Why does new technology take so long to pay off?

On stage, I tell rooms where this is all heading. AI stops being a project and becomes how you operate. "It is literally like electricity where we just plug it in and it just works."

That's where it ends up.

Getting there is the part nobody budgets for. In McKinsey's 2026 survey, about 80% of respondents said AI has improved their own productivity. Only 37% attribute any profit impact to it, about the same share as the year before.

Your people feel faster. Your numbers look about the same.

That gap has a cause, and it's happened before. With electricity, it took about four decades to show up in factory productivity. What finally closed that gap tells you where to look in your own company.

Faster people, same company

If you run a one or two person operation, or even a small team, you can see big gains from AI right out of the gate.

When you're small, the person who gets faster is the workflow. There's nobody to hand the work to and no approval chain to wait on. The time you save shows up in the business the same week.

That's been true in my own business. I didn't add AI to the way I used to find speaking opportunities. I rebuilt how I find and evaluate them, and that became my outreach engine.

Same with software. I stopped judging tools one at a time by their features. Now my tech stack and my business work as one world, and a new tool has to fit everything else before it comes in.

The larger the organization, the harder that gets.

Individuals still see real gains. In a study of 5,172 customer-support agents by economist Erik Brynjolfsson and two colleagues, AI assistance raised productivity 15% on average, and about 30% for less experienced and lower-skilled agents.

But in a big company, each person's work runs through handoffs, approvals and other teams. The gains are real, and they stay private.

The pattern I see most often is everyone speeding up alone. Each person makes their own piece faster with their own AI. Nobody shares what's working. The handoffs between them never change.

Getting individuals to use AI is the first phase, and everything starts there. Turning those gains into team and organization value is a different challenge.

Electricity shows why it takes so long.

Electricity, then computers

In 1899, electric motors supplied less than 5% of the mechanical power in American factories. They passed half around 1919 to the early 1920s, about four decades after the first central power stations opened.

Only then did electricity show up in factory productivity.

Part of the delay was price. Stanford economist Paul David, who studied how factories took up electricity, found adoption only gained real momentum after 1914 to 1917, when regulated electricity rates fell. Engineers also had to work out the new layouts plant by plant.

And the factories were already built. A steam plant ran on a line shaft: one long spinning bar along the ceiling, with a belt running down to every machine. Plants had the engines, the shafts and the buildings, and they all worked.

So retrofitting a steam plant "typically entailed adding primary electric motors to the original stock of equipment." A brand-new power source, running the exact same layout.

Sound familiar? That's most large companies with AI right now. The licenses are bought, the logins are handed out, and the work still runs through the same steps, handoffs and approvals as last year.

The line shaft in most companies today is the old workflow itself. Everyone's new speed still runs through it.

San Francisco Fed President Mary C. Daly put it in the same terms: using AI on the old work is like "replacing a steam-powered motor with an electric one but leaving the factory floor unchanged."

The change came with unit drive, which gave each machine its own motor. Once the shafts were gone, David wrote, factory structures "could be radically redesigned." An engineer reported in 1901 that the gain from the new layout could reach "20 or 30 percent or even more, with the same floor space, machinery, and number of workmen."

Same building, same machines, same people. Different design.

From 1909 to 1919, motors went from about a quarter of factory power to about half, and output per hour rose only about 12%. From 1919 to 1929, as unit drive spread, it rose about 72%. Those endpoints flatter the 1920s a little, and hours worked fell too, so part of the jump is fewer hours.

And the plants that changed early didn't wait forty years. A county-level study of US electrification found the gains "were rapid and long-lasting, in contrast with established narratives."

The long wait is an average. Early redesigners gained early.

Computers repeated the pattern. They spread through offices in the 1980s, and the gains only showed clearly in the late 1990s. In 1987, Robert Solow, the MIT economist who won the Nobel Prize that year for his work on economic growth, wrote, "You can see the computer age everywhere but in the productivity statistics." About thirteen years later, he told the New York Times, "You can now see computers in the productivity statistics."

Brynjolfsson and economists Daniel Rock and Chad Syverson explain the gap. Technologies like AI "enable and require significant complementary investments": new processes, new skills, new ways of organizing work. You pay first. The numbers catch up later.

They call it the productivity J-curve. In a factory, that investment was a new floor. In a company using AI, it's something more human.

Redesign is a people choice now

Here's where I part ways with the forty-year story. I don't think AI takes anywhere near that long to pay off.

