- GOVERNS
- A product bet
- CRITERION 1
- METRIC
- ____________________
- THRESHOLD + WINDOW
- ____________________
- ACTION
- ____________________
- CRITERION 2
- METRIC
- ____________________
- THRESHOLD + WINDOW
- ____________________
- ACTION
- ____________________
- CRITERION 3
- METRIC
- ____________________
- THRESHOLD + WINDOW
- ____________________
- ACTION
- ____________________
- WITNESS
- ____________________
- REVIEW DATE
- ____________________
The kill criteria worksheet
Decide when you would quit before you have to.
Kill criteria written mid-crisis are negotiations. Kill criteria written today are commitments. Write yours in ten minutes, while nothing is on fire.
Write your kill criteria
Step 1: What do these criteria govern
One document per bet. If two things could die separately, they get separate criteria.
Step 2: Write the criteria. One is the minimum. Three is the working number.
Criterion 1
Worked exampleMETRIC: weekly active users of the beta. THRESHOLD + WINDOW: under 150 by June 30. ACTION: kill the bet, fold the learnings into the core product.
What you will measure. A number a stranger could check.
The line and the date. Below X by DATE.
What happens when it trips. Kill, pivot, or escalate to a named person.
Criterion 2
Worked exampleMETRIC: activation rate from signup to first real use. THRESHOLD + WINDOW: below 20% for 8 consecutive weeks. ACTION: pivot the onboarding once, then kill.
What you will measure. A number a stranger could check.
The line and the date. Below X by DATE.
What happens when it trips. Kill, pivot, or escalate to a named person.
Criterion 3
Worked exampleMETRIC: monthly revenue attributed to the feature. THRESHOLD + WINDOW: under $2,000 by the end of Q3. ACTION: escalate to the head of product for a kill call that week.
What you will measure. A number a stranger could check.
The line and the date. Below X by DATE.
What happens when it trips. Kill, pivot, or escalate to a named person.
Step 3: Commit
Who sees this document today. Criteria nobody witnessed are criteria you will quietly edit.
The calendar day this document gets read again. A day, not a quarter.
Fill at least one criterion completely: metric, threshold, and action. The blank lines print blank.
Why criteria written early are the only ones that count
The version of you that will read this document in eight months is not the version writing it. That one will be six figures and two identities deep into the bet, and the sunk cost will do its standard work: every line that can be moved will be moved, every threshold that can be reinterpreted will be reinterpreted, and every deadline that can slide a quarter will slide a quarter. None of this will feel like cheating. It will feel like nuance, and updated context, and knowing things the old document could not have known.
Dating and witnessing the document removes the permission. A criterion with a date on it and a name beside it cannot be quietly rewritten, because quietly is no longer available: someone else read the original. That is the entire difference between criteria and intentions. Intentions are renegotiated in private by whoever holds them. Criteria are commitments with a witness, and the witness is not there to enforce anything. The witness is there so that moving the line has a cost you can see in advance.
Good criteria are boring
A good kill criterion reads like a compliance rule, not a mission statement. Cost per qualified lead above $80 for six consecutive weeks, kill the channel. No adjectives, no vision, nothing that would survive in a pitch deck. If a criterion needs interpretation, it will get interpretation, and the interpreting will be done by the person with the most invested in continuing, which is you, at the worst possible moment, which is when it trips.
That is why the examples above are dull on purpose. Every one of them is a number a stranger could check, a line with a date on it, and an action with a name attached. The test for each of yours is the same: could someone who has never met you read the criterion, look at the dashboard, and tell you whether it tripped. If the answer is no, the criterion is not finished. Boring is not the price of a good kill criterion. Boring is the evidence of one.
Kill criteria questions, answered
What are kill criteria?
Kill criteria are the conditions under which you will stop, written down before you start. The term comes from Annie Duke's work on quitting: decisions to stop run systematically late, because sunk cost, endowed identity, and status quo bias all pull toward persistence. The fix is states and dates set at deciding time. Each criterion names a metric, a threshold with a window, and the action that executes when the line is crossed.
What makes a good kill criterion?
Three parts, all of them checkable. A metric a stranger could verify: a number, not a feeling. A threshold with a hard date: below X by a calendar day, not by Q2-ish. And a named action: kill, pivot, or escalate to a specific person, not a meeting to discuss. If any part needs interpretation, the interpreting will be done in favor of continuing. A good criterion reads like a compliance rule and trips like one.
How are kill criteria different from tripwires?
Same discipline, different scope. A tripwire is the exact state and the exact date that would reopen a decision, named in advance, in writing, while you are cold, and it fires in both directions: wires can trigger doubling down as well as killing. Kill criteria are the subset pointed at stopping: the conditions under which the bet dies. The tripwire pattern comes from Annie Duke's kill criteria, so on this site every kill criterion is a tripwire whose named action is kill.
How often should I review kill criteria?
On the review date you wrote in the document, and at a standing cadence in between: weekly for a channel or campaign, monthly for a product bet or hire, quarterly for a startup. The check takes one minute per criterion: tripped or not. What you do not do is review them every bad morning. A criterion that has not tripped is a settled question, and doubt without a tripped wire is weather.
Get the Sunday issue.
One essay every Sunday on the decisions that define where your startup and your life actually go. Free.
No spam. Join 58,000+ founders.