"salaries\" and \"operating expenses.\" But if you isolate it, it usually looks like this: 4-6 months of salary + one burned-out team lead + one lowered standard for the entire team + your own personal time spent at 7:40am on Fridays, multiplied by the number of quarters. Paid in cash, out of runway.
\r\n\r\nYou're not keeping the weak player around out of kindness. You're keeping him because firing him means admitting you're the one who hired him; it's cheaper to keep paying salary than to pay the ego tax — until the till decides for you.
\r\n\r\nThe HR priest in the cardigan isn't malicious. He's functional. His function is converting cowardice into virtue, so both of you feel more comfortable sitting at that conference table. He's good at his job. He's inside your head, he's cheaper than a therapist, and he never sends an invoice — you send it to yourself later, and by then it's a great deal bigger.
\r\n\r\nRight now, the natural reflex is to think of someone specific. Whoever comes to mind first — Dmytro, Olena, Andriy — you already know the name. Fine. But the question isn't about him. The question is about you.
\r\n\r\nWhen was the last time you wrote out kill-criteria — observable exit signals — on the day the offer letter got signed? If the answer is \"never\" or \"in my head,\" then you don't have a stop-loss. You have an emotion that rewrites the standard every single time, in the moment of pain — and always in the same direction.
\r\n\r\nAnd there's an even more uncomfortable mirror: the person in that position isn't winning either. She gets months of uncertainty, where nobody says \"you're not a fit\" out loud, but every gesture is already saying it. It's a specific kind of hell — being at a company that tolerates you, humanely. An honest, timely exit is a release for her too, not just for you. Humanity isn't one more quarter. Humanity is clear, and now.
\r\n\r\nAveraging down on a person costs both of you. And here's the counterintuitive part: teams that know how to exit cleanly and on time hold together tighter, not looser. Because the strong ones stay where they can see the standard is real, not painted on the wall next to a poster that says \"we believe in people.\" Patience toward the weak link is a tax the strong ones pay. They notice it before you do.
\r\n\r\nNow, the strongest version of the counter-argument — because without it, this piece becomes just one more set of pretty rosary beads, hung from the opposite side.
\r\n\r\nFirst. Stop-loss in trading has a documented failure mode: the whipsaw — when market noise shakes you out of a position an hour before it turns back up. In people, this looks like: someone who accelerates non-linearly. The first two sprints are failures, because your onboarding is one Notion doc that's been stale since last year, and the teammate who's supposed to help \"doesn't have time.\" Month three, the turnaround. Month six, she's the strongest player on the team. A kill-criterion of \"two sprints\" would have killed her at sprint two. That's why the criterion on the card shouldn't be \"misses sprints,\" but \"misses with no change in trajectory\": measure the derivative, not the point. And run one procedural test before you judge the person: if the previous two people in this exact position also \"didn't cut it,\" the thing that needs firing is the role, not a third person in a row.
\r\n\r\nSecond — and this is a nuance that's cheap to ignore from an office with a view of the Dnipro. \"Exit fast\" carries a very different weight in California than it does in a frontline city, where the job market has shrunk to one factory and three coffee shops, and the person across the table has rent, two kids, and a mobilization context riding on their back. This doesn't cancel the stop-loss — it changes the severance package and the horizon. Write your kill-criteria equally honestly everywhere; but measure severance and job-search time against the market the person actually lives in, not against benchmarks from a quick online poll. Exit discipline with no exit conscience isn't trading anymore. It's just being cheap.
\r\n\r\nThird, and most unpleasant: a written card doesn't enforce itself. Ku, Malhotra, and Murnighan (Organizational Behavior and Human Decision Processes, 2005) looked at live and online auctions: people systematically blew past the limit they'd written for themselves before the bidding started — competitive arousal sweeps away a self-imposed threshold the moment a rival and a clock show up in the room. Your written kill-criterion, under quarterly pressure, is no better than the auction bidder's limit. That's why a guide carries a radio and has a person at base camp whose brain hasn't been cooking in the death zone. You need someone down at base camp too — a partner, a board member, an outside advisor, someone with nothing invested — who looks at the card once a month and asks exactly one question: what time is it.
\r\n\r\nThis isn't a checklist for \"how to fire someone.\" This is the Firing Kill-Criteria Card — a document you fill out on hiring day, not on the day doubt shows up. The logic is simple: on hiring day, you're not invested yet. You can still think like an investor, instead of like someone defending his own decision.
\r\n\r\nThree columns, to be filled in before the offer gets signed:
\r\n\r\n| Kill-criterion (observable, not moral) | \r\nReview deadline | \r\nWhat I'll tell myself then (name the alibi in advance) | \r\n
|---|---|---|
| Misses agreed scope with no warning for two sprints running | \r\nMonth 3 | \r\n\"The context was tough\" / \"I didn't explain it clearly enough\" | \r\n
| Three rounds of feedback showing the same pattern, with no change in trajectory | \r\nMonths 4-5 | \r\n\"He's trying\" / \"let's give it one more chance\" | \r\n
| I'm spending over 2 hrs/week thinking about this person in the context of a problem (not growth) | \r\nAny point at all | \r\n\"This is my responsibility as a manager\" | \r\n
The fourth row under \"What I'll tell myself\" is always the HR priest. Write him down in advance, so you can recognize him the moment he starts talking.
\r\n\r\nReview this monthly, not quarterly. The ratchet clicks faster than your quarterly review cycle. If two kill-criteria have already triggered, you're not \"thinking about whether to fire.\" You're already late. The only question left is by how much.
\r\n\r\nAnd one rule for the conversation itself: prepare it exactly like you'd prepare an investment decision. You're not saving the person from the truth — you're handing back their agency. \"We're exiting this position, because [specific, observable facts]. Here's the timeline. Here are the terms.\" No \"I believe in you,\" no \"this is really hard for me,\" no hidden subtext. The person in front of you is an adult — she deserves a direct conversation, not a piece of theater with coffee growing a skin on top.
\r\n\r\nAveraging Down on a Person isn't a weakness of character. It's a badly designed system, wearing the hat of morality.
\r\n\r\nYou don't have kill-criteria — so you can't exit on time. You can't exit on time — so you pay the ego tax. The ego tax accumulates — the team sees the standard isn't real. The team sees it — the bar drops. The bar drops — the next hire calibrates lower, because the new person sets their standard by what they observe, not by what's written in the job description.
\r\n\r\nThis isn't one bad quarter. It's compound interest, running in reverse.
\r\n\r\nOn the mountain, all of this arithmetic fits into a single night: 2pm, 4pm, the storm. At a company, that same night gets stretched across six quarters — and that's exactly why it's possible not to notice it, as long as you're paying for it monthly. Ask any founder who shut down and had the honesty to run a real post-mortem: where did it actually break? Most often, you won't hear about the product or the market — you'll hear about two or three moments when he should have exited a position, and instead said \"one more sprint.\" Not from a lack of information. From a lack of a stop-loss, written before entry.
\r\n\r\nThe till doesn't say \"I'm sorry.\" The till just decides for you, the moment the runway runs out. And by then the HR priest has long since gone home in his cardigan, and you're the one standing over the balance sheet with the phrase \"we did everything we could.\"
\r\n\r\nYou did. Just not the right thing, and not at the right time.
\r\n\r\n" }