Where is the pendulum of the AI world swinging?

We started at "all the value is in the big models." We moved to "the real value is with whoever generates revenue." And now we have landed, and it seems this is where we stay: "whoever holds interesting, unique data." (So say Sequoia, OpenAI, SSI, and everyone else.)

So instead of reciting what does not work, we focus on what does. One of the strongest strategies we have seen in our portfolio is the data moat.

The surprising part? A company's data strategy can, and should, be articulated on the day not a single line of code has been written.

We had a remarkable experience with Navina's founders. Already in the first meeting they told me: "We know how to build AI systems. But we will not write a single line of code this year before we get our data strategy precise." That is not just a nice sentence. It is a strategic choice that became a real competitive advantage.

Good data strategies are: unique and hard to replicate; improving with every new user; embedded deep inside the customer's workflow, so the moat keeps deepening.

But most founders do not start the journey with a backpack full of data only they have. So how do you create proprietary data? In our playbook we identified four common paths that actually worked:

1. Restructure public information in a smart, unique way. 2. A barter with your first customers. They give data, you return value. 3. Manual collection. Slow now, scalable tomorrow. 4. Using AI and the users themselves to improve the existing data and strengthen the moat.

Companies like OnFire, Protai, Limitless Labs, Alice, Nucleai and Opmed.ai built this moat for real. We wrote about it in the third part of our playbook.

And do not forget: unique data is not enough. You also need to know what to do with it.