A good demo tells you people might want the product. It tells you almost nothing about whether the company gets to keep selling it, or keeps any money when it does. Those are two different questions, and the second is where money is made or lost. This is how I think about it, not investment advice.
If you came from the video, that's the idea. Here's the part that doesn't fit in sixty seconds: the exact questions I ask, the reasoning under each, a full run on a real-feeling company, and the copyable list I take into a call so I don't get charmed into skipping the hard parts.
The questions I insist on
I assume the product is good, or there'd be nothing to talk about. So I spend my attention on the five things that decide whether a liked product is also a business worth owning a piece of.
Whose model is it built on?
Most AI startups run on someone else's model (OpenAI, Anthropic, Google). Normal and often smart, but it puts two levers you care about, cost and capability, in the hands of a supplier who is also a potential competitor. I want the founder to describe that exposure plainly: what breaks if the price doubles, and what they'd do if the provider shipped their exact feature as a checkbox. A founder who has thought about it has an answer. One who hasn't gets defensive.
Does it make money on each customer?
What does it cost to serve one user for a month versus what they pay? AI has a real per-use cost, so this is not a rounding error. If the company loses money on every customer, growth makes the hole deeper. I want gross margin after model and infrastructure costs, on paying customers, not the free tier. The margins guide covers what to subtract.
Why would a customer stay?
What stops a user leaving for the next tool, or for the model provider doing it natively? Something has to hold them: their data living in the product, a workflow the team runs on, integrations that took a month to set up, a habit. If the honest answer is "our model is a bit better," that's rented, not owned. Better models arrive every few months.
Is the growth real or bought?
Free credits, generous trials, and launch buzz inflate the numbers that look best in a deck. I want to see what happens after the discount ends: do people convert to paid, and are they still there in month six? One clean cohort chart tells me more than any total-signups figure.
Is there a legal or platform question hanging over it?
Training data, copyright, privacy, and the terms of whatever platform they build on. A lawsuit or a policy change can outweigh great product-market fit. I don't need certainty, I need to know the founder has named the risk and has a plan that isn't "hope."
A real one, run end to end
Abstract questions are easy to nod at and hard to apply, so here's one worked through. A startup is raising money for an AI note-taking app: it joins your meetings, transcribes them, and writes clean summaries with action items. The demo is genuinely good, and I still have to answer all five before I get excited.
- Whose model: it runs on one provider's speech-to-text and another's summarization. So I ask the cost per hour of audio today, what it does if that provider raises prices, and what's left when the provider adds "summarize this meeting" natively, which they're clearly heading toward.
- Per-customer economics: a heavy user sits in twenty hours of meetings a week. If the plan is twenty dollars a month and that user costs eighteen to serve, the average looks fine only because light users subsidize the heavy ones. That's a pricing bomb waiting for the wrong customer mix.
- Why they stay: the summary is a commodity. The lock-in, if any, is the searchable archive of every past meeting and the integrations that push action items into the tools a team already uses. I'd ask what percentage of paying teams connected an integration, because that's the moat, not transcription quality.
- Real or bought growth: these apps grow fast because one person adds the bot and the whole call sees it. I want paid conversion and month-six retention, not meetings joined. A bot in the room is not a customer.
- Legal and platform: recording people has consent rules that vary by place, and the video platforms they plug into can restrict bots whenever they like. I want their consent handling and their plan if a platform tightens the rules.
The demo answered none of these. Good products fail these tests all the time, which is why the demo can't be the thing you invest on.
The list I take into the call
This is what I paste into my notes before a founder call, so I ask the same things every time instead of getting talked past the awkward ones. If a founder can't answer three or more of these, that's the answer.
Company: [ ]
MODEL DEPENDENCE
- Which model(s) do you build on, and what % of your cost is that?
- What breaks if their price doubles?
- If they ship your core feature natively, what do we still have?
PER-CUSTOMER ECONOMICS
- Gross margin on a PAYING customer, after model + infra?
- Cost to serve your heaviest 10% vs. what they pay?
- Does the average margin hide a segment that loses money?
WHY THEY STAY
- What holds a customer besides "our model is better"?
- % of paying customers with data / integrations locked in?
- Net revenue retention over the last 12 months?
GROWTH QUALITY
- Free-to-paid conversion after credits/trials end?
- Month-6 retention by cohort, not total signups?
LEGAL / PLATFORM
- Any open question on training data, copyright, or consent?
- What happens if a platform you depend on changes its rules?What goes wrong when you skip this
The most common mistake I see, and one I've made, is investing in the demo and calling it diligence. The product is delightful, the founder is sharp, and somehow nobody asks what it costs to serve a customer. Delight is a real signal about demand. It says nothing about margin, retention, or defensibility, and those are what you're buying. Separate "do I like this" from "is this a good business," and ask the second one out loud.
The second trap is total-user numbers. A company leads with millions of users because that figure flatters, and it usually means the paying and retention numbers are worse. When someone quotes reach instead of revenue, treat the silence as the data: ask for paying customers and month-six retention, and watch how quickly the answer gets vaguer.
Third, people wave away model dependence because everyone depends on the big providers right now. True, but "everyone" is not a plan. The question isn't whether they depend on a provider, it's what they built on top that the provider can't trivially copy, and whether their cost survives a price change they don't control. If the whole answer is a thin wrapper, the provider is your real competitor.
Last, don't confuse a big valuation with a proven business. A number that's mostly "reportedly" plus a story about a huge market is a story, not evidence. Before you get attached, run the whole thing through the research a business framework, and if you want an assistant to gather the filings and coverage for you, the research assistant setup does that legwork.
Everything here comes down to one line: does the company own something, or is it renting a moment? A rented moment is a model that's better this quarter and a wave of curious signups. Something owned is customer data, workflow lock-in, and margin that survives a price change it doesn't control. You're buying the owned part. If you can't find it, you're paying for the moment.
And once you've decided the business is worth backing, know exactly what your money buys, because a check doesn't always mean what people assume. That's its own guide: what you actually own when you invest in a startup.