Most businesses we meet are not short of AI ideas. They are short of a way to tell the good ones apart from the ones that simply sound impressive in a meeting.
The difference matters, because a fashionable AI project costs the same to build as a valuable one. It just does not pay you back.
Start with friction, not technology
A real opportunity almost always begins as a complaint. Someone re-keys the same information into two systems. Quotes take three days because one person has to touch every one. Nobody can answer a simple question about the pipeline without exporting a spreadsheet.
That is friction, and friction is measurable. A fashionable opportunity usually begins the other way around: with a tool someone saw, looking for a job to do.
If you cannot name the person whose week gets better, it is not an opportunity yet.
Size it in four numbers
Before anything is built, we put every candidate through the same four questions. They are deliberately blunt.
- How many hours a week does this consume today, and at what cost?
- What revenue is currently lost or delayed because of it?
- What decision gets made better, or sooner, if this is fixed?
- What will it cost to build, run and support for twelve months?
You do not need perfect figures. You need honest ones. A rough estimate that everybody agrees with is far more useful than a precise number nobody believes.
When the first three clearly outweigh the fourth, you have an opportunity. When they do not, you have an interesting idea that can wait.
The tests a fashionable idea fails
Three patterns come up again and again in ideas that should not be built yet:
- The volume is too low. Automating something that happens four times a month will never earn back the build.
- The process is not agreed. If two people do the same task two different ways, AI will simply reproduce the disagreement faster.
- The data is not there. If the information the model needs lives in someone's inbox, the first project is fixing the data, not adding AI.
Build the smallest thing that proves the return
Once an opportunity clears those tests, resist the urge to solve the whole department. Take one workflow, instrument it so you can see the before and after, and ship it.
A working improvement in a single process does two things a strategy document cannot. It returns money, and it earns the trust you need for the next one.
That is the whole method: find the friction, size it honestly, build the smallest version that proves the return, then measure it and go again.