Line up electricity's brakes against AI, and most of them have come off. Trying AI costs little, and it doesn't need a new building.

The one that's left is people. They still have to work out the new layout, one team at a time. That brake moves as fast as leaders decide it should, and no vendor can make that choice for you.

In 1920, the new layout was an engineering job. Architects and electrical engineers redrew the floor, and the building changed.

With AI, the floor is how people work together: who hands what to whom, what needs approval, what gets shared. That layout is made of habits.

And habits change when people see someone else change first.

That's why, when a CEO tells me they've had AI for a year and don't see it in the numbers, I ask two questions. How many of your leaders are using AI themselves, and showing their people how it helps them? And how often do you let your employees and teams show off what they're using AI for, and recognize it, both what's working and what they learned from the experiments that didn't?

Those two questions get at whether individual gains can turn into shared ones. When leaders go first, people see it's safe to try. When teams show their work, one person's speed-up becomes something the whole team can build on.

Once it's shared, a team can rebuild the workflow around it. That's the step from individual adoption to team and organization value.

The factory data says the same thing. A Census Bureau study of tens of thousands of US manufacturers found that plants adopting industrial AI usually dipped before they grew. Older plants took the biggest hit, partly because routines like tracking KPIs and production targets broke down during the switch.

That was industrial AI, from 2017 to 2021. The hard part is the same today: the routines, and the people who run them.

Isn't AI different?

In two ways that matter. People adopt it far faster, and economists disagree on how big the total payoff will be.

Start with adoption. According to the St. Louis Fed, 54.6% of US adults 18 to 64 used generative AI by August 2025, about three years after ChatGPT. At the same point, PC adoption was 19.7% and internet adoption 30.1%.

Faster adoption doesn't mean faster redesign on its own. Mostly it means more people speeding up alone.

But it's also why this can move faster than electricity did. The people are already using it, so a team can redesign as soon as its leaders make room for it.

So how big will the payoff be? Economists don't agree. One of the most skeptical is Daron Acemoglu, the MIT economist who shared the 2024 Nobel Prize in economics.

He estimates AI will add no more than 0.66% to total factor productivity (the extra output an economy gets from the same people and equipment) over ten years, and possibly under 0.53%. That's less than one percent across a whole decade, and it's the low end of a live debate.

Whatever the number turns out to be, it'll show up unevenly. The companies that change the work are the ones that capture it.

Paul David himself warned against stretching the analogy. "Computers are not dynamos," he wrote. But he also wrote a line worth keeping on every leader's desk:

Factories wore out and buildings got replaced. Your workflows won't wear out on their own.

Making the speed-ups visible

The history turns into a short plan.

First, stop judging AI by the installation. Licenses and logins tell you the motor is in the building. They don't tell you whether the floor changed.

Then make the individual gains visible. Give people a regular place to show what they're using AI for. Recognize both what works and what they learned from the experiments that didn't. Leaders go first, with their own examples.

Once the speed-ups are out in the open, rebuild one workflow around them, on purpose. Ask what the work would look like if you designed it today, with AI in the room from the start. The workflows that tend to show results run often, pass through a lot of hands, have an owner who wants them changed, and already have a result you can measure.

Plan for the dip. Nobody can promise how long it lasts, and in the Census factory data it played out over years. Plan for it anyway, so nobody kills the change before it pays off.

And measure the work itself: cycle time, quality, and what your people do with the time they get back.

What this means for you

If you run a small team, the gains are there right now, so take them.

If you lead a larger organization, getting people to use AI is phase one, and it's worth doing well, because everything starts with individuals. Phase two is getting those gains to add up, and that's a leadership choice.

The motors are already in your building, and your people use AI every day. Electricity's forty-year wait was an average, and this time most of the brakes are off.

How long it takes you comes down to whether someone redraws how the work gets done.

So who's redrawing yours?

The payoff came the day someone redrew the floor.

Key takeaways

1. 01Individual AI gains come first, and they're real. Small teams feel them right away.
2. 02In larger organizations, those gains stay private until the work around them changes.
3. 03Electricity took about four decades to show up in factory productivity. The payoff came when factories redesigned the floor, and early redesigners gained early.
4. 04AI doesn't have to take forty years. Most of electricity's brakes are gone, and the one left is people.
5. 05Redesign is a people choice now. It starts with leaders using AI themselves and teams showing what they're trying.

Questions

Sources