GPS told us where every van was. It did not tell us which one was carrying what the customer needed.

Almost twenty years ago I was running operations and IT for a medical equipment delivery company. We put GPS on the fleet, and dispatch did the obvious thing with it. When a call came in, they sent the closest van.

The closest van was often the wrong one. A delivery needs the right equipment, and the nearest truck might not have it on board. A pickup needs room, and the nearest truck might already be full. So the closest van drove back to the warehouse, loaded up, and went out again, or went back to unload before it could pick up a larger item.

GPS answered the question it was built to answer, which is where everything is. That was not the question dispatch was asking. Dispatch needed to know which vehicle could actually complete the call, and location is only half of that.

The other half was already in the building. We had bar codes on the equipment and scanners in hand, and every item was scanned out of the warehouse and scanned again at delivery. We had great data on where items were. So I added one rule. Every van became a warehouse. Anything loaded onto it was scanned in, and anything that came off was scanned out. A delivery counted as a scan out to the facility.

Once that data existed, the team could ask the question dispatch actually had: which van with the equipment was closest, or would be closest soon. Sometimes a van without the equipment was right next to the facility, and a van farther away on another run had it. That van got the dispatch, because it had what was needed. For a pickup, we looked for the closest van with room for the item. That cut out trips back to the warehouse several times a day.

The tool arrived before the question. We installed GPS because tracking the fleet was obviously useful, and it was. But we had not named the decision it was supposed to serve, so its first use answered an easier question and surfaced the real one.

I see the same pattern with AI tools now. The capability shows up first, it is genuinely good at something, and that something becomes the use case by default. Nobody stops to ask which decision it was supposed to improve.

The fix is usually smaller than another purchase. Ours was a scanning rule we already had, applied to one more place.

What decision is your newest tool supposed to improve, and has anyone written that decision down?