Historic lessons on why AI’s productivity payoff is an organizational problem, not a technology one
Moacyr Galo
Chief AI Officer / Head of sa.global labs
In 1900, 20 years after Edison patented the lightbulb, electricity was simultaneously everywhere and nowhere. The dynamo was the wonder of the age, and yet only about 3% of American homes had electric lighting, and electric motors drove less than 5% of the nation’s factory power. The most transformative technology of its century had arrived, and the economy could barely feel it.
We are living the same moment with a different machine. More than 80% of organizations have piloted tools like ChatGPT or Copilot, and boards talk about little else. Yet MIT’s 2025 GenAI Divide study found that roughly 95% of enterprise AI pilots produce no measurable impact on profit and loss.1 The technology is undeniably powerful. The results, so far, are undeniably thin.
The gap between those two facts is the most important thing a leader can understand about AI right now. And we have seen it before.
The electricity 40-year wait
Electricity didn’t show up in the productivity statistics in any serious way until the 1920s, roughly 4 decades after the lightbulb. The Stanford economist Paul David explained why in his 1990 paper The Dynamo and the Computer2, and the explanation is worth more today than when he wrote it.
The reason for the delay was not the motors. It was the factory.
Nineteenth-century factories were built around the steam engine: a single, massive central power source that drove every machine through an elaborate system of rotating shafts, belts, and pulleys. Everything followed from that one constraint. Machines were crammed close to the drive shaft because power was lost over distance; buildings went vertical and dense to keep the linkages short. The entire architecture was a monument to the location of the engine.
When electricity arrived, the first instinct was the obvious one: pull out the steam engine, drop in one big electric motor, and connect it to the same shafts and belts. The factory was now “electrified”. And it did almost nothing, because nothing about the work had changed. Owners had bought a new power source and bolted it onto an old design.
The breakthrough came a generation later, when a new set of managers stopped asking “How do I plug electricity into my factory?” and started asking “What would a factory look like if it had been born electric?” The answer was the unit drive: a small, independent motor on every machine. The central shaft disappeared, and the factory was free to be rearranged around the flow of the work rather than the flow of the power – single-story, spread out, lit, ventilated, organized for efficiency instead of proximity to an engine.
That redesign, not the motor, is what roughly doubled manufacturing productivity. The technology had been available for 40 years. The leverage only appeared when organizations rebuilt themselves to deserve it.
The AI pilot phase is the steam-engine swap
This is exactly where most organizations sit with AI today, and the MIT data describes it with uncomfortable precision.
We have collectively electrified the factory without changing the factory. A copilot gets dropped on top of an unchanged process; a chatbot is wired into a workflow designed for a world without one. It is the 1890s move, a new power source connected to the old shafts and belts. So, predictably, over 80% of organizations have adopted these tools while only individual employees feel much benefit; the enterprise P&L stays flat, and only about 5% of the more ambitious enterprise systems ever reach production1.
MIT is blunt about “why”. The core barrier is not infrastructure, regulation, or talent. It is learning – both the systems inability to retain context and improve, and the organizations inability to reshape themselves around the technology. Seven of nine major sectors showed heavy pilot activity but little structural change: experimentation without transformation.
The unit-drive lesson even shows up in the failure modes. More than half of AI budgets go to glamorous front-office demos while the real returns sit unclaimed in unglamorous back-office processes. Companies that partner with specialized providers succeed about 2x as often as those trying to build everything internally. And a “shadow AI” economy has emerged, where individual employees quietly capture the exact gains their organizations cannot capture at scale. The motors are running. The factory has not been redesigned.
How to avoid waiting 40 years
The lag is not a law of nature. It was a choice, made four decades running, by factories that refused to change shape. We can choose differently, and we can compress the timeline dramatically, because the diagnosis is already in hand and the technology now diffuses in months rather than decades.
What that demands:
- Ask the born-electric question: For each core process, stop asking “Where can I insert AI?” and start asking “What would this look like if it had been designed for AI from the beginning?” That is the unit-drive question, translated. This type of question requires a new team mindset where intellectual curiosity, agency, systemic thinking and reasoning by first principles skills will be highly valued.
- Redesign the work, not just the tooling: A model layered onto an unchanged workflow is the single big motor on the old drive shaft. The gains live in re-sequencing the work around the new capability.
- Go where the ROI actually is: Aim at the high-value, often invisible back-office processes and meaningful initiatives that “move the needle”, not the visible front-office theatre.
- Distribute the motors: Anchored in a strong technical foundation, push ownership out to enabled and empowered teams closest to the work, rather than concentrating everything in a central AI lab.
- Build for learning, and measure P&L: Favor systems and processes that retain feedback and improve over time; hold every effort to a business number, not a successful demo. The most important metric in AI era that drives business leverage is “revenue per employee”.
The call to action
The most expensive mistake an organization can make right now is to wait for the technology to get better. It is already good enough. The differentiator was never the engine – not in 1900, and not today. It is who rebuilds the factory around it first. That rebuild is precisely what sa.global empower is built to deliver.
empower is an operating system for AI:
It combines a technology accelerator: the platform that lets you put intelligence on every machine on the floor, with an “outcomes-based methodology” that keeps the focus where it belongs, on business leverage rather than pilots. It is the difference between swapping the motor and redesigning the factory: empower gives organizations both the engine and the operating model to reorganize around it, and ties the work to P&L outcomes from the start.
So, get in touch with any sa.global local office in Americas, Europe, North Africa or Asia-Pacific. Our team will pick one process that matters. Ask what it would look like if it had been born for this new technology paradigm. Redesign it around that answer using AI, give the work and transfer the know-how to the people closest to it, and hold the result to a number. Then do it again.
The companies that take electrification’s lesson seriously won’t wait 40 years for their productivity dividend. They’ll take it from the ones that are still swapping motors onto the old drive shaft and wondering why nothing has changed.