When to Give Up on a Business Idea Using Kill Criteria

Give up on a business idea when it misses a number you wrote down before you started, on the date you wrote down, and not a day later. If you never set those numbers, set them now for the stage you're in. Once you've put months into an idea, your judgment leans toward continuing, so it's the wrong tool for this decision.

Why can't you trust your judgment once you've started?

The research on this goes back almost fifty years. In Barry Staw's 1976 study, 240 business school students ran a simulated investment decision. The people who put the most money back into a losing course of action were the ones personally responsible for the bad results. That describes most founders. You picked the idea, so its failure is your failure, and your brain works hard to avoid admitting it.

Annie Duke, the former poker pro who wrote a book on quitting, explains why this happens. As long as there is hope, a turnaround always seems possible, and moving from "failing" to "having failed" hurts so much that people keep going until they're sure they had no choice. That certainty only arrives very late.

So the most common way of deciding, "I'll know when it's not working," is also the weakest. When you finally know, you have paid for the lesson several times over.

What does a good kill criterion look like?

Duke's format is "states and dates". A state is a measurable condition you either hit or miss. A date is when you check. Her version with a budget works even better for founders: if you haven't reached X by the time you've spent Y in money, effort or time, you quit. She also argues that making the commitment in advance makes you more likely to follow through, even when you're behind.

A usable criterion has four parts: a metric, a threshold, a deadline and a spending cap. Compare these two:

• Weak: "If people don't seem interested by summer, I'll rethink it."
• Usable: "If fewer than 5 of the people I pitch pay a $50 deposit by March 31, or I've spent $500, whichever comes first, I stop."

The weak version has no number, no date anyone can check, and an action ("rethink") that lets you do nothing. The usable version can be checked by someone who has never met you, and that's the test.

What thresholds should you start with?

These are my defaults. They're opinions, not research findings, and they're meant to be edited before you start, never during. Each stage costs more than the one before it, so the aim is to kill the idea at the cheapest stage that can tell you no.

Stage	State you must reach	Deadline	Spend cap
1. Problem	15 conversations with the target segment; at least 5 describe the problem unprompted and already spend money or time on a workaround	3 weeks	$100
2. Demand	At least 300 targeted visitors to a landing page; at least 5% leave an email or click a pricing button	3 weeks after stage 1	$300
3. Commitment	At least 5 people pay a deposit or preorder	4 weeks after stage 2	$0 new spend
4. MVP	At least 10 paying customers, and at least 60% of them renew once	90 days after launch	$2,000 and 200 hours
5. Business	Revenue up at least 10% month over month for 3 straight months	6 months after stage 4	Set before stage 5 starts

Three rules make the table work:

A near miss is a miss. Four preorders when you needed five means no. If "close enough" counts, the threshold turns into a suggestion, and escalation creeps back in through that gap.

Passing a stage earns you the next stage's budget, nothing more. Hitting 5% on the landing page doesn't mean "keep going." It means "you may now spend up to the stage 3 cap."

Adjust for price before the clock starts. A $400 a month B2B tool might reasonably need only 3 preorders at stage 3. A $4 consumer app needs far more than 10 customers at stage 4. Change the numbers on day zero, write down why, and then leave them alone.

Why does a slow no cost more than a fast one?

A fast no and a slow no usually tell you the same thing: the demand wasn't there. Nine months doesn't teach you much more than six weeks. It teaches you the same lesson for more money.

Here's a hypothetical. Say you build on the side for 12 hours a week, and you value your time at $50 an hour, roughly what you'd earn freelancing instead.

	Fast no (fails stage 2, week 6)	Slow no (drifts to month 9)
Hours	6 × 12 = 72	39 × 12 = 468
Time cost at $50/hr	$3,600	$23,400
Cash (ads, tools)	$400	$1,500 ads + $30 × 9 months tools = $1,770
Total	$4,000	$25,170

The slow no costs about 6.3 times as much. That still isn't the biggest cost. The biggest cost is the number of ideas you get to try.

