The Idea Triage Framework: Score, Kill, or Fund in One Hour

Deciding which startup idea to pursue comes down to scoring every candidate on four criteria before anything gets built: demand evidence, founder advantage, willingness to pay, and cost to test. Ideas that clear your bar earn a small, capped test budget. Everything else gets killed in writing that same week — no prototype, no domain purchase, no "just a quick landing page to see."

That one habit protects more runway than any amount of clever engineering, because most wasted build budgets aren't lost during validation. They're lost upstream, at the moment a founder decides an idea is worth validating at all.

Ideation got cheap. That moved the bottleneck.

AI can hand you a dozen coherent startup ideas before lunch, each with a confident pitch, a plausible market, and a feature list. The dangerous part is not that these ideas are bad. It's that they're plausible. A language model is trained to produce things that sound viable, and "sounds viable" is exactly the trap that gets a solo founder three months and five figures into a build nobody asked for.

Evaluating AI-generated business ideas is also a different job than evaluating your own. You have less emotional attachment, which helps. But you also have less context — the model doesn't know your network, your savings, your distribution channels, or what you're uniquely positioned to see. So the scarce founder skill has shifted. It's no longer ideation. It's ruthless, repeatable triage: a startup idea prioritization framework you run the same way every time, so the decision doesn't depend on your mood that morning.

Here's the one we use.

The four criteria that predict whether an idea deserves money

Score each idea 0–2 on four axes. Maximum score is 8. The anchors matter more than the numbers — vague scoring is how bad ideas sneak through.

1. Demand evidence (0–2)

What proof exists that anyone has this problem badly enough to act on it?

0 — The only entities who believe in this problem are you and the AI that suggested it.
1 — Secondhand signals: people complaining in forums or communities, competitors existing, visible search interest.
2 — You have personally heard a specific human describe this problem unprompted, or you can point to people already paying for a worse workaround.

That last signal — money already flowing to a clumsy alternative, a spreadsheet, a virtual assistant, a duct-taped process — is the strongest evidence you'll get before launch. Existing spend beats stated interest every time.

2. Founder advantage (0–2)

Here's the uncomfortable math of AI ideation: the model that generated your idea will generate roughly the same idea for thousands of other people prompting it this month. The idea itself carries almost no edge. Your advantage is the only unique input.

0 — Anyone could build this, and dozens probably will. You bring nothing structural.
1 — You have relevant domain experience or technical skill that shortens your path.
2 — You have unfair access: an existing audience, proprietary data, a distribution channel, deep regulatory or industry knowledge that takes others years to acquire.

We see founders skip this criterion most often, and it's the one that separates "good idea" from "good idea for you."

3. Willingness to pay (0–2)

Not "would people use this." Usage is free. The question is: who writes the check, out of what budget, replacing which existing line item?

0 — "Consumers might pay a few dollars a month." No named buyer, no named budget.
1 — A plausible buyer type with a plausible budget, but you can't name real instances.
2 — You can name three actual people or companies, and you know which budget the money comes out of — tooling, headcount, marketing spend, whatever it displaces.

If you can't finish the sentence "this replaces the money they currently spend on ___," score it honestly.

4. Cost to test (0–2)

This one is inverted: cheap-to-kill ideas score high. Ask what the least expensive experiment is that could disprove demand.

2 — Killable in under a week for a few hundred dollars: ten customer interviews, a concierge version done manually, a waitlist that asks for a deposit.
1 — Testable in a few weeks for low four figures.
0 — You genuinely cannot learn anything meaningful without months of building first.

An expensive-to-test idea isn't a bad idea. It's a bad idea for now, for a founder without the capital to burn on the answer. Backlog it and move on.

The kill line: what a passing score actually buys

Three rules, applied without negotiation:

Any single 0 kills or backlogs the idea, regardless of total. A 7/8 with zero founder advantage is someone else's business.
6 or above earns a capped test budget — a fixed dollar amount and a fixed calendar window, sized to your runway. For most early-stage founders we work with, that's in the hundreds of dollars and one to two weeks, not thousands and a quarter.
4–5 goes to a backlog, with one written line: what new evidence would change the score.

Note what the passing score buys: the cheapest possible demand test. Not a prototype. Not an MVP. The build decision is a separate, later gate that the demand test has to unlock.

And write the scores down. The written record is what stops you from re-litigating the same seductive idea every month when the AI serves you a fresh variation of it.

Decide the exit before you enter

Startup idea validation criteria only work when they're paired with kill criteria for startup ideas — written before the test starts, while you're still objective. The format is a falsifiable sentence with a number and a date:

"If fewer than 5 of 20 interviewees describe this problem unprompted by the 25th, this dies."
"If the pre-order page converts under 2% of qualified visitors after 200 visits, this dies."
"If none of the ten manual concierge clients asks to continue past the free period, this dies."

The pre-commitment is the entire point. Two weeks into a test, you will be tempted to move the goalposts — every founder is, us included. A number you wrote down on day zero is much harder to argue with than a feeling you have on day fourteen.

The mistakes we keep seeing

Scoring in your head. Unwritten scores drift toward whatever you already wanted to do.
Letting your own excitement stand in for demand evidence. Enthusiasm is fuel, not data.
Running three tests at once. Attention is the real constraint; parallel tests usually means three half-tests.
Building the prototype as "the test." A prototype tests whether you can build the thing. You can. That was never the question.
Treating the AI's market analysis as evidence. Model-generated market sizes and "growing demand" claims are hypotheses to check, not facts to cite. Verify every number before it influences a score.

Make it a standing habit

How to choose which startup idea to pursue stops being an agonizing decision when it becomes a weekly hour: score whatever new ideas accumulated, check active tests against their kill criteria, then kill, backlog, or promote. Most weeks the honest output is a stack of kills and zero new builds — and that's the framework working, not failing. If you want a second set of eyes on the scoring, or CTO-level direction on what the cheapest credible test looks like for a specific idea, that's exactly the decision layer we help founders get right before a single line of code gets written.