What we learned building our own AI products

We build AI products for ourselves and websites and apps for clients, which means we get to make every mistake twice. Here are the lessons that cost us the most to learn, shared in the hope they cost you nothing.

1. The model is not the product

Every AI tool has access to roughly the same models. The product is everything around the model: the defaults, the constraints, the workflow, what happens when it goes wrong. QuickAIConvert works not because it does conversions, any model can, but because you never have to leave your sentence to get the answer. Friction, not capability, is what you're designing.

2. Ship the narrow version

Wordflow AI started as a plan for an everything-writing-platform. What actually launched did a few writing jobs well. The narrow version taught us which features people used (drafting, rewriting, tone control) and which they'd never touched (most of the rest). You cannot learn this from a roadmap. You can only learn it from a launch.

3. Free tiers are a product decision, not a pricing decision

The free tier isn't a discount, it's your demo, your word of mouth and your support load rolled into one. Set it too small and nobody experiences the value. Set it too big and you've built a charity. Our rule: free users should hit the limit at the exact moment they'd miss the product if it vanished.

4. The first support question predicts the next feature

When three people independently ask the same question, that's not a documentation problem, it's a product telling you what to build next. Our best features started life as support tickets. Keep a simple tally of questions; it beats most analytics.

5. Speed is a feature users can feel

Nobody praises fast tools in reviews, they just keep using them, and they abandon slow ones without saying why. Time-to-first-answer turned out to correlate with retention better than any other metric we tracked. When in doubt, make it faster.

6. Trust is earned in the error messages

Any tool looks good when it works. Users decide whether to trust you in the moments it doesn't: what happens when the model is down, when the input is weird, when the upload fails. Clear, honest, human error states buy more loyalty than any feature launch.

7. Small teams win on focus, not effort

There are four products in our catalogue, built by a small team in Coventry. That's only possible because we say no constantly, to features, to platforms, to whole product ideas. Big companies out-resource you; they can't out-focus you. The constraint is the strategy.

None of this is complicated, which is exactly why it's easy to skip. Build narrow, ship early, watch the support inbox, make it fast, and protect your focus like it's the budget, because it is.