Why I build my own Ai tools
A while ago I needed a way to show customers what our tools actually did for their business.
The obvious answer was a spreadsheet. I've seen how those go. It sits in a shared folder, the third person to open it breaks a formula, and nobody touches it again.
So I built a tool instead. A Replit app with five inputs, kept that small on purpose. You put your numbers in and it shows hours saved per week, NPS movement, extra revenue and churn reduction.
It took a couple of hours to build and deploy. Sales were using it the same day.
Our sales conversations changed. We stopped talking features and benefits. We showed customers their own numbers, projected forward. (Funny how much more interested people get when it's their own numbers.)
That build led to the next one. Then the next. Proofs of concept, internal tools, content ideas. Most got binned quickly. A few I now use every day.
What it's changed about how I work
Most product people I know sit one step away from what they're building. They write specs, review designs and sit in demos. Someone else makes the thing. I've worked that way too, and it works fine.
Claude, Cursor, Replit and v0 mean I can get from an idea to something working in an afternoon. So now I build first. Three things came out of that.
I feel the friction myself. Five inputs felt minimal to me. For a sales rep mid-call, even five needed explaining. I'd never have spotted that in a requirements doc. I spotted it watching someone use it for the first time.
I know how Ai breaks. When an output is wrong, I can usually see why. Most of the time the input data was too thin. So when I decide where Ai fits, I go off what I've seen break.
I get real reactions. Show someone a mock-up and you get polite feedback. Run their own numbers through a working tool and you get objections. The objections are the bit I need to hear.
None of it is polished
Some of what I've built are internal scripts. Some are one-off React apps in a folder on my laptop. One is a call transcript analyser that flags competitor mentions and compliance keywords. I built it over a weekend and I've used it in every quarterly review since.
None of them needed an engineering sprint. A few are public now, and you can try them on the tools page.