The Multiplier Myth.

The thing you are sitting in front of right now is a forty-year-old logistics network pretending to be a workspace.

Mouse, keyboard, screen, and underneath them a stack of standards (TCP/IP, browser engines, file formats) that the world has spent trillions of dollars building, layer on layer, since about 1985. All of it does one unglamorous job: get a messy idea out of one head, across an undersea cable, and into another.

We called that efficiency. It was a high-friction detour, and we took it for forty years because there was no other road. We slowed our thinking down to the speed of the interface.

That is what is about to change. And almost nobody in the boardroom is reading the change correctly.

This is where the boardroom miscalculation happens, and it happens almost universally.

I sit in a lot of meetings where senior leaders look at this shift through the narrowest lens there is: the spreadsheet. They are treating a cognitive multiplier as a margin-chopper. The first question they ask is "how many headcount hours can this save me in Q3?" rather than "what new market can my people now reach that they couldn't last year?"

That single question reorientation — multiplier vs. margin — is, in my read, one of the more expensive strategic mistakes on the table right now. And the uncomfortable thing is that the evidence is sitting in plain sight, in three different waves of technology, all telling the same story.

The ATM was supposed to eliminate bank tellers. Between the late 1980s and 2010, hundreds of thousands of ATMs were installed across the United States. The average number of tellers per branch did fall. But the banks that read the shift correctly did not pocket the savings. They responded to the lower per-branch overhead by expanding their physical branch footprint by something on the order of 40%, aggressively going after regional market share. Total teller employment went up over the same window — on the order of a hundred thousand more roles. The ATM did not kill the teller. It collapsed the unit cost of branch operations, and the winners used that compression to expand the surface area of their business. (One caveat: teller headcount began a slow decline through the 2010s as branches consolidated — the parable describes the ATM era, not a permanent verdict.)

The spreadsheet was supposed to eliminate bookkeepers. It did — several hundred thousand fewer routine clerical roles between 1990 and 2015. But during the same window, financial managers, analysts, and accountants who used the same tool to do something larger grew at roughly 3% annually. The financial industry grew anyway: the spreadsheet expanded what counted as financial work, and the market for strategic financial analysis followed.

The current AI cycle is the same dynamic on faster compounding. The most useful data point here is from the PwC 2026 AI Performance Study: roughly three-quarters of the total economic value being generated by AI right now is being captured by something like a fifth of organisations. The leaders in that cohort are markedly more likely than their peers to use AI to reinvent their core business model, to redesign operational workflows, and to increase the volume of decisions executed autonomously inside a real governance frame. They used the tool to change what the business does.

The Gartner data should worry everyone running the defensive playbook. When an AI deployment is targeted at individual task efficiency — the "everyday AI" frame — the measured productivity leakage tends to run somewhere from a quarter to, in the loosest cases, well over half. Translation: most of the time savings get eaten by administrative friction before they hit the P&L. CFOs in those deployments report headcount reductions in the low single digits, if any. The savings story largely doesn't show up.

The risk story, on the other hand, does show up. A tribunal found Air Canada liable for negligent misrepresentation after its customer-service chatbot invented a retroactive bereavement-refund path that its actual policy didn't offer. DPD disabled the AI part of its chatbot when screenshots of it swearing at a customer reached over a million views. The Commonwealth Bank of Australia reversed 45 customer-service layoffs because the automation was producing systemic operational failures. Across all three, the pattern is the same: pure cost-cutting AI deployments produce small savings, big tail risks, and zero structural advantage.

The pattern that works is structurally different. McKinsey tracked 80 global commercial banks from 2018 to 2022. Digital leaders — the ones treating digital infrastructure as a growth platform rather than a cost-saving project — delivered 8.1% annual total shareholder return against 4.9% for digital laggards. They grew retail revenue at +0.8% annually while laggards lost −1.4%. They expanded their share of digital sales from 40% to 70%; laggards moved from 8% to 17%. Same technology. Same five years. Completely different outcomes — entirely a function of whether leadership read the shift as a margin-chop or a market expansion.

So the job of senior leadership in 2026, as I see it: stop counting the hours an agent saves, and start counting the markets your team can now reach because the cognitive bandwidth constraint has finally moved. The brief becomes: what would we go after if the limiting factor was not how many things our best people could hold in their heads at the same time? Because that limit is no longer the limit.

Treating that shift as an OpEx exercise is like treating the automobile as a way to save 10% on horse feed. It is technically correct, and it misses the entire point.