When a team comes to me asking to buy an AI tool, I don’t start with the tool. I ask four questions, all about the workflow.
First. What’s the current workflow? Not in the abstract. For three real outputs from last week, write down step by step what a specific person actually did, with what inputs. Not a diagram. A list. Most proposals stall here, because nobody has written down what the work actually is. Half the value of the question is in forcing that list into existence.
Second. Which step is the tool supposed to replace, and what does that step cost right now? Minutes per week, or dollars per output. Something you can count. If the answer is a feeling about productivity, the team is trying to buy a tool to replace a feeling, and the tool won’t land anywhere specific enough to be measured later.
Third. What’s the workflow after the tool is in place, and who owns the change? Write that down too. If the answer is ‘the team will figure it out’, the team will use the tool for a month, abandon it, and nobody will know why. Someone has to own the redesigned workflow. Ideally that person isn’t the same person making the case for the tool, because the person pushing the purchase is exactly the wrong person to judge whether it’s working.
Fourth. If the tool disappears in twelve months, what’s still in place? Tools get acquired, deprecated, repriced, or folded into something else. The workflow redesign should outlast the particular tool that triggered it. If the answer is ‘nothing’, the team was renting a workflow change rather than buying one, which is a much smaller bet than the purchase order implied.
None of these questions need great answers on the first pass. They just need answers. A team that can’t answer them isn’t ready to buy the tool. A team that can usually doesn’t need my approval anymore, because the thinking the questions were meant to force has already happened. Procurement is the easy part.