Episode Details

• • •

The woman who ran quality on our second shift could hear a bearing starting to fail. Not with instruments—with her ears, over the ambient roar of an extrusion line pushing polymer at four hundred degrees. Three shifts before the vibration sensor flagged the anomaly, she would walk over to the maintenance lead and say, “Pull 2-B tonight or you’ll lose it by Thursday.” She was right so often that the maintenance team stopped questioning her and started scheduling around her. Twenty-three years on that floor had trained her nervous system into a detection instrument no algorithm has yet replicated.

Last year, a corporate team visited the plant to demo a predictive maintenance platform. Slick interface. Real-time dashboards. Machine learning trained on vibration, temperature, and amperage data from six hundred installations. The regional VP asked me afterward, “So how many heads can we take off the floor once this is running?”

He did not ask what she knew. He did not ask what she was worth. He asked how quickly she could be removed.

That question—and the accounting failure it reveals—is the subject of a study that dropped this week and should stop every manufacturing executive mid-sentence.

• • •

Thomas Davenport—the Babson professor who essentially invented the phrase “competing on analytics”—and Laks Srinivasan, co-founder of the Return on AI Institute, surveyed 1,006 C-suite executives across eleven countries and thirty-two industries. The headline finding, reported in Fortune’s CFO Daily: when CFOs are responsible for scoring AI outcomes, 76 percent of companies achieve a great deal of value. When anyone else owns it—the chief AI officer, the chief data officer, the CTO—the numbers crater.

Only 2 percent of companies actually give the CFO that responsibility.

Let that sit. The most effective lever for AI value extraction is used by one in fifty organizations.

Srinivasan explains this with a phrase that should be tattooed on every boardroom wall: “When finance gets involved, it brings institutional credibility behind numbers.” Davenport, for his part, notes a pattern he has studied for decades—technical capabilities arrive before the management systems to harness them. What makes AI different, he says, is how many consequential decisions, especially on workforce, organizations are making before those systems catch up.

And here is where the study turns from interesting to devastating.

• • •

Ninety percent of the organizations surveyed have already reduced or frozen hiring in anticipation of AI productivity gains. Not in response to demonstrated AI performance. In anticipation of it. Meanwhile, only 2 percent have made large headcount cuts tied to actual AI implementation. The actual headcount reductions in anticipation of AI are thirty times the number made from actual AI deployment. Srinivasan himself calls the gap plainly: headcount reductions and hiring freezes are running way ahead of evidence.

Two percent acting on evidence. Ninety percent acting on faith.

And the faith is already proving fragile. Some of the companies that laid off workers citing AI have quietly walked it back—reopening closed roles and rehiring the people they let go, after discovering that the AI tools required more human insight than anticipated. One survey found that a third of employers have already rehired between a quarter and half of the roles they initially cut. This is not a correction. It is a confession.

You would not write down a $50 million production line on a hunch. You would not scrap a CNC machine because a vendor showed you a brochure for a better one. Every piece of capital equipment on your balance sheet has a depreciation schedule, a maintenance history, a residual value assessment. Before you dispose of it, your CFO demands justification. Your auditors demand documentation. Your board demands a business c