A new Forbes piece on manufacturing’s AI spending stopped me cold with a single number: zero.
In a Grant Thornton survey, 64% of manufacturers said AI delivered efficiency gains. Not one of them could document a meaningful revenue or cost saving from it. Across other industries, 12% could. On the floor, we have a name for a result everybody feels but nobody can find in the numbers. We call it a story.
The article’s diagnosis matches what I see. Nearly half of these projects are stuck in pilots. The most common reason companies bought in wasn’t a specific, costed problem — it was competitive pressure. They bought the tool first and went looking for a use.
One line from the piece is going on my wall: “a pilot with no P&L target can’t really succeed or fail.” It can only continue.
That’s not an AI problem. It’s an operations-discipline problem — and it’s the one I keep hammering. Before you build or buy anything, name the value out loud, and name which kind it is: a hard dollar on the P&L, capacity you’ll actually redeploy, a capability you didn’t have before, or a better decision that avoids a failure. If you can’t name the number the tool is supposed to move, you don’t have a project. You have a subscription.
“Felt faster” is the trap. AI almost always feels faster — it’s fluent, quick, and it takes work off your plate. But feeling faster and moving a number are two different things, and only one of them shows up at year-end. The efficiency was probably real. It just never got converted — nobody redeployed the freed hours, nobody tied it to a metric, so it evaporated.
The fix in the article is refreshingly boring: treat AI like any other capital investment. Start from a costed problem someone actually feels. Name the metric before the pilot starts. Give it an owner and an exit date. That’s not a data-science move — it’s the same rigor you’d bring to a new line or a capex request. You already know how to do this.
So what? A pilot that can’t fail isn’t a safe bet. It’s a slow leak — budget and attention draining into something that will “continue” until someone finally kills it.
Now what? Pull up your AI pilots and ask one question of each: what number is this supposed to move, and by when? The ones that can’t answer aren’t pilots. They’re line items waiting to be cut. Give them a number, or give them an end date.
Source: “Manufacturers Rushed Into AI. The Returns Aren’t Showing Up,” Robert Szczerba, Forbes, July 2026.
Leave a comment