Industry 4.0's Productivity Paradox

When factories first electrified, they took out the central steam engine and put in one big electric motor. Productivity did not follow. Electricity was commercially viable by the end of the 19th century, but its payoff in manufacturing did not show until the 1920s and 1930s, 25 to 30 years after the electric motor. It came when factories put a motor at each machine and rearranged the floor around them. Hillel Zand of Maniv told the story to explain a puzzle, and co-host Jan Marchewski of Hitachi Ventures gave the puzzle a name.

The puzzle is in the statistics. US manufacturing labour productivity grew 3.4% a year on average between 1987 and 2007, Hillel said. From 2010 to 2022 it fell by an average of half a percent a year. In the EU, growth dropped from 1.5% a year in 1999 to 2008 to less than 0.4% in 2024. All this while investors like the two hosts put money into what was called Industry 4.0. Replace a pick-and-place worker with a pick-and-place robot, Hillel said, and some gains follow, but not the kind that show up in the macroeconomic figures. Floor layouts, labour incentives and org charts are part of the problem.

The line Ford drew

Jan called it the Solow paradox. Firms invest heavily in new technology and the gains appear in productivity statistics only decades later. Car plants still follow the logic of Henry Ford's moving assembly line of 1913, from stamping and the body shop to the marriage of body and powertrain and final assembly. Most innovation since has swapped in cheaper or better components without changing that architecture. Incumbents are bad at architectural change, Jan said, because the shop floor mirrors how they make decisions and set budgets. Changing one step affects the people who control the steps upstream and downstream, so a restructuring has to come first.

The counter-example was Tesla's unboxed process, planned for the Cybercab. Five modules of the car are built in parallel and joined at the end. Tesla says it can cut the cycle time of mass production from 60 seconds to 10. Jan said that is still to be proven.

Hillel made the same point about digital twins. You can put a twin on any process, but if nobody rethinks the process, and management resists what the twin suggests, "you're putting lipstick on a pig." New entrants find it far easier to design a new floor than a Ford, a GM or, with apologies to the co-host, a Hitachi. Hillel named Tesla and Hadrian as examples. A founder building an electric aircraft said the same. Even at Airbus's scale, much of the layering of wings is still done by hand. Cars take a few hundredths of a full-time worker per seat to build, aircraft still whole ones. The founder treats the plant as the company's second product and sees automation as the way to keep manufacturing in high-wage Europe.

You don't really want to see that much productivity gain

A case against the premise

Participants offered other explanations. One said the big gains of the eighties and nineties came from just-in-time and kanban, adopted from Japanese manufacturers, and nothing comparable has followed while products have grown more varied. Another recalled that a BMW 3 Series could be ordered in 3.5 million configurations years ago. An investor said older decision-makers see no reason to change a process that works. Another said plants in rural America can no longer staff a third shift, so machines stand idle for part of the day although the demand is there.

The sharpest challenge came from a participant who had spent years living in factories. Surface-mount machines for circuit boards replaced hand work, but what they delivered was lower cost, and customers constantly demand lower cost. "You don't really want to see that much productivity gain," they said, because goods would become expensive and stop evolving. Agriculture and construction were different, they added, since cutting waste there is how productivity rises. A founder whose company typically replaces three to five operators per line with one person or a robot asked why robot-heavy China does not change the picture. Nobody had the figures to hand.

Start with data

One participant saw the biggest opportunity in digital twins that hold the complete history of every component. In injection moulding, as someone added, that means the mould temperature, the flow, the chemistry and what made a part fail. Such data could feed models that design better products than engineers can, as semiconductor makers already do before they design their equipment. Another participant said the difference between today and Ford's era is data, and that factories should collect it before thinking about physical AI or cobots.

Asked what is investable now, participants pointed to coatings, adhesives and other layers that are still applied by analogue means, and to quality control, because a line that runs three or five times faster gains little if poor yield means whole batches must be run again. The most telling example came from nuclear maintenance. For every hour an operator spends on a reactor, one participant said, seven go on reports, and the process often ends with a printed document that has to be signed. Industry 4.0 has been talked about for decades. On paperwork like that, it has barely begun.

This Ripple was hosted by Hillel Zand (Maniv) and Jan Marchewski (Hitachi Ventures) at The Drop 2026 on 16 September.