Kill Criteria: Decide When to Stop Before You Start

Kill criteria are the conditions, written down before a project begins, under which you will stop it. A time limit, a cost ceiling, and the specific result that would mean no. Set in advance, they turn stopping from a judgement made under pressure into a check against a standard you already agreed with yourself.

Most projects have none. They have a start date, a hope, and a vague intention to "see how it goes." That phrase is where the trouble begins.

How do you know when to kill a project?

You know when the evidence you set out to find has arrived, or failed to arrive by a date you fixed in advance. You do not know by feel, and you cannot know by feel, because feel is the instrument that has been compromised.

Consider what the decision looks like from inside. Six months in, the numbers are mediocre. Not catastrophic, which would make the call easy, but mediocre. There is a plausible story in which the next quarter turns it around. There is also a plausible story in which it does not. You have two readings of the same data and no neutral party to choose between them, because the only person qualified to judge is the person who has spent six months and a good deal of money on it.

So you give it another month. Then another. Each extension feels like prudence. None of them is a decision. They are the same deferral, renewed.

Why is it so hard to stop once you've started?

Because the investment changes what you are able to see.

The mechanism has been studied for decades. In 1976 Barry Staw ran an experiment in which participants allocated research funds to a division, were told the investment had done badly, and were then asked to allocate again. Those who had personally made the first allocation committed more money to the failing division than those who had not. Same facts. The only difference was ownership of the earlier choice. Staw called it escalation of commitment, and the finding has been replicated in many forms since.

A neater version came from Hal Arkes and Catherine Blumer in 1985. People who had paid full price for a season theatre subscription attended more performances than people who had been given discounted tickets, even though the seats and the plays were identical. Money already spent, and unrecoverable, was steering future behaviour. Economists call it the sunk cost fallacy. It is a fallacy because the spent money is gone whatever you do next, and the only rational question is what the next hour or pound will return.

The trouble is that knowing this does not protect you. You can recite the sunk cost fallacy and still be inside it, because the pull does not feel like a bias. It feels like loyalty to a reasonable plan, or a reluctance to waste what you have built. Concorde is the standard public example. Britain and France kept funding the aircraft long after the commercial case had collapsed, and the project gave the effect its other name in those countries: the Concorde fallacy. Two governments with every analytical resource available could not stop. A founder with a Tuesday afternoon and a hunch has no better odds.

Why don't willpower and good intentions fix it?

Because the person doing the judging is the person with the stake in the answer.

The usual advice is to be more objective, to step back, to ask what you would do if you were starting fresh. It is sound advice and it asks the compromised party to audit itself. You are being told to discount your own investment using the faculty that the investment has distorted.

The advice also arrives at the wrong time. By the point you are asking whether to stop, the cost of stopping has already risen: the sunk money, the public commitment, the story you have told colleagues or customers, the identity you have attached to the work. Each of these makes an honest "stop" more expensive to say aloud. A decision that is expensive to make is a decision that gets postponed.

Solo operators feel this most sharply. There is no board to overrule you, no budget committee to pull funding, nobody whose job is to say that it has run long enough. The only brake is your own, and you are the person least able to apply it.

What should kill criteria actually contain?

Three things, fixed before you begin, written somewhere you cannot quietly revise.

The first is a time box. A date on which the project is reviewed, chosen while the work is still an idea and your attachment is low. Not "a few months." A calendar date.

The second is a cost ceiling. The most you will spend in money, hours, or both, before the project has to justify itself. Hours matter as much as money, and they are the ones that go unrecorded. If you will not write the number down, you have not set a ceiling; you have set a mood.

The third is the signal that means no. This is the one people skip, and it carries the most weight. Name the specific result that, if you see it by the review date, ends the project. Ten paying customers. A reply rate above a stated level. A deadline met. Choose something observable, and choose it before you know the outcome, so that it cannot drift to fit what happens. A threshold you invent after the data arrives is not a threshold. It is a rationalisation with a number in it.

Then make the review binary. On the review date there are two options: continue or stop. There is no third option called "give it another month." If the criteria were met, continue, and set the next set of criteria. If they were not, stop. Extensions are how kill criteria die, and the discipline lies in refusing the third door.

What does stopping well look like?

It looks undramatic. You open the note you wrote at the start, you compare it to what happened, and you do what it says. Nothing about the result needs to be a verdict on you. The project was a test, the test returned an answer, and you acted on the answer.

It also frees capacity. A project that should have ended and has not is not a neutral object sitting in the corner. It is an open loop that takes attention every time you pass it, a standing question about whether today is the day to decide. Closing it returns that attention to work that has earned it.

None of this requires you to become more ruthless or less sentimental. It requires you to move the hard decision to the moment when it is cheap, which is before you start, and to leave yourself a record that the later, more invested version of you cannot argue with.

That is the work Decision Resolution exists to do. It is a written framework for fixing the criteria in advance, including the exit conditions, so that stopping becomes a check you run rather than a verdict you agonise over. The judgement you apply on day one is cleaner than the judgement you will have on day two hundred. Build the structure while it is still available.