Kill Criteria Belong in the Project Brief
A project brief should state the evidence that earns another 30 days and the conditions that end the work. Precommitment keeps optimism from rewriting the test after results arrive.
Founder
6 min
The short version: Put the exit conditions in the project brief before the team becomes attached to the project. Define what must be true by day 30, day 60, and day 90; cap the money and founder time available for each stage; and pre-authorize four decisions: continue, narrow, pause, or kill.
Most briefs are excellent at describing the future if the project works. They list the opportunity, milestones, team, launch plan, and upside. They say much less about what weak evidence will look like. That omission turns normal optimism into a governance problem. Once people have shipped code, made a deck, or recruited a pilot, stopping feels like admitting failure rather than completing the test.
Write the death conditions while everyone is still optimistic
The first page of the brief should contain the thesis in one sentence and the falsifier directly underneath it. A falsifier is observable evidence that would make the team stop funding the current shape of the project. It is not “the team loses confidence” or “market conditions change.” It is a missed behavior, economics, distribution, or risk threshold on a fixed date.
This sits below the company's default-alive runway model. Runway tells the founder how long the company has to change trajectory. Kill criteria stop one attractive experiment from quietly consuming that time.
The one-page precommitment block
Thesis: which customer, painful job, and repeatable behavior the project expects.
Proof: the smallest observable behavior that supports the thesis.
30/60/90 gates: dates, thresholds, and the system where evidence will be recorded.
Budget: cash, external commitments, founder hours, and team capacity released for each stage.
Owner: one person who presents the evidence and one person with authority to end the work.
Non-negotiables: trust, security, legal, margin, or brand boundaries that trigger an immediate stop.
Salvage path: code, research, relationships, content, or process improvements worth preserving after a kill.
My hard rule is that two consecutive missed gates end the current project shape. The team may propose a narrower thesis, but it cannot rename continuation as a pivot without a new brief, new budget, and new clock.
Day 30: prove the problem, not the presentation
The day-30 gate should test whether the named customer recognizes the problem strongly enough to change behavior. Interviews count only when they produce evidence tied to the thesis. A polite “interesting” is not proof.
For an illustrative B2B workflow project, I might precommit to 15 interviews with the target role, at least five examples of the same costly workaround, and three organizations willing to provide real data for a constrained pilot. If the interviews reveal five different problems or nobody will share a real workflow, the choices are narrow or kill. Building more screens does not answer the question.
Day 60: prove repeated use under real conditions
By day 60, the team should have moved from stated pain to observed behavior. Define the action that demonstrates value: a weekly report reviewed, a recommendation accepted, an exception resolved, or a manual step removed. Track the burden around the action too.
In the same example, the gate could require three live pilots to complete the workflow for four consecutive weeks, at least 70% of outputs accepted without full rework, and no more than 30 minutes of manual repair per account per week. These are example thresholds, not universal benchmarks. Their purpose is to stop the team from celebrating logins while hidden service labor carries the product.
This is especially important when deciding when not to productize a service. If customers value rare judgment and the exception queue refuses to shrink, better internal tooling or a better-packaged service may be the winning outcome.
Day 90: prove an economic and distribution path
The day-90 gate asks whether the behavior can support a business or a strategically justified capability. Require evidence of willingness to pay or an explicitly valued internal outcome, a credible acquisition or rollout path, and fully loaded delivery economics. Review cost per completed task, not only software subscriptions: include review, support, retries, data repair, founder intervention, and incident work.
An illustrative gate might be two paid conversions from the pilots, delivery cost below 35% of revenue at the tested scope, and one repeatable channel that produced at least five qualified opportunities. Missing one threshold may justify narrowing. Missing the behavior gate again after day 60 ends the project shape under the two-strike rule.
Do not move the goalposts in the review meeting
Teams often discover a “better metric” the week an original metric misses. Sometimes they are right. The protection is procedural: preserve the old threshold in a decision log, record why it was flawed, show the evidence available before the result, and make any extension a new allocation decision.
I allow one extension of up to 14 days only when a named external dependency delayed the test and the underlying customer signal remains intact. More time for polishing, recruiting a broader segment, or changing the success metric is not an extension. It is a new experiment and should compete for resources as one.
What a clean kill looks like
Stop new spend and revoke production access on the decision date.
Notify pilots and partners with a clear transition or data-return plan.
Archive the brief, raw evidence, actual costs, and the final decision.
Extract reusable assets without keeping the project alive through maintenance.
Return named people, budget, and calendar space to the portfolio.
For a portfolio such as Equinox Ventures, this is attention allocation as much as capital allocation. Work across ventures such as AI Synergy may leave useful infrastructure even when one product thesis dies. The salvage path lets the learning compound without letting the project linger.
A kill is not the opposite of conviction. It is conviction about the test. The brief promised that a specific amount of time and money would buy specific evidence. When the evidence arrives, honor the promise. Founders do not need less optimism; they need a date when optimism has to face the numbers written before the story became personal.
