Part 1 — Cast the Net

Six to ten real candidates, sourced on purpose — not the one idea that happened to walk in the door.

Most AI initiatives start backwards. Someone in leadership hears about a tool, or a vendor gets fifteen minutes in a meeting, and that idea becomes “the AI project” — not because it was the best use of the next quarter’s attention, but because it was the only one in the room. The fix isn’t a better idea. It’s more of them, gathered on purpose, so the one you commit to actually beat something instead of arriving unopposed.

An inventory doesn’t need to be exhaustive. It needs six to ten real candidates, sourced from three places that consistently produce better material than a brainstorm.

Shadow use first. People are almost certainly already running ChatGPT, Copilot, or something similar on personal accounts for parts of their job. That’s free market research: wherever someone already reaches for AI without being told to, the appetite and the workflow fit are proven before you’ve spent a euro.

The complaint log second. Whatever customers, staff, or partners repeatedly flag as slow, error-prone, or frustrating is a use case wearing a disguise. Pull the last two months of support tickets, internal gripes, or the line staff say out loud in every second meeting — “we always lose time on X” is a use case statement, even when nobody phrases it that way.

The report audit third. Ask which spreadsheet or report takes the longest to compile by hand each week. These are usually strong candidates for a structural reason: the data already exists somewhere, it’s just not connected to anything — which means the hardest part of most AI projects, finding usable data, is already solved.

For each candidate, write one sentence: who does this, how often, and what it currently costs them in time or money. Not a business case yet — just enough to compare it to the others on the list. A one-line entry that says “invoicing follow-up, done twice a week, four hours each time, chased manually” is worth more at this stage than a polished pitch for a single idea with nothing beside it to compare against.

By the end of this step you should have a written list of candidates — not one you’re already attached to, and not the most impressive one, just the most complete one you could put together in an afternoon of asking around.