Say you have two years of side-project time, about 104 weeks. If each failed idea dies in 6 weeks, you get 104 ÷ 6 ≈ 17 attempts. If each one drags on for 39 weeks, you get 104 ÷ 39 ≈ 2. Now assume, purely for illustration, that 1 in 10 ideas you take on would pass every stage:

• 17 attempts: 1 − 0.9¹⁷ = 1 − 0.167 ≈ 83% chance at least one works
• 2 attempts: 1 − 0.9² = 1 − 0.81 = 19% chance at least one works

Your hit rate is the same in both cases. The only thing that changed is how fast you quit, and the odds of ending up with a working business go from about one in five to better than four in five. Quitting slowly doesn't make you persistent. It makes you less likely to ever find the idea that works.

How do you write criteria you'll actually obey?

Follow this procedure on day zero, before you write code or buy a domain:

1. Run a pre-mortem. Gary Klein's method assumes the project has already failed and asks why. He cites research showing that imagining an outcome has already happened increases the ability to correctly identify reasons for future outcomes by 30%. It can take as little as 20 to 30 minutes. Picture your stage 4 deadline arriving with the idea dead, and list every reason.
2. Turn each reason into a state. "Nobody would pay" becomes "fewer than 5 deposits." "Too hard to reach buyers" becomes "cost per landing page signup above $6."
3. Attach a date and a spending cap to each state, using the table above as the starting point.
4. Write the action for a miss as a single verb: stop. Not "reassess," not "consider a pivot."
5. Give the list to someone outside the project, and ask them to check your numbers on each date. Outside evaluators applying fixed criteria do surprisingly well. Analysts at the Canadian Innovation Centre forecast the commercial success of inventions with approximately 80% accuracy. That program evaluated 11,000 inventions, and its economists still had to account for inventors not following the advice given to them. A good evaluation only helps if you act on it.
6. Put every deadline in your calendar now, with the threshold written in the event title.

Isn't quitting early how you miss the idea that would have worked?

This is the strongest objection, and it rests on survivorship bias. You hear from the founders who pushed through and won. The ones who pushed through and lost don't give talks.

Kill criteria don't remove persistence. They point it at something. Hitting the stage thresholds is where persistence belongs, and grinding through 30 cold pitches to get 5 deposits counts. What the criteria stop is persistence without evidence.

Organizations that run experiments for a living build this in on purpose. At X, Alphabet's moonshot lab, Astro Teller's advice is to run at all the hardest parts of the problem first. He also says they have bonused every single person on teams that end their projects. Solo, nobody pays you a bonus for quitting. The equivalent is treating a clean kill as a finished task, not a personal failure.

That leaves pivots. My rule: a pivot is a new idea, not an extension. If stage 2 shows that accountants want your tool when you built it for lawyers, you haven't earned more time on the lawyer idea. You've found a new candidate. Give it its own criteria, its own caps and its own clock, starting from stage 1. Without this rule, "pivoting" becomes a way to keep an idea alive indefinitely without ever facing a deadline.

What do you do the day a criterion trips?

Stop within 48 hours, before you can talk yourself out of it. Then do three things.

First, write a five-line autopsy: which state you missed, by how much, what you assumed going in, what turned out to be true, and what you would test first next time. It shouldn't take more than 15 minutes. The value is in writing it while the details are fresh.

Second, keep what's reusable. An email list, a landing page template, ad targeting you've learned about, a codebase with working auth and billing: all of these carry over to the next idea and shorten its stage 1.

Third, pick the next candidate and write its kill criteria before you do anything else. If your list of ideas is empty, Unthinkable's feed of business ideas built from demand signals is a reasonable place to find the next one. Whatever you pick, the day-zero rule still applies: thresholds, dates and caps go on paper before you open your editor.