Kill Criteria for Startup Ideas: Set Them Before You Start

Kill criteria are the specific, pre-committed conditions that tell you to drop a startup idea before you sink another quarter into it. You write them while you're still skeptical, before conviction, sunk cost, and identity make quitting feel like personal failure. Set them first, and a weak idea dies in days instead of years.

Quick Answer: Kill criteria are pre-committed thresholds, for demand, unit economics, distribution, and your own motivation, that trigger a stop. Define them before you start, make each one testable and dated, and give someone the standing to hold you to them.

Why founders need kill criteria before conviction forms

You need kill criteria before conviction forms because the founder who has already fallen for an idea can no longer read the evidence that it's failing. Conviction is an asset when you're executing a validated plan. It's a liability when you're still deciding whether the plan deserves to exist.

Here's the trap. Every day you work on an idea, you accumulate reasons to keep working on it: hours spent, a name you like, a pitch you've told friends, a small win you can point to. None of those are demand. But all of them raise the emotional price of walking away, so you quietly lower your standard for what counts as "working."

By the time the idea is obviously in trouble, you're the least qualified person in the room to call it. That's not a character flaw. It's the predictable result of asking a committed brain to grade its own commitment.

And the danger usually isn't a dramatic blowup. Most ideas don't die from a single catastrophe you'd notice; they fade, absorbing another month and then another while never quite failing hard enough to force a decision. That slow fade costs more than a fast, clean "no," because the currency you're burning is the one you can never get back: your own runway of attention and calendar. A quick death frees you to place the next bet; a lingering one taxes it.

Kill criteria fix this by moving the judgment earlier, to a moment when you're still cold. You decide now, while you can think clearly, what future evidence would change your mind, and you write it down so your later, more attached self can't renegotiate it. It's a version of what decision researchers call a Ulysses contract: you bind your future self while you're still rational, the way Odysseus had himself lashed to the mast before the sirens sang.

In Thinking in Bets, poker player and decision scientist Annie Duke names the failure this defends against: "resulting," the habit of judging a decision only by how it turned out. A founder who resulting-judges their own idea will keep a bad bet alive as long as any single number looks good, and kill a good bet the first bad week. Pre-set criteria force you to grade the decision on the evidence you named in advance, not on whichever outcome your mood is reacting to today.

Picture the mechanism in action: two founders run the identical weak launch, one with a pre-set demand threshold and one without. The first reads the miss against a line drawn in a calm moment and calls it, painful but clean. The second, judging live and attached, finds the one channel that looked promising, decides the test was "contaminated," and green-lights another month. Same evidence, opposite decisions, and the only difference was who had committed to a standard before the outcome arrived.

If you've started companies before, you already know the feeling of realizing eighteen months late what you suspected in month two. This is the discipline that closes that gap, and it's the backbone of any serious second-time founder validation playbook: decide the exits before the emotions arrive.

The four threshold types: demand, economics, channel, and personal

The four threshold types are demand, economics, channel, and personal, and nearly every durable kill criterion falls into one of them. Naming the type first keeps you from writing five criteria that all test the same thing while a fatal risk sits completely unmonitored.

Each type answers a different question your idea has to survive. Demand asks whether anyone actually wants the outcome. Economics asks whether the math can ever work. Channel asks whether you can reach buyers repeatably. Personal asks whether you'll still want to be doing this in two years. Skip any one and you've left a blind spot exactly where founders most often crash.

The table below maps each threshold to what it tests, what a breach tends to look like in the wild, and the specific rationalization it's designed to overrule.

The takeaway: write at least one criterion per type. A demand test tells you nothing about whether you can afford customers, and glowing economics on paper mean nothing if you have no repeatable way to find those customers. Coverage across all four is what turns a pile of metrics into an actual safety net.

One caution on balance. In practice, founders over-index on demand criteria, because demand is the fun part to test, and under-invest in the personal threshold, because auditing your own motivation feels self-indulgent. It isn't. A founder who has quietly stopped caring will find a way to fail the other three tests on purpose, so treat your own energy as a first-class signal rather than an afterthought.

There's also an order to them. Demand and channel usually deserve the earliest, cheapest tests, because an idea nobody wants or nobody can reach is dead no matter how elegant the economics look on a spreadsheet. Economics can often be estimated early and confirmed later, while personal thresholds run continuously in the background, checked at every review rather than resolved once. Sequencing your criteria this way means you spend scarce validation time disproving the fastest, most fatal risks first.

