When Do We Stop? Writing the Kill Criteria Before You Write the Cheque

Nine other decisions in this series are about starting well. This one is the cheapest of the ten and the one skipped most often, because on the day a clinic pilot begins nobody wants to talk about how it ends.

That is exactly when it costs nothing to say. In month nine it costs someone their reputation, which is why in month nine it does not get said at all.

Projects do not fail, they fade

A clinic AI pilot rarely produces a clean negative. It produces a partial: two clinicians use it, the number moved a little, flu season was unusual, and the next quarter would be a fairer test. Every one of those statements is usually true, and together they are indistinguishable from a project that is not working.

Without a written ending, continuing is the default and stopping requires an act of will from a named person who will be remembered as the one who ended it. Nobody volunteers for that, so the renewal quietly happens. The cost is not only the licence. It is that the next proposal arrives into a partners’ meeting that has learned these things do not conclude.

What drift actually costs

A pilot nobody ended

$28,800

$2,400 a month, month after month

12 months · $28,800 6 months · $14,400 24 months · $57,600

The licence is the small half. The larger half is the attention of whoever keeps half-owning it, and the credibility the next project inherits.

A stop condition is three things

Not a feeling that it is not working. A sentence: the number, the level, the date.

“If notes are not being closed the same day by October 31, we stop.” That is checkable by anyone, requires no interpretation, and converts an uncomfortable judgement into arithmetic. It also needs a fourth element that most versions omit: who is allowed to call it. A criterion nobody owns is one everyone assumes somebody else is watching, which is the same failure as an AI system with no named owner arriving one level up.

This only works if the number was baselined before you started, which is the discipline a proof of value exists to enforce. Without a baseline there is nothing to measure against and every result is arguable, which is its own way of never having to stop.

The result is in. What does it mean?

Notice that two of the three endings keep something. Stopping is not the opposite of succeeding, it is one of three ways a well-run experiment concludes, and the only genuinely bad outcome is the fourth one where nobody decides.

A project that cannot end cannot really be evaluated. Writing the ending is what makes the rest of it a decision rather than a commitment.

What to do this week

Write one sentence for the project you are running or about to approve: if this number is not at this level by this date, we stop. Send it to whoever owns the project, so it exists somewhere other than your head.

If you cannot fill in the number, you have found something more useful than a stop condition. You have found that the project does not yet have a definition of working, which is the same gap the first decision in this series was meant to close, and it is far cheaper to discover today than in month nine.

---

Paying for something you suspect should have ended already? Book a free consult and we will look at the number, the baseline, and the honest ending with you, including whether the right answer is to shrink it to the one clinician it works for rather than kill it. See more on Second Opinion.