The Two Questions That Kill Bad AI Proposals in Under a Minute

By Steve Schroeder  •  EliteFlow Consulting

I sat through a vendor pitch last year that was, by the standards of the genre, excellent. Clean deck, credible logos, a demo that did what demos do. The proposal was to drop an AI layer across a logistics company’s planning workflow, and the room was leaning in. The COO turned to me at the break and asked what I thought. I said I had two questions, and if the proposal survived both I would stop being the skeptic in the room. It did not survive the first one.

I have watched a lot of AI proposals in the last few years, in a lot of rooms, and I have stopped evaluating them the way the deck wants me to. The deck wants a conversation about the model, the accuracy, the integration timeline. None of that tells you whether the thing will help. Two questions do, and they take under a minute, and they work precisely because the strong proposals answer them cleanly and the weak ones fall apart on contact.

Question One: What Does the Human Still Decide?

The first question is: what does the human still decide? Ask it and watch which way the answer goes. A proposal built on a real understanding of the work has a crisp answer. The AI flags the anomalies, and a planner decides which ones are worth acting on. The AI drafts the forecast, and a human owns the call to commit inventory against it. The boundary is drawn, and the drawing of it is a sign that someone thought about where machine output ends and human judgment begins. The weak proposal cannot answer this, because its unstated promise was that the human stops deciding. That is the tell. When a vendor cannot name what your people still own after the system goes in, they have not sold you a tool. They have sold you the fantasy that judgment is a cost you can delete, and judgment is the one thing in your value stream that was never the bottleneck.

Question Two: What Happens Without the Map?

The second question is the one that killed the logistics pitch: what happens without the map? Meaning, before we accelerate this workflow, do we know where value actually waits in it? Every AI proposal is a proposal to speed something up. So the only question that matters is whether you have measured where the slow part is, or whether you are about to spend a budget speeding up a step that was never the constraint. The logistics vendor had no answer, because the diagnosis had never been done. They were proposing to optimize a workflow nobody had mapped, which means they were proposing to make an unknown constraint arrive faster. Deployed without flow-based diagnosis, AI accelerates waste, and the second question is how you catch that before the invoice rather than after.

Deployed without flow-based diagnosis, AI accelerates waste — and the second question is how you catch that before the invoice rather than after.

Both Questions Belong to the Buyer, Not the Vendor

Here is the uncomfortable part, and it is not the vendor’s fault so much as the room’s. Both questions are ones the buyer should have answered before the vendor ever walked in, and almost nobody has. We evaluate the proposal on the vendor’s terms, the model and the accuracy and the timeline, because those are the terms on the slides. The two questions move the evaluation back onto the ground where it belongs: your work, your constraint, your judgment. A proposal that cannot say what the human still decides does not understand your work. A proposal that cannot say what happens without the map has not looked at your work. Neither is a model problem. Both are diagnosis problems wearing a model’s clothes.

The Questions Don’t Kill Good Proposals — They Reveal Them

The reframe is that these two questions are not a way to say no. They are a way to say yes with your eyes open, and they cost nothing to ask. The proposals that survive both are the ones worth funding, and they exist. When a vendor can tell you exactly what your planners still decide, and can point to a map that shows where the workflow actually waits, you are no longer being sold decoration. You are being offered decision time, and that is worth buying. The two questions do not kill good proposals. They just make the good ones easy to see by clearing the room of the ones that were hoping you wouldn’t ask.

You are no longer being sold decoration. You are being offered decision time — and that is worth buying.

So the next time a clean deck leans the room forward, you do not need to know anything about the model to know whether the proposal is real. You need under a minute. What does the human still decide, and what happens without the map?

#AIStrategy #ValueStreamMapping #FlowIntelligence #DigitalTransformation #OperationalExcellence #AIProcurement