Most organizations investing heavily in AI aren't getting the return they expected. By whatever measure you pick, most companies aren't yet seeing it: only around 6% report significant enterprise value from AI, fewer than a third report meaningful ROI from generative AI, and over 80% see no real bottom-line impact at all, even as nearly all of them now use AI in regular operation (McKinsey, 2025; WRITER, 2026).
The instinct, when the numbers look like that, is to assume the technology fell short of the promise. It usually didn't. The most consistently cited reason these initiatives underdeliver isn't the model, the tooling, or the integration. It's that organizations bought and deployed before they decided — in any concrete, business-anchored way — what they were trying to accomplish, and clearly communicated that to the people whose work would change to bring them along.
The technology is sometimes the problem, e.g. data that isn't ready, or systems that won't connect. But far more often, the real limit is the one nobody put on the project plan: a strategy that ties the experiments back to something the business actually values, and the work of helping people understand how their roles will adapt to support that strategy.
None of this is news. We've seen this before.
The same lesson, relearned and re-forgotten
This isn't the first technology that arrived promising to remake how organizations run, and it isn't the first time the promise broke on the same ground.
In the early 1990s, business process reengineering made the case that you had to redraw the whole operating model around the new capability, or you were just, in the language of the time, paving cow paths. The movement produced real results and a great deal of wreckage, and its post-mortems landed on a now-familiar verdict: where reengineering failed, it usually wasn't the redesign that sank it but the organization's inability to carry its people through the change. The lesson was there to be learned. Then the field moved on, and the next wave started over.
Enterprise resource planning ran the same course later that decade and into the 2000s. The post-mortems converged on a finding that has aged remarkably well: ERP was never really a technology project. It was an organizational transformation that happened to involve technology, and the human and organizational factors mattered several times more than the technical ones to whether the investment paid off.
Digital transformation made the case a third time in the 2010s and reached the same verdict. Widely cited research put the share of efforts that fell short of their goals high and consistently found that the reason was culture and people, not the technology. The differentiator was rarely the digitization method and almost always the organization's capacity to absorb the change.
Three waves. The same two instincts each time — redesign everything, or ship something small and let it spread. The same dueling research. And, underneath the method debate, the same finding in the wreckage: the method was never what separated the wins from the losses. What separated them was whether the organization did the slow work of deciding what it was actually trying to achieve, and whether it carried its people through the change or skipped both to start building.
What's genuinely different this time (and what isn't)
It'd be too neat to say AI is simply the old story in new clothes. Two things really have changed.
The first is the cost of trying. Reengineering and ERP made "redesign first" attractive partly because experiments were expensive; you couldn't easily test your way forward, so a large up-front bet felt rational. AI inverts that. You can stand up a pilot in an afternoon, which means the incremental path is cheaper and more available than it has ever been. That should make disciplined, strategy-anchored experimentation easier. Instead it has made scattershot experimentation easier, because the friction that used to force a moment of planning is gone.
The second is that AI increasingly acts. Prior waves automated and connected; agentic systems take actions on their own, which introduces a governance and accountability problem the earlier transitions never had to solve.
But the thing that hasn't changed is the thing that matters most. In every wave, the organizations that won were the ones that decided what they were doing and why before they committed, and that treated the human side of the change as central rather than as training to be scheduled later. The technology changed. The real limit didn't.
Why the lesson never sticks
Here's the question worth asking: if this lesson has been documented in three consecutive technology waves by major research firms, repeatedly, why does every new wave's organizations behave as though they're the first to run into it?
It isn't ignorance. The research is not hidden. Most leaders, asked directly, will tell you the people side is what makes or breaks this kind of change. They already know.
The honest answer is that the strategy-first, people-first path is the one the incentives quietly punish. Doing the up-front work of deciding what business outcome an AI investment is actually for, and how the work and the workforce have to change to deliver it feels like delay at exactly the moment everyone wants to be seen moving. A pilot you can announce. A strategy you can't. When roles, budgets, and bonuses are on the line, the rational individual move is often the visible one, not the valuable one. And a series of rational individual moves is how an organization ends up with twelve disconnected pilots and no return.
That's not a flaw in any particular leadership team. It's a standing feature of how organizations behave under pressure, and it's precisely why the lesson evaporates between waves. Each generation rediscovers it in its own post-mortems, writes it down, and then the next pressure cycle rewards the visible bet all over again.
So here we are again
Which brings us back to where we started. Fewer than a third of companies seeing real return from significant AI investment; the rest having rushed to tools and projects ahead of any strategy that connects the work to value or brings the workforce along. It reads as a new crisis. It's a thirty-year-old pattern wearing this decade's technology.
The solution isn't a new framework. The organizations that'll get a return on AI are the ones willing to do the unglamorous thing the winners did in every prior wave: stop long enough to decide, honestly, what they're trying to accomplish and whether they've prepared their people to deliver it before they decide what to buy or how much to rebuild.
It's a discipline, not a tool. And the reason it's worth saying out loud one more time is that being able to recite the lesson has never been the hard part. Holding onto it when the pressure is on is.