Most AI transformations are solving the wrong problem

We are told the winners of the AI era will be the firms that deploy the most, fastest. Roll out copilots, find internal champions, drive usage, track productivity gains. None of that is wrong. It is also just table stakes. The real divide is between firms that use AI to do today's work faster and firms that use it to build an organisation that gets better each time the work is done.

History offers a warning in the form of General Motors. In the 1980s it poured $45 billion into factory robots, more than Toyota and more than any manufacturer had yet committed to automation. Market share, quality and productivity barely improved. Toyota outperformed it with far less capital because it paired automation with fewer management layers and tighter feedback from the factory floor to decision-makers. GM invested in the technology. Toyota redesigned the work around it.

GM's mistake shows a pattern rather than a one-off. Nokia spent more on research than Apple before the smartphone era ended its dominance. Sears launched digital initiatives years before its decline became terminal. Blockbuster built an online service, yet its store network still defined the business. In each case, the pattern is consistent: firms see the technology, fund it and still lose, because they remain loyal to an operating model built for the previous age.

The AI era is following the same script. The mistake is simple: most firms are digitising a funnel, not building a loop. Work goes in, output comes out faster and cheaper, but the organisation itself learns little that sustainably improves the next cycle. The advantage does not compound.

The mistake in the AI era is simple: most firms are digitising a funnel, not building a loop.

The PC era already showed why this matters. Personal computers stripped out much of the work that had justified layers of middle management: gathering information, pushing it upward, and translating decisions back down. The layers that remained were the ones still adding judgment. AI is now moving onto that ground. This is not just another wave of automation. It is a deeper challenge to the logic of the hierarchy itself. A firm that deploys AI into an unchanged structure is repeating GM's mistake. One bottleneck is removed, but the organisational constraint stays in place.

Those organisational constraints usually show up in two places. The first is structure: how work is divided, how decisions travel, where coordination costs sit, and which layers still exist because the old technology required them. The second is incentives: what gets measured, rewarded, protected, and punished. Most transformations fail when firms change the tools but leave either of those two things intact.

That is why so many transformation programmes stay cosmetic. The technology may improve. The slides may improve. Even local productivity may improve. But if the structure still slows the right decisions, and the incentives still reward the old model, the organisation cannot compound what it is learning. The strategy points forward. The system remains anchored in what made the organisation work in the past.

Firms that break this pattern do something different. They redesign work so each cycle leaves something useful behind for the next one. What the organisation learns is captured as part of doing the work, kept accessible, and returned quickly enough to shape the next round of decisions. That is what turns output into capability. Miss any link in that chain and the work may get done, but the organisation does not get structurally better at doing it.

The test for any AI transformation is blunt. After the work is done, what remains besides the output? If the answer is only a completed task, the organisation has built a faster funnel. If the answer is better judgment, reusable context, and sharper coordination for the next cycle, it may be building a compounding loop.

That question has practical force. It tells you where to look. Where does useful judgment vanish? Where does work get completed without leaving the system any better? Where do incentives still favour preserving the old model? That is where the real transformation work still sits. In most AI transformations, the binding constraint is no longer technical. It is organisational.

The firms that win the AI era will not be the ones that moved first or spent most. They will be the ones that use AI to build organisations that learn faster, decide better, and coordinate more effectively with each cycle of work. That is the real choice hidden inside AI transformation. Not whether to deploy the technology, but whether you are willing and able to redesign the system around it. Treat AI not as a tool to bolt onto the organisation you already have, but as a stress test of whether that organisation still deserves to exist in the market AI is reshaping.