Writing testable kill conditions, step by step

A testable kill condition names one metric, one threshold, one deadline, and one action, so that when the date arrives there's nothing left to argue about. If any of those four is missing, you don't have a kill criterion. You have a hope with a due date.

The single most common mistake is writing conditions that can only be evaluated in hindsight, like "if we don't get traction." Traction is a mood, not a measurement. When the review comes, a motivated founder will always find some angle from which this month looked like traction. Vagueness isn't caution; it's a pre-installed escape hatch.

Build each condition in the same order:

1. Start from the riskiest assumption. Ask what single belief, if false, kills the whole idea, and test that first, not the assumption that's easiest or most fun to check.
2. Pick a leading indicator you can read soon. Choose something measurable in weeks, not quarters, so the signal arrives while a pivot is still cheap.
3. Set the threshold cold. Before you run the test, write down the specific result that would honestly change your mind. Decide it now, because after the data lands you'll be tempted to move the line.
4. Attach a real deadline. A criterion with no date never triggers; it just gets "a little more time" forever.
5. Predefine the action. State in advance whether a breach means kill, pivot, or persevere, and be honest that all three are on the table.
6. Write it somewhere you can't quietly ignore. A criterion living only in your head is one you'll revise without noticing.

Notice what this order protects against. By starting from the riskiest assumption, you avoid the comfortable habit of testing the things most likely to pass. By choosing a leading indicator, you get your answer while course correction is still cheap, instead of waiting for a lagging revenue number to confirm what you could have known months earlier. And by predefining the action, you remove the worst conversation in startups: the one where a team stares at a clearly failing result and debates, from scratch and under stress, what it's allowed to mean.

Choose the leading indicator carefully, because the wrong one quietly guarantees a false pass. Page views, sign-ups for a free thing, and polite "I'd totally use that" replies are the classic traps: they move easily and correlate with almost nothing that pays. Anchor your criteria to signals that cost the customer something real, a deposit, a scheduled call, actual usage, a signature, because willingness to spend a scarce currency is the only demand signal that survives contact with a budget.

The difference between a criterion that saves you and one that doesn't is almost always specificity. The comparison below shows the same intent written weakly and then written so it actually bites.

The pattern is clear: strength comes from deciding everything before the result exists. A helpful companion exercise is to run a pre-mortem on the startup idea first, imagining the venture has already failed and working backward to the causes. Those imagined causes are your best raw material for criteria, because they surface the failure modes you're currently most inclined to wave away. For the full sequence of tests that surround these conditions, the complete guide to startup idea validation walks through where kill criteria sit inside the broader validation loop.

Review cadence and who holds you accountable

Kill criteria only work if you schedule the review before you need it and give someone else the standing to hold the line. A criterion you evaluate alone, on a date you set in the moment, is a criterion you'll evaluate favorably. The cadence and the witness are not optional add-ons; they're what make the whole thing binding.

Match the review rhythm to how fast your signals move. Fast, cheap tests, like landing pages or outreach experiments, deserve a weekly look, so a dead end doesn't quietly eat a month. Slower bets, like a pilot with a design partner, are better reviewed at fixed milestones. Whatever the interval, put the review on the calendar the same day you write the criterion, not when you sense trouble, because the moment you "feel like it's time to review" is exactly the moment your judgment is already compromised.

Then borrow the accountability structure Duke recommends in Thinking in Bets: a small group of people who agree to tell each other the truth, including the truths nobody wants to hear. Pick at least one person who is not emotionally invested in your idea and give them a real job, to read your pre-committed criteria back to you and ask, plainly, whether they were met.

Their value is precisely that they weren't in the room when you fell in love. Choose someone who will actually push, another founder, an advisor, a skeptical friend, not a cheerleader who'll help you rationalize. And write your prediction down before each review, so afterward you can't rewrite what you "always expected." A clear record is the only defense against a memory that edits itself to protect the idea.

Keep the review itself brutally simple. Read the criterion exactly as written, state the actual result next to it, and name which of the three outcomes applies: kill, pivot, or persevere. Resist turning it into a strategy session, because a long discussion is usually the sound of a breach being talked out of existence. The evidence either cleared the bar you set or it didn't; the interpretation you're tempted to bolt on afterward is the exact thing the pre-commitment was meant to prevent.

Common ways founders rationalize past their own criteria

The most common way founders rationalize past their criteria is by quietly moving the threshold after they've seen the result. You set a bar, you miss it, and within a day you've discovered five reasons the bar was set wrong. The criterion is still technically there. It just doesn't do anything anymore.

