Five Questions to Answer Before You Build Anything with AI
Before you write a line of code or pick a model, answer these five questions. They are the difference between an AI project that ships and one that becomes an expensive science experiment.
Enthusiasm is the most dangerous phase of an AI project. Everyone wants to start building immediately, because building feels like progress. But the projects that succeed almost always spend their first days not in code, but in a handful of uncomfortable questions. Here are the five I insist on answering before anything gets built.
1. What decision does this improve?
Not what does it predict — what does it change. Every AI system exists to make some human or automated decision faster, cheaper, or more accurate. If you cannot name the specific decision and who makes it today, you are building a solution in search of a problem. Name the decision first.
2. What happens when it is wrong?
Every model is wrong sometimes. The question is what that costs. A recommendation engine that occasionally suggests the wrong film is harmless. A model that denies someone a loan is not. Your tolerance for error should dictate your entire design — how much human review you build in, how you handle edge cases, and how transparent the system needs to be.
3. Do you actually have the data?
Not data in the abstract — the specific, labeled, accessible data this problem requires, delivered continuously. Most stalled projects discover too late that the data they need is trapped in a system no one can export from, or does not exist at all. Verify the data pipeline before you fall in love with the idea.
The model is 20% of the work. The data pipeline and the workflow are the other 80%.
4. Who owns it after launch?
A model is not a project with an end date; it is a system that needs an owner for its entire life. Someone has to watch for drift, retrain on fresh data, and respond when it misbehaves. If you cannot name that owner before launch, the system will quietly rot after it.
5. How will people actually use it?
Intelligence that lives in a separate dashboard nobody opens is worthless. The best AI disappears into the tools people already use — the CRM, the inbox, the workflow they live in all day. Design for the point of decision, not for a demo screen.
Answer these five honestly and you will kill some bad ideas early — which is exactly the point. The ideas that survive all five are the ones worth building, and they tend to ship. That is not a coincidence.