Every single week, a client asks us to "add AI."
We get it. AI is everywhere, and the pressure to be seen doing something with it is real. But after building real systems for real businesses, we've learned something worth saying out loud: most businesses don't have an AI problem. They have a data problem.
Here's the honest framework we use — the four questions we ask before we ever recommend AI.
1. What problem are we actually solving?
Nine times out of ten, the answer is "we want AI" — not "we have a problem AI solves." Start with the problem. A chatbot sitting on a website where nobody asks questions isn't innovation, it's decoration.
2. Can a simple rule or query do it?
This is the uncomfortable one. A lot of "AI" requests are really just: organize this data, filter this list, match these records, or summarize this report. A well-built query, a simple algorithm, or a dashboard does that — for a fraction of the cost, with predictable results. AI is a hammer. Not everything is a nail.
3. Is the data good enough?
AI learns from your data. If your records are scattered across Excel files, WhatsApp messages, and paper, no model in the world will save you. The boring prerequisite is clean, structured data. We've turned down AI projects because the data wasn't ready — and it was the right call.
4. Does the output need to be explainable?
If a wrong answer costs money — pricing, compliance, approvals — you need systems that explain themselves. Generative AI produces confident answers, not guaranteed ones. For anything that touches money or rules, you want deterministic logic with AI layered on top, not AI deciding.
So when should you use AI?
Real AI problems are things like: extracting structure from documents and images, routing messages, predicting stock or demand, detecting anomalies, and generating first drafts for humans to review. Those deliver real value.
If you're being honest, most of what you call AI is probably workflow automation wearing a costume. And that's fine — automation is often the smarter, cheaper, faster win.
We'd rather tell you that than bill you for a chatbot you don't need.