The one assumption that can kill your startup — and how to test it this week

Startups don't usually die from bad execution. They die because a core assumption everyone treated as fact turned out to be false — and nobody checked until it was expensive. The most valuable skill an early founder can build is finding that assumption early and testing it for almost nothing.

Every idea is a stack of bets

Write your idea down and you'll notice it's really a pile of assumptions stacked on each other: people have this problem, they'll trust AI to solve it, they'll switch from what they use now, you can reach them affordably, the model is reliable enough. Each one is a bet. Some are safe. One is usually load-bearing — pull it out and the whole thing collapses.

The goal isn't to test everything. It's to find the load-bearing bet you're least sure about, and aim there first.

Sort by "fatal if wrong" and "least certain"

Run each assumption through two questions: If this were false, is the product dead? and How sure am I, really? Plot them. The assumption that's both fatal and shaky is your riskiest — that's where the first experiment goes. Everything comfortable or non-fatal can wait.

It helps to bucket them by type: desirability (do they want it?), viability (can it make money / can you reach them?), and feasibility (can it be built reliably?). For most new AI products the riskiest bet is desirability — whether the pain is real and sharp enough that people will change behaviour. Founders love to jump to feasibility because it's the fun part. Resist that.

Design the cheapest possible test

A good experiment has three properties: it's cheap, it's fast, and it produces a clear signal. The best ones need no product at all.

• Assumption: people will trust AI for this. Test: do it manually behind the scenes ("concierge") for ten users and see if they accept the output. If they won't take it from a human, an AI won't fix that.
• Assumption: they'll switch from the incumbent. Test: interview fifteen current users about what would actually make them move. Watch for real friction, not politeness.
• Assumption: there's demand. Test: a landing page that describes the outcome and asks for an email or a small pre-payment. Commitment is signal; a "cool idea" is not.

Notice none of these require building the product. The point of a test is to buy certainty at the lowest possible price.

Decide the kill criteria first

This is the step founders skip, and it's the one that makes the whole thing work. Before you run the test, write down the result that would make you stop or pivot. "If fewer than 5 of 15 describe this problem unprompted, I rethink the audience." Deciding in advance protects you from the universal temptation to reinterpret a weak result as encouraging. Without a line drawn beforehand, every outcome looks like a reason to keep going.

Run it, then believe the answer

The hard part isn't running the experiment — it's believing it. A lukewarm result is information, not an insult. Founders who treat a failed test as "we just need to explain it better" are the ones who spend a year building something the market already told them it didn't want. The ones who move fast take the answer, adjust the assumption, and test the next one.

Testing your riskiest assumption isn't a phase you finish before building. It's a habit you keep — the cheapest insurance a founder can buy against the most expensive mistake there is.

Want your assumptions surfaced and ranked, each with a concrete test? The free Riskiest-Assumption Finder names the one most likely to sink your idea and the cheapest way to check it. To pressure-test the whole idea with us, that's AI Product Strategy.