Read ten pitch decks this month and I'd bet nine say "AI-powered" on the first slide. A few of those companies have built something hard to copy. Most have put a nice interface on a model anyone can rent.
You won't see the difference in the deck. You'll see it two years later in their margins. These are the eight things we check on every startup so you can see it before you invest.
1. Traction
Is there real demand, or a good demo? Look at active users, revenue growth and retention. AI products often get a spike of curious sign-ups that fades fast.
2. Business model
How does the company make money, and does the AI actually drive that revenue? Check pricing, unit economics and whether customers would pay without the AI features.
3. Real AI or wrapper
Does the company own anything a competitor couldn't copy in a month? Proprietary data, fine-tuned models, deep workflow integration and distribution all count. A prompt on top of a public model does not.
4. AI models used
Which models does the product depend on, and from which providers? What happens if one provider raises prices, changes terms or degrades quality? A healthy company has a fallback.
5. AI costs
What does it cost to serve one customer today, and at ten times the usage? Inference costs can quietly eat gross margin as a product grows.
6. Data rights
Is the company allowed to use the data its AI is built on? Scraped, licensed and customer data each carry different risks.
7. Security
Can the AI be tricked into leaking data or taking actions it shouldn't? Check agent permissions, prompt-injection defences and how customer data is separated.
8. Regulation
Is the product exposed to rules like the EU AI Act or US state AI laws? Hiring, credit and education are high-risk categories with real obligations.
How long it takes
Done manually, this review takes weeks. AI Due Diligence runs it on any startup in about 10 minutes and gives you a red, amber or green score for each point, so you know where to dig deeper.