AI's greatest challenge is managerial, not technological

A recent IBM survey of 2,000 executives on their expectations for artificial intelligence in 2030 revealed something noteworthy. They unsurprisingly predict that AI investment will surge (from already high levels) and that 79% expect AI will contribute significantly to their revenue. But strikingly, only 24% "clearly see" where that revenue will come from.Such lack of clarity might seem like a bad sign when most AI projects have failed to generate a return on investment, but it’s actually exactly what we should expect from a truly revolutionary innovation, and it makes clear that the greatest businesses challenges posed by AI will be managerial, not technological.

Revolutionary innovations rarely announce their business models — or even their use cases — in advance. They usually start with a simple one-to-one replacement where the innovations are a better or cheaper way of doing something that companies already do. Over time, users realize that they offer new and powerful capabilities. That’s when their real impact kicks in, and it explains why the same survey reports that executives expect their AI spending to shift from efficiency gains to product and service innovation.

Properly utilizing these new capabilities, however, usually requires businesses to reorganize in every way. That shift is usually the biggest roadblock to a new technology fulfilling its promise.

We’ve been here before. When electricity first entered American factories in the late 19th and early 20th centuries, its economic returns were disappointing. Thomas Edison invented the electric light bulb in the 1870s, but by 1900 less than 5% of the power used by American factories came from electric motors. Instead, power came from steam engines — often ones was shared by multiple factories — which drove machines via line shafts. Electric lighting allowed factories to work far more efficiently at night, but their fundamental operations remained unchanged.

The real transformation came later, once smaller and cheaper electric motors made it possible to abandon centralized power. Machines could be distributed more flexibly, and workflows could be rearranged to follow the logic of production. Just as important, these new processes could only be properly utilized by workers who were trained differently, had more independence and, eventually, were paid better. That reorganization made possible the moving assembly line and, with it, mass production on an entirely new scale. And it’s when electricity became a truly revolutionary innovation. It didn’t just make factories cheaper to run, it changed what factories could produce, how quickly they could adapt and which firms survived. Entire industries followed.

AI may be inching out of its replacement phase and into its re-architecture phase. Automating routine tasks and optimizing workflows is like swapping steam engines for electric motors. Useful, necessary and limited. The harder and more consequential work lies in redesigning processes, products and decision-making when machines can generate content, interpret unstructured information and act autonomously within defined boundaries.

That kind of redesign is uncomfortable and often expensive (pity the early 20th-century industrialist whose factory had to be rebuilt thanks to some dancing electrons). It doesn’t fit neatly into existing organizational charts or capital budgets. It often produces temporary declines in measured productivity as firms experiment, fail and relearn how work should be done. When information technology was being rolled out in the 1970s and 1980s, this effect was so pronounced that Nobel Prize-winning economist Robert Solow quipped, "You can see the computer age everywhere but in the productivity statistics." And the combination of the difficulty of redesign and the unknowability of what these new capabilities will be used for makes it difficult or impossible for executives to answer the seemingly basic question of where, exactly, the revenue is going to come from.

History suggests that demanding too precise an answer too early is a mistake. The executives who installed electric motors in their factories in 1905 had no way of knowing that assembly lines would soon transform manufacturing or that their factories would soon be producing products that didn’t exist yet. The ones who required a detailed forecast before reorganizing production were the ones most likely to be left behind.

The same risk exists today. If you only see AI as a cost-cutting tool, you might protect margins in the short run, but you might also be trapped by optimized versions of soon-to-be-obsolete business models. Once competitors begin to redesign offerings, pricing and customer relationships around AI-enabled capabilities, incremental efficiency gains will no longer be enough.

None of this guarantees that today’s AI bets will pay off. Many early electrified factories failed. Many — likely most — AI-driven initiatives will too. But the beginning of wisdom is admitting what you don’t know, and the executives admitting they don’t yet know how AI will generate revenue are the ones who understand that the next phase of this technology isn’t about execution within known boundaries, but about discovering entirely new ones.

By the time the destination is obvious, someone else will have already reached the goal.

This column reflects the personal views of the author and does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners. Gautam Mukunda writes about corporate management and innovation. He teaches leadership at the Yale School of Management and is the author of "Indispensable: When Leaders Really Matter."