Write the Stop Condition Before the Project Starts

Ask a leadership team how an initiative began and the answer is usually a date, an owner and a slide. Ask how the last one ended and the answer is vaguer. It stopped getting airtime. Its owner moved to something else. The budget line was folded into next year's plan without a decision that anyone could point to. Most initiatives are not stopped. They are left to expire, and expiry is slower and more expensive than a decision would have been.

How a project outlives its reason

The reason an initiative was started is usually a belief about the world: that customers would adopt a feature, that a new segment would convert at a workable rate, that a platform migration would free a team by the third quarter. Beliefs of that kind get tested by events, and events rarely arrive as a verdict. They arrive as a slightly weaker adoption curve, a slipped milestone, a workaround that needs one more sprint. Each is small enough to explain away, and each explanation is offered by the person whose name is on the project.

An initiative is rarely wrong all at once, so it is never obviously wrong on any given Tuesday. The evidence accumulates below the level at which anyone is willing to say it aloud, and the spending continues in the meantime. Priority lists grow because adding an item costs a leader nothing today. The mirror image is that removing an item costs something in the room, at once, and so the removal keeps being postponed.

The remedy is not a braver leadership team. Courage in the sixth month of a project is expensive and unevenly distributed. The remedy is to move the decision to a point where it costs nothing, which is the moment before the work starts, when the sponsor has no sunk cost to defend and no peers watching them concede.

What a usable criterion looks like

A kill criterion is a sentence with a condition and a date. It names something observable, the level it has to reach, and the point by which it has to reach it. Three kinds cover most cases.

• An evidence test. By the end of the second month, a named number of pilot customers is using the feature without being prompted, or the premise that they would has failed.
• A cost ceiling. The initiative has a fixed amount of budget and senior attention, and spending all of it without the milestone it was meant to buy is the stop.
• An assumption that no longer holds. The case rested on a partner, a regulation, a pricing position or a particular person, and if that one thing changes, the case is reopened before another week is spent.

The wording matters more than it looks. "The project will be reviewed if it runs into trouble" is not a criterion, because trouble is defined afterwards by the people who most want to continue. "If fewer than four of the ten pilot accounts are using it weekly by the end of March, it stops" can be read by anyone, including the person it is about.

Teams that do write criteria tend to spend their argument on the threshold, four accounts or six. The threshold is the negotiable part. The date is the part that does the work, because a criterion without a date is a promise to look at the project at some unspecified point, and that point does not arrive.

A review date is a decision date

The date has to be a real slot in the calendar, in the meeting where trade-offs for that period are already being made, so that the question of whether an initiative continues is asked alongside every other call on the same people. That ties the stop rule to the cadence rather than to a special occasion. A review that needs its own meeting is the first thing moved when the week gets busy.

Two outcomes are allowed

At the date, the project continues because the condition was met, or it stops because it was not. The outcome that has to be refused is the extension, granted because the team argues that it was close. An extension without a new criterion is a decision to delete the old one. If the evidence is honestly ambiguous, the response is a second, narrower condition with its own date, agreed in the same meeting.

Even with a criterion agreed in advance, the meeting where it is applied is uncomfortable. A team has spent months on the work, and the sponsor is hearing that the premise did not survive. The discomfort does not go away with better process. What changes is what it is about. Without a criterion, the room is deciding whether the sponsor was wrong, which is personal. With one, the room is checking whether a condition both sides wrote down was met.

A stop that follows a written condition is a matter of record rather than a verdict on a person.

What the stop releases

The other half of the decision is what happens to the work and the people. A stopped initiative usually leaves something usable behind: research, a partly built component, a customer conversation that changes another team's plan. Naming what is salvaged in the same meeting turns the stop from a loss into a reallocation. The released hours need a named destination as well. A leadership team can carry only so many priorities, and capacity freed without an assignment refills with whatever request is loudest that week.

The rule also settles a quieter problem. Where two live initiatives contend for the same team, the one whose condition fails first gives way, and neither sponsor has to argue seniority to get the outcome.

Most companies have a rule for starting an initiative: a business case, an approval, a budget code. Few have a rule for ending one. The result is a portfolio in which the oldest projects carry the weakest evidence and the strongest attachments, and the senior team's attention goes into maintaining them.

The first piece of work is small. Take the current initiative list and write, beside each item, the condition under which it would stop and the date on which that condition is read. The items whose sponsor cannot write one are the ones to look at first. That exercise sits inside our consulting work on growth discipline, and it tends to produce the first honest cut of the list without an argument about which initiatives are worth less.