John McIntyre
Founder, HotPMO
George Sivulka, who runs the AI company Hebbia, opened an essay this spring posing a simple question:
AI just made every individual ten times more productive. No company became ten times more valuable. Where did the productivity go?
It is a fair challenge. Individuals who are using AI insist that it is making them more productive. PMO Analysts speak of the time savings on meeting minutes, creating reports etc. Project managers similarly celebrate the ease at which they can now create project documentations and reports. Data appear to back this up: The Work AI Index 2026, a survey of 6,000 digital workers across the UK, US and Australia from Glean’s Work AI Institute, found that 87% of digital workers now use AI at work and 75% say it makes them more productive. Ask how many can point to their organisation performing significantly better as a result (the survey’s own wording) and the number falls to 13%.
Business leaders tell the same story. KPMG’s Global AI Pulse for Q2 2026, which asked 2,145 senior leaders across 20 countries, found 76% claiming AI delivers meaningful business value, up twelve points in a single quarter. But ask who has established, measurable ROI and the numbers collapse to a small fraction of that number.
Analysts and Project Managers feel it. Leaders claim it. But the P&L can’t find it. We have seen this before, and last time it took the best part of thirty years to sort out.
The mills got new motors. The owners kept the factory.
1900: electricity reaches the factory floor. What happens next?
By the early 1900s, mills like these were the highest concentration of advanced industrial technology on the planet: steam-driven, thunderous, running around the clock. They all ran on the same architecture. One enormous engine driving hundreds of machines through overhead shafts and leather belts, and the whole building, floors, layout, jobs, was designed around getting rotational power from that single source to everywhere it was needed.
Electricity reached the mill towns from around 1900, and what happened next is one of the best documented puzzles in economic history. The economist Paul David wrote it up in 1990 piece “The Dynamo and the Computer.”
Factory owners electrified fast, within a decade of the technology arriving. Most did the obvious thing: pulled out the steam engine, bolted an electric motor into exactly the same spot, and changed nothing else. Same buildings, same shafts, same jobs. For nearly thirty years, productivity barely moved. The payoff arrived only in the 1920s, when a new generation redesigned the factory itself: a motor in every machine, layouts arranged for the flow of work, different jobs entirely. Manufacturing productivity surged at over five per cent a year for a decade once that happened.
Two ways to lose an industrial revolution, then. (1) Bolt the new technology onto the old factory, and lose thirty years finding out that doesn’t work. Or (2) defend the old design so completely you barely bolt anything on at all, and by the time a leaner competitor rebuilds around the new technology, there’s no version of you left to catch up. Both are failures of the organisation, not the technology.
Why the hours vanish
If individuals really are faster, the missing value has to be leaking somewhere, and the Work AI Index found the leak. Workers report saving around eleven hours a week through AI. They also report spending 6.4 hours a week on what the report calls botsitting: feeding AI missing context, supervising its output, debugging its mistakes and cleaning up after it. Nobody budgets those hours, so they quietly eat the saving. I’ve written more about botsitting, and its uglier sibling botvomiting, in a separate piece.
The rest of the leak is organisational. A systematic review in the Project Management Journal this year (Huzooree, 2026) concluded that GenAI adoption is an organisational design problem rather than a technology rollout, and that value depends on governance, processes and individual practice moving together. Deploy at the individual level only, which is what a Copilot licence for everyone amounts to, and you get what the paper calls sociotechnical drift: confident individuals, unchanged processes, absent governance. That drift is Sivulka’s missing 10x, wearing academic dress.
What the organisations seeing returns do differently
Two findings in the KPMG data are worth highlighting. (1) Organisations with clearly defined accountability for AI outcomes report established ROI at three times the rate of those without (14% against 4%). And (2) Organisations with full visibility of their AI operating costs report it at five times the rate (15% against 3%).
Contrast those figures with the fact that: 75% of CEOs say they own AI as a strategic priority, yet only 24% of organisations can name the CEO or executive committee as accountable for decisions made using AI. Ownership is a press release or words on a page, unless it is backed by organisational design.
If you want to evidence your own AI ROI rather than just cite KPMG’s, it’s the same two ingredients at PMO scale:
Ensure you have a named owner for every AI-touched process, licence and token spend tracked against the hours it actually replaces, and a before-and-after baseline, hours saved, work shipped unverified, actions closed on time, on whichever process you start with. A one page report that you can share with your sponsor and the business. That’s tangible evidence that can drive real decisions – something not many organisations are doing yet with respect to AI investment initiatives.
If this sounds familiar, that’s because it should. We are simply talking about accountability structures, cost visibility, process redesign, benefits tracking. None of it is a technology capability. All of it is what a decent PMO does every day. This is why I think the AI era makes the PMO more important rather than less. Somebody has to redesign the factory floor, and the vendors only sell motors.
The 1920s: the factory floor redesigned around the new technology, and productivity finally moves.
The practical move for the PMO finding their way with AI is firstly to figure out which of the two mill failures your organisation is currently rehearsing. If AI spend is rising while your processes stay untouched, you’re the 1905 factory with a shiny new motor and the same old shafts. Whereas, if you’re waiting for the technology to settle down before acting at all, you’re the other failure, and history has a clear view on how that one ends too.
If your PMO is working out where it stands, and what redesigning the floor would actually involve, that’s the work we do with clients every week. Get in touch and we’ll have a conversation.
Sources
- Sivulka, G. (2026). “Institutional AI vs Individual AI”, 12 March 2026.
- Work AI Index 2026, Glean Work AI Institute.
- KPMG Global AI Pulse, Q2 2026.
- Huzooree (2026), Project Management Journal, DOI 10.1177/87569728261447861.
- David, P.A. (1990), “The Dynamo and the Computer”, American Economic Review 80(2).
- Feature image:  – Plate from More Pictures of British History, London, 1914, p.61. Publisher: London. Adam and Charles Black. 1914 (Public Domain)
