The skill that changed my business wasn't a tool. It was learning where I could trust it, and where not to.
Recap for new readers: six months into owning a forty-year-old company, I used ChatGPT to extract our entire order history from a legacy system whose vendor wanted ~$10,000 in a matter of weeks, just to let us keep accessing the records we’d already created. Last week ended on the catch: I'd gotten the data out as more than 600 raw CSV files, which my non-technical team had no idea how to actually use.
So I went back to ChatGPT and asked for the part I needed to make use of the files: can you build a user-friendly interface around this?
The results were… well, they didn’t exactly spark a hallelujah moment. I told it what I wanted most: the ability to look up a serial number and check whether a product was still under warranty. The first few passes couldn’t read the data at all. The next gave me results that were pretty clearly wrong. But the key was that I could fix the problems the way I fix everything else: in plain English. This part's wrong. Do it this way instead. The same sort of conversation I’d have with a software developer; the AI tool knew how to execute, but I needed to be the one shaping the output.
I was lucky. I'd bought from this company before I owned it, so I knew exactly what my order record should say. When the tool returned it correctly, I knew we were on the right track. After some more spot-checking, I knew we had something. It wasn’t perfect. But it was usable.
Two observations from that stretch worth noting for any small operator.
First, on adoption. My team is far from tech-resistant, but they were skeptical of what AI could actually do to move the needle at an old-school consumer goods business. I knew our size relative to our competitors meant we’d have to take every advantage that tech could give us, but at the same time the team was busy and didn’t have the time to burn on experimental tools that didn’t deliver results. What changed the conversation wasn't an argument about what AI could give us in theory. It was watching a serial number come back instantly instead of walking a customer through finding a sticker on their equipment. Buy-in requires a visible win on a real task. Everything else is theory, and theory doesn't move people.
Second, on the decision itself. With the data extracted and a working interface, I cancelled the legacy contract on December 31st. We saved about $10,000, but that wasn’t the biggest return. It was that we redefined what was within our reach as a small company. A job which had lived in our "can't afford to do it" column got done in-house, by someone who doesn't know SQL, coding, or database engineering. And we learned something that still reshapes how I make decisions: with these tools, a small business can build the bespoke solutions it once either had to buy or make do without. The build-versus-buy line moved.
That should be the end of the story. It isn't, quite.
Because that first tool was crude. It did one thing, it occasionally was wrong, and we had to learn to double-check it. For about eighteen months, I left it that way. But in that time, I learned to work with these tools on everything else.
That's the part that actually changed the business. Not the one tool. The skill.
We’ve reached the point where AI adoption is no longer optional for small businesses in competitive industries. The productivity impact is too great to ignore. But there’s risk in the other direction. The moment you take whatever an AI model spits out at face value is the moment it quietly hands you a confident, wrong answer and you run with it. The real skill isn't prompting. It's balancing risk versus returns, guardrails versus new possibilities. Don’t treat AI as an oracle. Treat it as the smartest intern you’ve ever had: trust, but verify.
Eighteen months in, the AI models had gotten dramatically better, and so had I. So I came back to that crude little interface and had Claude rebuild it properly. The new version isn't just a taped-together serial-number lookup tool. You can search by name, by phone number, by email. Full order history, clean contact records, a UI my team actually likes using. The thing I'd built in a panic to beat a deadline became, on the second pass, something that truly works.
So now, when a customer who spent real money with us calls in, we already know them. We say, I can see exactly what you bought, and when. How can we help? At our price points, that recognition demonstrates we value that customer.
Here's the thing I keep returning to, and the reason I'm writing this newsletter at all.
The fear about AI is all about the work it takes away. My experience has been the opposite. Every meaningful thing it's done for us is work that otherwise would not have gotten done. Too technical, too expensive, too far outside our ability to manage. It hasn’t replaced anyone. It’s let a handful of non-technical people do things we'd otherwise have simply skipped.
Until next week,
Zain
P.S.: What's something you've always assumed you'd have to buy, that you're now wondering if you could build? Reply and tell me.