1 April 2026
Your data is about to matter a lot more
Stuart Wan
I spent most of this week untangling data.
Not the fun kind. The kind where three people track the same thing differently. Where critical business logic lives in someone's head and a spreadsheet from 2019. Where a number is off by a little bit, and nobody can tell you why.
This is what working inside real businesses looks like. And I'm starting to notice something that makes this unglamorous work matter more than most people realise.
Something interesting is happening at the top end of tech.
Companies like Cursor, Intercom, and Pinterest are taking open-source AI models and specialising them on their own data. They're training smaller models to be really good at the specific tasks their customers care about. Then they pair those with powerful general models for the heavy reasoning.
Why? Two reasons. Cost, because sending everything through a frontier model at their scale gets expensive fast. And privacy, because your core business data, your pricing, your margins, your customer patterns, is not something you want sitting on someone else's servers.
So they keep the private knowledge in models they control. And use the general intelligence when they need the horsepower.
That's their moat. Not the AI itself. The specialised knowledge they feed into it.
Now, you might be thinking: that's great for Pinterest. What does this have to do with my 50-person company?
Fair. You're probably not going to self-host an open-source model. You don't need to.
But the raw material is the same.
Every business I work with has accumulated years of operational knowledge. Pricing logic. Supplier relationships. The weird exceptions that exist for good reasons nobody wrote down. Customer service that runs on tribal knowledge and good memory.
Most founders have never thought of this as a technical asset. It's just "how we do things."
But that knowledge is what makes AI actually useful for your business. Without it, ChatGPT gives you generic answers. With it, you get something that actually understands your operation. The form factor might be different from what Pinterest is doing. Maybe it's a well-structured knowledge base. Maybe it's clean data in your systems instead of scattered spreadsheets. Maybe it's business rules captured in software instead of one person's head.
The point is the same. Your private knowledge is the foundation. AI is just the amplifier. And you don't need a custom model to benefit from that. You just need your knowledge in a form that any tool can use.
This is also why I think the typical vendor model is breaking down.
A vendor asks you what you want. You tell them. They build it. The problem is most business owners can't articulate what they need in technology terms. They know their pain. They don't know the solution.
So the vendor builds exactly what was asked for. And it kind of works. But not really.
General AI has a similar blind spot. ChatGPT can write your code. It can't sit in your office for three months and notice that your team tracks the same customer differently across four systems. It can't figure out why your quoting process breaks when a supplier updates their pricing.
The hard work is getting knowledge out of people's heads and messy processes and into something a system can use. That requires someone who understands both the technology and the business well enough to say "this is what we should actually build" before anyone asks for it.
I see many business owners running companies doing $5-50M a year. Decades of industry knowledge. Customers who depend on them. But the business struggles to scale because that knowledge doesn't transfer. Every new hire takes forever to get productive. Every process lives in someone's head instead of a system.
A lot of them are now looking at AI thinking it will speed things up. And it can. But not if you skip the first step. You can't amplify knowledge that isn't captured. AI on top of messy, undocumented processes just gives you faster mess.
The first step is boring. Define your business rules. Clean up your data. Get the processes out of people's heads and into structured systems. Then AI becomes genuinely useful, because now it has something real to work with.
Start there. Start messy. Start imperfect. Just start.
Weekly notes from conversations with my business partner, where we share what we're seeing on the ground across clients and the market - not the hype, the real shifts.