Room Left to Change

In a Fortune essay published yesterday, senior advisory partner Ming Lu argues that the faster we innovate, the slower we grow.¹ While artificial intelligence is spreading rapidly, its effect on average productivity barely registers. A recent survey by McKinsey found that 88% of organisations use AI in at least one business function, but only 6% of all organisations surveyed are high performers.²

»Companies aren’t ancient civilisations, and inefficiency isn’t collapse. But the underlying problem is recognisable: every process exists for a reason, and as they accumulate, the problems they were built to solve may already have changed or disappeared«

There are two ways to read this. One is that AI has yet to deliver on its promises. The other is that while a handful of companies are doing extremely well, most are getting nothing out of it – and the average between the two blurs them into a single disappointing number. In »The Median Isn’t the Message«, the evolutionary biologist Stephen Jay Gould argued that variation isn’t an imperfection surrounding some underlying truth. It’s the reality. The average is the abstraction.³

This way of thinking has an unexpectedly specific history. In 1835, Belgian astronomer Adolphe Quetelet introduced the idea of l’homme moyen – the average man. Quetelet borrowed a method astronomers used to pin down a star’s true position: average enough imperfect measurements and the errors cancel out, leaving the real value behind. He then applied the same thinking to people, assuming that behind every person stood a »true« average man, and that each person’s distance from it was an error, rather than a fact about them.⁴ That was of course a mistake: there’s no single ideal human that real people are flawed copies of. The deviations weren’t errors of measurement; they were just the differences between actual human beings.

We still tend to repeat Quetelet’s mistake. The more interesting question about the McKinsey study isn’t the average, but what separates the two groups beneath it. Lu’s own answer points towards organisational design: the companies that benefit most from AI have used it to redesign their workflows from the ground up, while the others have simply introduced it into their existing structures.

Why should that make such a difference? One possible answer comes from a very different field: archaeology. In The Collapse of Complex Societies, archaeologist and anthropologist Joseph Tainter proposed that societies respond to problems by becoming more complex – specialising, adding rules, building new coordination structures. Each such addition can solve a real problem, but complexity comes with a maintenance cost, and the benefits of adding to it eventually diminish. In the end, a society can reach a point where its accumulated machinery consumes more resources than its solutions are worth.⁵

Tainter’s own example is the late Roman Empire: faced with invasion and a shrinking surplus, successive emperors responded the way the empire always had – with a larger army, more administration, expanded territory – until there was nothing left to conquer that was worth the cost of conquering it. The only options left were debasing the currency and taxing an exhausted population harder, just to keep paying for a complex apparatus that no longer paid for itself.⁵

Companies aren’t ancient civilisations, and inefficiency isn’t collapse. But the underlying problem is recognisable: every process exists for a reason, and as they accumulate, the problems they were built to solve may already have changed or disappeared.

Which brings us back to AI. A company with room to reorganise can use it to change how its work is being done. One already entangled in layers of procedure and coordination may simply add AI as yet another layer, acquiring a more powerful tool bolted onto a system that stays the same.

This is my own interpretive leap, not a conclusion in Lu’s figures: the survey doesn’t show that accumulated complexity explains the gap between the high-performing 6% and everyone else. But it reframes the question worth asking. Perhaps what separates them isn’t so much how well each company understood the technology as how much institutional room each had left to change before the technology arrived.

If that’s right, the companies struggling with AI may be failing at something considerably older – the same thing that eventually slows any institution that keeps solving problems by adding to itself without ever subtracting. Companies that fail to profit from AI may not fail at the technology. Some ran out of room before it even arrived.

References