Recognizing the maneuvers is half the defense, because they feel like insight in the moment rather than avoidance. Watch for these:

• Moving the goalposts. You miss the number, then decide the "real" target was lower all along.
• Redefining the metric. Sign-ups looked weak, so now the thing that matters is "engagement," which happens to look better this week.
• Blaming execution, never the idea. Every miss becomes a marketing problem, a copy problem, a timing problem, anything except a demand problem.
• The "just one more experiment" loop. You extend the deadline for a single decisive test, and then again, and then again.
• Reframing dread as grit. Your own falling motivation, a real personal-threshold breach, gets recoded as the noble suffering all founders endure.

The defense is structural, not heroic. Write the criteria in ink, review them with your witness, and separate the moment of judging the evidence from the moment of feeling the fear. Willpower is not a reliable counterweight to a brain that's protecting a two-year investment; a pre-committed process, evaluated with someone watching, is.

Experience makes this harder in one specific way. A second-time founder carries more sunk credibility, the reputation, the "I said this would work," the people who joined because of your last outcome, so the social cost of calling a stop feels heavier. That weight is real, but it's a reason to make the criteria more binding, not less. The founders who protect their track record longest are the ones who let a written rule, not their pride, decide when a bet is over.

It helps to reframe what honoring a kill criterion actually means. Killing an idea that hit its stop condition isn't quitting on your dream. It's freeing the months you'd have spent grinding on a proven dead end and redirecting them at a bet that can actually pay. The founders who win more over a career aren't the ones who never quit. They're the ones who quit the wrong things faster.

Key Takeaways

• Write kill criteria before conviction sets in. A committed founder can't grade their own idea honestly, so the only reliable moment to decide your exits is while you're still cold and skeptical.
• Cover all four threshold types. Demand, economics, channel, and personal each test a different way an idea dies; monitoring three of them just means you'll crash into the fourth.
• A real criterion has four parts. One metric, one pre-set threshold, one deadline, and one predefined action. Miss any one and you have a hope, not a kill switch.
• Set the threshold before you see the result. Deciding in advance what would change your mind is the whole mechanism; deciding afterward is just rationalization wearing a number.
• Schedule the review and recruit a witness. Put the review date on the calendar the day you write the criterion, and give an unattached person the standing to read it back to you.
• Watch for the goalpost shuffle. Moving the bar, redefining the metric, and blaming execution all feel like insight and function as avoidance; catch them by keeping the criteria in writing.
• Quitting a dead end is an offensive move. Honoring a kill criterion isn't failure; it's reclaiming the months a proven-wrong idea would have quietly consumed.

Frequently Asked Questions

What is an example of a kill criterion for a startup?

A concrete kill criterion pairs a specific leading indicator with a pre-set threshold, a date, and an action. For example: "If, after a fixed number of qualified sales conversations by a set review date, not one prospect commits to a paid pilot, we kill this direction." The key traits are that it's measurable, decided in advance, and dated, so it can actually trigger.

When should you set kill criteria for a new idea?

Set them at the very start, before you've built anything or told many people, while you're still able to be skeptical about your own idea. Kill criteria written after you're attached are almost always too lenient, because you've already accumulated reasons to protect the idea. The earlier and colder the decision, the more honest and binding the threshold will be.

How is a kill criterion different from a goal or KPI?

A goal states what you're aiming for; a kill criterion states the result that should make you stop. KPIs track ongoing performance, usually with an implicit assumption that the venture continues. A kill criterion is the opposite posture: it names, in advance, the specific evidence that would end the bet. Goals pull you forward; kill criteria tell you when forward is the wrong direction.

What if my kill criteria are triggered but I still believe in the idea?

That tension is the whole point, and the honest move is usually to respect the criterion you set while calm. Belief that survives a breached threshold is exactly the belief you can't trust, because it's the attachment talking. If you're genuinely certain the criterion itself was mis-designed, change it openly with your accountability partner and document why, rather than quietly ignoring it.

Do kill criteria make founders quit too early?

Not if they're written well, because a good kill criterion tests a specific fatal assumption, not a passing bad week. The risk of quitting too early comes from vague, mood-based judgments, which is exactly what pre-committed thresholds replace. By defining in advance what a real, disqualifying result looks like, you protect an idea from being killed by noise just as much as you protect yourself from grinding on one long past the evidence.