How I choose a company: 8 rounds of questions

Having ideas is easy. The hard part is comparing a collections agency, an accounting firm and a medical membership model with the same yardstick. This is the yardstick I use.

It isn't one matrix: it's eight rounds, and each one asks something the previous one didn't. That's the whole point. If two rounds in a row ask the same thing, the second one filters nothing — it just makes you more confident. That's how I went from 109 ideas to 7: 109 → 41 → 19 → 16 → 7.

The Excel template below has one tab per round, with the formulas already set up.

Stage 0: fill the list

Volume, not judgment. Two questions to generate:

• What behavior change will feel obvious in hindsight?
• Is AI exactly what makes it possible today?

Tag every idea by layer (L1 runs the business, L2 infrastructure, L3 platform) and horizon (now, emerging, moonshot).

Round 1: the six criteria (1 to 5, out of 30)

One by one, yes / no / maybe. Each criterion is scored 1 to 5, total out of 30:

1. Growth
2. Revenue
3. Durability
4. Proven problem
5. Market size
6. AI advantage

The market doesn't kill anything here. What dies is whatever doesn't excite you and whatever anyone could do as well as you.

Round 2: group into families

Put the survivors into families and score the family, not the idea:

• How high is the ceiling?
• Is it feasible for you, with what you have today?
• What's the honest moat? ("AI" is not a moat)
• Does it survive the risk that kills it?
• How fast do you learn inside it?
• Is the window open or already closed?

Round 3: demand test

Stop grading the idea and grade the demand:

• Is that money already being spent, or does the budget have to be invented?
• How slow is the incumbent?
• Can you name your first 10 customers, by name?
• Can it be falsified in 60 days for under $50K?
• If the idea dies, what's left?
• What capability does it leave you with?

Plus three "can I reach them" lenses: the gap with the best competitor, your unfair access, and whether that competitor's scale is worth anything in your city.

Round 4: pain × frequency × price

Three numbers, multiplied, not added. A zero in any of them kills the business:

• How much does the problem hurt?
• How often is it felt?
• What does it cost today not to solve it?

Everything preventive, and anything felt twice a year, falls here.

Round 5: the real numbers

• Transactions per year.
• Dollars per transaction.
• How much of that you can capture.
• Market size.
• How many customers you need for $10M and for $100M.

That last one organizes everything: some businesses need tens of thousands of customers bought through ads, others need forty.

Round 6: verify outside

Deep research on the ones still alive: market, competition, capital. Not to have a better opinion, but to see whether reality backs yours.

Round 7: the four underlying frameworks

• The oracle test: does something external grade the result with no opinions? (the payment lands, the filing is accepted, the code compiles). Without an oracle, AI never learns from its own work and the business doesn't compound.
• The Warp test: an agent on top of someone else's system is a wrapper; being the system of record is a company.
• AI-native services: be the firm, don't sell software to the firm.
• The Cognition/Windsurf lesson: buying distribution wins. If someone already has it, a better product isn't enough.

Round 8: one deep dive per idea, with a gate

The last one isn't a matrix: it's a yes-or-no gate, idea by idea. You define up front what has to be true to keep going, and if it isn't, it's out.

What kills the most

Out of 109, almost all of them died for four reasons:

1. No oracle.
2. Distribution already owned by a platform or a lab.
3. People buy too rarely.
4. Too much capital or too many people for thin margins.