Setting Kill Criteria Before You Start Something New

Why this matters

Shops rarely kill a new service line, location or role on time. They start it with energy, it underperforms, and it limps along for another year because there is always a reason to give it a few more months: the season was odd, the marketing just started, the tech is still learning. By the time it ends, it has used the owner's best hours and the cash that should have gone to something that worked. The way out is to decide, before you start, what result by what date ends it, and to write that down while you are still neutral. Kill criteria do not make the decision for you later. They make it possible to make it at all.

The idea

The former professional poker player and decision writer Annie Duke, in Quit (2022), argues that people quit too late far more often than too early, and that the remedy is kill criteria: a state and a date, set in advance, that trigger a decision to stop. "If we are not at X by Y, we stop." The reason it works is timing. Before you start, you can judge a result coldly. Once you are in, every month of effort makes the thing feel more valuable and stopping feels more like failure. The money already spent pulls the same direction; that trap has its own card (see related: Sunk Cost in a Bad Hire, a Bad Truck and a Bad Customer).

A kill criterion is not a target or a hope. A target is the result you want. A kill criterion is the result below which you would not have started if you had known it in advance.

The case

A six-person plumbing shop (owner, office manager, three techs, one apprentice) decides in January to add water-treatment work: softeners, filtration and reverse-osmosis systems, sold mostly to existing customers. One senior tech gets trained; the shop buys a small stock of units. Every step below is one field of the card the owner wrote before the first install, and the filled card follows the steps.

Step 1: Name the one number that says it is working

Pick a single primary measure that would have to be true for this to be worth doing. Not five measures; if you have five, the thing can always be "working" on one of them. Secondary measures are allowed as guardrails, but one number decides.

For the water-treatment line, the owner chose installs per month, averaged over the latest three months. Revenue would mix in large commercial outliers. Leads would count interest rather than sales. Installs are what pays back the training and the stock.

What breaks if you skip it: at review time, the champion of the new line points to whichever measure looks best. "Installs are down but leads are way up."

Step 2: Set the threshold from what it has to earn, not what you hope

The threshold is the level below which the line is not worth the capacity it uses. Work it out from what the line displaces. Here, the trained tech's hours on water treatment come out of regular plumbing service, which runs at about 40% gross margin and has a waiting list in summer. For an install line to beat that use of his time at the margins the shop expected, the owner calculated he needed about 8 installs a month.

Add guardrails that would end the line even if the primary number passed:

• Gross margin on water-treatment work at or above 35%, because a line that hits volume by discounting is not working.
• Owner time at or under 6 hours a week on the line after month 3, because the owner's hours are the scarcest thing in a six-person shop.

What breaks if you skip it: the threshold drifts down to whatever the line is actually doing. A line that does 5 installs a month gets described as "steady."

Step 3: Set the date, and make it a date on the calendar

Pick when the primary number is judged. Long enough for a fair test, short enough that a failure does not consume a year. For a new service line sold to existing customers, six months of selling is usually enough; a new location needs longer. The owner set June 30, judged on the April to June average, which leaves the first three months for training and ramp and does not count them.

Put the date in the shop calendar with the card attached. A kill criterion with no calendar entry gets reviewed when someone remembers, which is usually after the thing has gone worse.

What breaks if you skip it: "six months" quietly becomes "after the summer," then "after the holidays."

Step 4: Write the middle zone, and allow one extension only

Results rarely land cleanly above or below a line. Decide in advance what happens in between. The owner's rule:

• At or above 8: continue, and set the next review for December.
• 5 to under 8: one extension of three months, with one named change to try, and a new date.
• Under 5: stop.

"One extension only" is the most important clause on the card. Without it, the middle zone becomes a permanent home: every review lands in the middle, every review grants three more months.

Step 5: Name who decides, and make it not the champion

The person who championed the new line, often the owner, is the worst-placed person to judge it alone, and the one most likely to find a reason to extend. Name a second person whose agreement is needed to override the card: a partner, the office manager, the bookkeeper, a peer owner. They do not make the call; they hold the owner to the one already made.

