Exit Criteria: Why Your Future Self Needs a Pre-Written Permission Slip to Quit

You’re launching a pilot. New onboarding flow, run with a slice of customers for one quarter, and going in you set the bar: if it doesn’t lift activation by ten points by the end of March, you pull it and go back to the old flow. It’s a clean line, and an easy one to say in January, when you have nothing riding on it yet.

March arrives and activation is up four points against a target of ten. But the team has learned a lot, the last two weeks trended better than the first six, and one enterprise account loved the new flow enough to mention it on a call. So you give it another six weeks. At the new deadline it’s up six, and you’ve just shipped a fix you’re sure will close the gap, so you let it run to the end of the next quarter. Somewhere in there the pilot stopped being a pilot. It’s just the onboarding flow now, underperforming the thing it replaced, and nobody ever decided to keep it.

Look closely at what happened. You never made the call to continue past your bar. There was no meeting where you looked at four points and chose to override ten. The bar moved on its own, a little at a time, and every move felt local and sensible. That drift is the thing to understand. It’s what the decision produces when you make it live, one sensible step at a time, with no single failure of nerve to point to.

The evidence

Start with how escalation actually happens, because it doesn’t happen the way the cautionary tales suggest. People stuck with a failing course of action rarely sit down and consciously decide to throw more after it. What they do instead is make a run of small decisions, each one marginal: a little more time, a little more budget, a slight loosening of what counts as success. No single increment is big enough to trip a real reassessment. The sum is enormous and you never see it land, because it never lands all at once. This is escalation of commitment, and goalpost drift is its engine. The conditions for quitting are never met because the conditions keep quietly getting redefined.

Several forces push the redefinition. One is self-justification: continuing keeps your original call looking right, while stopping concedes it was wrong, so your future self has a standing interest in finding reasons to go on. Another is completion proximity, the sense that success is just around the corner, which is treacherous precisely because it renews itself at every stage. At four points you’re sure six is reachable; at six you’re sure ten is. A third is identity. The longer the pilot runs, the more it tangles with how you see yourself as the person who launched it, so pulling it starts to feel like a verdict on you rather than a read on the numbers.

Now the counter-move, and it’s better supported than most people expect. Plans built in the form “when situation X happens, I will do Y” produce far higher follow-through than general intentions to do the right thing. These are implementation intentions, and they work by wiring a specific trigger to a specific action ahead of time, so that when the moment comes you don’t have to deliberate your way to the action. The deliberation is exactly the step that drift hijacks. The broader work on commitment devices, the arrangements people use to bind their own future behaviour, points the same way: pre-commitments that are specific, time-bound, and tied to something you can observe beat vague resolve by a wide margin. The active ingredient is always that the decision was made in advance and pinned to a condition, not left to be argued in the moment.

How it works

The reason this matters comes down to the gap between two versions of you. You decide very differently in a cold state, calm and uncommitted, than in a hot one, invested and attached. Cold, you can look at a pilot that’s six points short and say plainly that it should stop. Hot, with your quarter and your credibility and your sense of yourself wrapped into it, that same conclusion is almost unreachable. And the stop decision always falls due in the hot state, long after the cold judgement that should have governed it.

A kill-condition set in advance is how you move the decision out of the hot moment and back to the cold one. You write it when you have no skin in the game, and you hand the hot version of you a verdict that’s already been reached by the clearer version. That only holds if the condition is specific enough to resist reinterpretation. “I’ll stop if it’s not working” hands your future self unlimited room, because every small positive signal becomes evidence that it is, in fact, working. “I’ll pull it if activation is below a ten-point lift on 31 March” leaves no room, because the number is either met or it isn’t and the date either comes or it doesn’t. What you’re protecting against is good judgement applied to a question that’s already been bent by the time you reach it.

The point of writing the kill-condition down is to let the clear-headed version of you decide the question once and bind the invested version who has to live with the answer.

How to use it

Write the kill-condition before the pilot starts, in the same breath as the success target. The moment you have no sunk hours to defend and no story to protect is the only moment you’ll author it honestly. Phrase it so a stranger could rule on it: a metric, a threshold, a date. “If activation isn’t up ten points by 31 March, the pilot ends and we revert.” Not “if it’s underperforming,” which is just an invitation to argue. Every bit of vagueness you leave in is room your future self will use to keep going.

Separate the decision from the feeling. When the condition fires, the feeling will say one more sprint, the next fix changes everything, we’re so close. The condition says stop. Honouring it doesn’t mean the cold version of you is always right. It means that overriding it should cost something deliberate: a written note, with new evidence that wasn’t available when you set the line, reviewed by someone who isn’t you. If the only thing that’s changed is that you’re now attached, you don’t have new evidence at all, just the drift showing up exactly when the line was built to catch it.

Build the override clause on purpose, because pilots do sometimes surface something that genuinely rewrites the case. The flow flops on activation but uncovers a support saving that dwarfs it, and you should be able to act on that. The point isn’t to be rigid, it’s that a real override clears a real bar: name in advance who can authorise it, and what kind of evidence counts. When the move from “stop” to “continue” needs a different person’s signature and a fact you didn’t have in January, you’ve kept the escape hatch without leaving it permanently propped open. Most of the time the hatch stays shut, and that’s the whole value.

Why it matters

The professional world runs on stories of people who refused to quit, and almost never on the ones who quit the wrong thing in time. So persistence gets read as character and stopping gets read as weakness, which is exactly backwards for anyone running real experiments. A pilot is a bet you took because you weren’t sure. Saying out loud what failure looks like, before you’ve spent anything, is the honest admission that you might be wrong and that you’ve thought clearly enough about being wrong to name it in advance.

A pre-written kill-condition doesn’t buy you discipline in the hot moment, which you mostly won’t have. It buys you a way to skip needing that discipline at all. The clear-headed version of you, the one with nothing on the line, makes the call once and pins it to something the invested version can’t renegotiate. The pilots that drift into permanence, draining a roadmap quarter after quarter, almost never get one bad decision behind them. They get a hundred reasonable ones, and no line anyone drew early enough to make the hundredth one visible.