Poll the room before you pick an AI pilot
The people who do the work know which task is worth automating. They will tell you, if their name is not attached to the answer.
Every AI kickoff has a moment where someone senior asks the room which problem to solve first. The room looks at the person who asked, then at the person who sponsored the meeting, and then agrees with whichever of them spoke last.
Nobody lied. They just answered a question about the org chart instead of a question about their week. That answer is the one that gets funded, and it is often the wrong one.
The loudest idea is not the most painful workflow
The ideas that arrive at a kickoff come from the people who were in the meeting where AI was discussed. The pain lives with the people who were not. The analyst who rebuilds the same report by hand every Monday, the engineer who reads every incoming ticket to route it, the contracts person who retypes the same clauses — they know exactly which hour of their week they would hand to a machine, and they are the least likely to say so in front of their manager’s manager.
We have written about how an assessment narrows a list of ideas. This post is about the step before that: getting the right ideas onto the list in the first place, from the people whose day would actually change.
Ask questions people can answer honestly
The fix is cheap. Put a few questions in front of the room and let everyone answer anonymously from their phone, before anyone discusses anything. Show the results, then talk. The order matters: once the sponsor has explained what they think, every answer after that is a vote on the sponsor.
The questions that tend to earn their place:
- Which recurring task would you give away tomorrow if you could?
- Where does the information for that task live today — a system, a spreadsheet, or somebody’s head?
- If a machine drafted it, would you check the draft, or redo it from scratch?
- What would have to be true for you to trust it without checking?
None of these ask about AI. They ask about work, which is the thing the room is actually expert in.
Read the disagreements, not the winner
The top answer is useful. The spread is more useful. When leadership’s favorite idea barely registers with the people who would use it, that is a pilot with no champion, and you learned it for free. When most of the room says the information lives in someone’s head, that candidate has a data problem before it has an AI problem.
The trust question is the one to watch. A task where people would redo the draft from scratch is a task where the time savings will not show up, however good the model is. A task where people would check and correct is where a pilot can prove something quickly.
What a poll does not replace
A poll is a signal, not a business case. It tells you where the pain is and where the trust is; it does not tell you what the thing costs to build or to run in year two. That still needs someone to sit with the data and price it. And in a very small room, anonymity is thin — people can guess who answered what, and they know it.
Used for what it is, though, it moves the conversation from who is in the room to what happens in their week. That is what FloorPoll is for: a six-character code on the screen, nothing to install, no account, anonymous by default. It fits a facilitated workshop as well as a conference talk.
If you want help running that conversation and turning its answers into a pilot, it is where our AI strategy work starts.
Written by Tommy Shrove, BluuAlpha Technologies.