The owner named the office manager, who also tracked the installs in the job system. That made the number hard to argue with: she was reading the same report the card referenced.

Step 6: Decide what stopping means before you need to

"Kill" is not one action. Write down what winding down involves, so stopping is a plan rather than a defeat: what happens to customers already sold, to the stock, to the trained tech's time, to any service agreements. A clear exit makes it easier to pull the trigger, because nobody is imagining an abandoned customer.

For water treatment: honor every warranty and service agreement already sold, keep servicing installed units, sell remaining stock back to the supplier or use it on service calls, and move the tech's hours back to the plumbing waiting list.

The card

KILL CARD - WATER TREATMENT    Written Jan 8
Primary:   installs/month, 3-month avg
Threshold: 8 or more = continue
           5 to under 8 = ONE extension,
             3 months, one named change
           under 5 = stop
Guardrails: margin 35%+, owner <= 6 hr/wk
            after month 3
Judged:    June 30, on Apr-Jun average
Decides:   owner + office manager
Stop means: keep warranties and service,
  return stock, tech back to plumbing

The card took about 40 minutes to write. Notice what it leaves out: no revenue target, no lead count, no "if the season is unusual" clause. Each of those is a door the champion could use to walk around the primary number, and each was considered and dropped on purpose.

What happened

April, May and June installs: 4, 5 and 7. 4 plus 5 plus 7 is 16, and 16 divided by 3 is 5.3 a month, inside the 5-to-8 band. Margin came in at 38%, which passed, and owner time at about 9 hours a week, which failed its guardrail.

The card said one extension. The owner and office manager named the single change: stop the owner selling systems personally, and have the trained tech offer a free water test on every plumbing call in hard-water neighborhoods. That also attacked the failed owner-time guardrail. New date: September 30, on the July to September average, threshold still 8.

July, August and September: 6, 5 and 7. 6 plus 5 plus 7 is 18, and 18 divided by 3 is 6.0 a month. Better, still under 8, and the card allowed no second extension.

This is the moment the card exists for. The owner's honest reaction was that the line was "just getting traction," and the trend from 5.3 to 6.0 was real. But at roughly 0.7 more installs a month per quarter, it would take about three more quarters to reach 8, if the trend held. The card had already answered the question of whether that was worth waiting for, back in January, when the owner was not attached to the answer. The line was wound down in October as written: warranties kept, stock returned, the tech back on plumbing service with a three-week summer backlog to work off.

What changes the answer

New information, not new hope. If something outside the shop changes the economics after the card is written, such as a utility rebate program for water treatment that cuts the customer's cost substantially, it is legitimate to rewrite the card. The test: would this information have changed the threshold in January? A rebate would. "We are getting better at it" would not; that was always expected.

A line with a long sales cycle. A commercial service or a second location may need a year before the primary number means anything. Then set an earlier leading indicator criterion (quotes sent, proposals accepted) at month 4, so you are not flying blind for twelve months.

A guardrail that fails alone. If the primary number passes and a guardrail fails (installs at 9 but margin at 28%), that is not a pass. Decide in advance whether a guardrail failure triggers the middle zone or a stop. This card did not say, and the owner had to make that call on the day, which is exactly what the card was meant to prevent.

How to tell your card will hold

You can state the primary number and its threshold in one sentence. The date is on a calendar. The middle zone says "once." Someone other than the champion has to agree to override it. And you would accept the stop outcome if it came tomorrow; if reading your own card makes you want to soften the threshold, it was set honestly and should stay where it is.

References

• Duke, A., Quit: The Power of Knowing When to Walk Away (2022)
• See related: Sunk Cost in a Bad Hire, a Bad Truck and a Bad Customer
• See related: Running a Small Experiment Instead of Arguing
• See related: Stress-Testing a Plan With Three Scenarios
• See related: Forecasting a New Service Line From a Reference Class