Twenty-seven American manufacturing industries have enough data to compare how fast their productivity grew before 2005 against how fast it grew after. Two of them got better: tobacco, and leather and hide tanning.1

None of the other twenty-five improved, and that includes the industries the country talks about most. Computer and electronic product manufacturing averaged 1.9 percent annual productivity growth from 1987 to 2005, then averaged negative 1.2 percent from 2005 to 2023.1 Aircraft manufacturing gave up roughly the same three percentage points.1 Those figures were assembled in July by the Information Technology and Innovation Foundation, in a piece arguing that Congress should hand firms a 25 percent tax credit for adopting productivity-enhancing technologies like industrial robots and AI,1 so you know going in what the people doing the counting wanted the count to show. The series underneath their analysis is the government's own.

This is the sector that produced the improvement methods everyone else borrowed. Most of the vocabulary a software team reaches for in a retrospective, standard work and flow included, came off a factory floor. And the floor has spent close to two decades producing almost nothing extra per hour.

Where the gain came from

In March the Bureau of Labor Statistics published an article asking whether manufacturing productivity was recovering. Its long-run figure, covering the first quarter of 1987 through the third quarter of 2025, was average annual growth of 2.1 percent.2 Its figure for the business cycle running from the end of 2007 to the end of 2019 was 0.1 percent.2 For the cycle that began at the end of 2019 and ran through the end of 2024, the figure was 0.0 percent.2 Adding the first three quarters of 2025 lifts the current cycle to 0.4 percent.2

The headline number matters less than the chart sitting underneath it. That chart sets productivity, output and hours worked all to 100 in the first quarter of 2007 and runs them forward. By the third quarter of 2025, productivity is at about 106.6, while output has fallen to about 96.0 and hours worked to about 90.1.2

Read those three together and the arithmetic says something the ratio on its own never will. Output per hour rose across eighteen years during which the sector was making less than it made in 2007. Hours fell by close to ten percent, and that is the reason the ratio moved.

A productivity number cannot tell you whether an organization got better at something or only got smaller. Both push the number the same direction, in the same report, on the same day.

Five hundred people, then fifty

In a session I sat in on this spring, someone walked through what remote operations had done to offshore oil and gas. Crews of around five hundred on a rig had come down to something closer to fifty, with basic maintenance and inspections now worked through on video calls, and helmet cameras pairing one person offshore with someone on shore. The savings were itemized down to helicopter flights and beds offshore. I do not have a note of anyone giving the production figure.

Cost is what an operator meters continuously, because finance owns it and it arrives on a schedule whether anybody asks for it or not. Output needs a unit definition and somebody willing to argue about what counts, so it gets refreshed when someone gets around to it. The bottom half of the ratio is instrumented and the top half is estimated, and that asymmetry shapes your reporting more than most of the analysis built on top of it.

Nearly every efficiency figure a senior leader reads is built this way. Revenue per head, cost per unit, tickets closed per agent, story points per sprint. Each one improves when the top half grows, and just as readily when the bottom half shrinks, and the second route is faster and available in any quarter you happen to need it. Anybody who has run delivery has watched a per-person number improve in the quarter a team lost someone, and has watched nobody in the room say so out loud.

The three quarters that look different

Which is what makes 2025 worth pulling out of that BLS article. Measured at annualized rates, output grew 3.9 percent in the first quarter, 2.9 in the second and 3.0 in the third, while hours worked moved 0.4, then 0.0, then negative 0.7.2 For three consecutive quarters the sector made meaningfully more with roughly the same labor. On the productivity line that reads as more of the same, only steeper. Underneath, it is a different organizational event, and it is the only stretch since 2007 in which output grew that strongly while hours held flat or fell.

So the question to put to any efficiency claim is dull and it works. Which term moved? Holding output steady with fewer people can be a legitimate gain, but it is one you can only repeat so many times. Getting more output from the same people builds something that compounds. On a dashboard those two arrive as the same green arrow pointing the same way.

I wrote a week ago about how an adherence rate reaches a senior leader while the work it claims to describe stays on the plant floor, and I am not going to run that argument again here, because this one is narrower. The number is accurate. Two opposite things produce it. A cousin of the same problem turns up in schedule reporting, where I looked at programs counted as on time after resetting their own delivery dates and then measured against the new ones.

Here is what I would go do this week. Take the efficiency metric your leadership team is proudest of, find the two numbers it was built from, and write them down separately with dates against them. If the one on top has not moved in a year, what is it you are celebrating?