What a kill criterion actually is, and why most founders skip it.

Startup kill criteria: the framework, the common mistakes, and the evidence that separates a defensible answer from a confident one.

A kill criterion is not a risk factor. It is not a bullet point that says "competition is intense" or "regulation could change." Those are conditions. A kill criterion is a specific, testable statement of the form: if X is true, the business cannot work as described.

The difference matters because risks are infinite and largely inarguable. Kill criteria are finite and testable. You can go find out whether they are true.

Here is an example. Suppose you are building a workflow tool for compliance teams at mid-market fintechs. A risk might be: "Large competitors have more sales resources." A kill criterion is: "No compliance officer at a company between 50 and 500 employees controls a budget above $12,000 per year for new software tools." If that is true, the price point that makes the unit economics work cannot be charged to the buyer who would benefit most. The business model breaks, not because of competition, but because the buyer does not exist at the required price.

Naming a kill criterion forces you to answer three questions: who is the buyer, what do they control, and what does this product need to cost to be a real business. Most founders answer the first question. Fewer answer the second. Almost none answer the third before building.

The reason kill criteria get skipped is not laziness. It is framing. Founders are told to be optimistic, to find reasons to believe, to construct the bull case. The investor's job is to poke holes, and the founder's job is to defend. That adversarial split means founders spend their preparation time building the case for the idea rather than pre-empting the cases against it.

The problem is that an investor who identifies a kill criterion you have not addressed will not fund you. And an investor who misses it will fund a company that later fails for a reason that was visible from the beginning.

There is a better framing. Before you pitch, name the three conditions that would make this idea not work. Not risks. Conditions. If the addressable market is actually 3x smaller than the figure you cited, does the model break? If the sales cycle to your target buyer is 14 months, does the cash runway hold? If the largest competitor copies the core feature in 90 days, is there anything left?

If you can name those conditions, you can address them. You can find data that contradicts them. You can restructure the model to survive them. You can pivot before you burn 18 months finding out the hard way.

Verdikt names kill criteria in every report because that is the information a founder actually needs. Not a score. Not a heat map. A specific, falsifiable statement about the conditions under which the idea fails. With that in hand, you can either disprove it or decide not to proceed. Both are good outcomes.

The memo you should be able to defend is the one that names the ways you are wrong, not just the ways you are right.

A pre-seed founder building a vertical SaaS for boutique law firms. The kill criterion is: "If fewer than 6 of 25 firms we cold-call in the next 30 days agree to a 20-minute discovery call about the workflow we are pricing at $400 per seat per month, the wedge is wrong." That is specific. The 25-firm sample is defined. The 6-firm threshold is the bar. The price band is the variable. After 30 days, you have evidence one way or the other, and you can keep going, change the price, change the wedge, or stop. The Mom Test by Rob Fitzpatrick is the cleanest discipline for running these conversations without leading the witness.

A seed-stage founder claiming a 10× advantage in latency for an AI dev tool. The kill criterion is: "If our p95 latency on the same prompt set is within 1.4× of Cursor by the end of Q2, the latency claim does not survive and we re-pitch on workflow specialization." Benchmark grids beat positioning slides every time. You cannot argue with a measured number.

A Series A founder pitching network effects. The kill criterion is: "If month-over-month new-user activation does not improve as the network grows from 1,000 to 10,000 active accounts, the network claim is decorative and we should pitch on switching costs instead." NfX has a useful taxonomy of network effects in their NFX Manual that distinguishes direct from indirect from data network effects, and the wrong taxonomy hides the wrong claim.

A risk is a thing that might happen. A kill criterion is a measurable threshold at which a thing has happened. "Competition is a risk" is a risk. "If GitHub Copilot ships Swift-specific tuning in their next release, the latency moat collapses" is a kill criterion. Investors at the Series A stage will not push back on risks because risks are universal. They will push back on the absence of a named threshold because the absence tells them you have not done the work to know when the answer changes.

The other useful distinction is between a kill criterion and an assumption. Steve Blank’s customer development frame in The Four Steps to the Epiphany separates hypothesis from test. The hypothesis is the assumption. The test produces the data. The kill criterion is the threshold at which the data falsifies the hypothesis. The three live together. You do not get to skip writing down the kill criterion and still claim the test was useful.

AI research tools, including Verdikt, produce a lot of evidence quickly. That cuts both ways. It accelerates the work, and it accelerates the temptation to produce a confident memo before the kill criteria are named. The discipline that holds the work together is naming the threshold before running the test, not after. A confident memo with no falsifier is the textbook output of an AI tool used badly. A defensible memo with named kill criteria is what the same tool produces when the human running it has done the work first.

This is the same discipline that distinguishes Verdikt’s methodology from a generic AI summary. Every Verdikt verdict ships with kill criteria named on the cover page, the thresholds at which the recommendation would change, and a re-run hook for the three weakest claims. The memo is not the end of the work. It is the start of a watch.