McKinsey published this year’s AI survey on 25 August. Eight in ten of the people in it say AI has made them personally more productive. The share of organisations who can point to AI in their bottom line sat at 37%, which is exactly where it sat a year ago.
Almost everyone is using it. The people using it feel measurably faster at their own work. The company accounts look the same.
That’s not an AI problem. That’s a strategy problem.
Why isn’t AI showing up in the results?
Because most businesses are using AI to do the same things faster. Drafting emails faster, writing social posts faster, processing data faster. The tool changes and the output doesn’t, because the thinking underneath it hasn’t changed either.
There’s a name for the pattern. McKinsey used it in April, in a piece called Where AI will create value, and where it won’t. They call it the Solow Paradox, after the economist Robert Solow, who wrote in a 1987 book review that you could see the computer age everywhere except in the productivity statistics. Companies spent heavily. Productivity barely moved. They computerised the old way of working instead of redesigning the work around what computers had made possible.
The number McKinsey open that article with is the one worth writing down. As of the end of 2025, almost nine out of ten companies had deployed AI in at least one business function, and 94% of survey respondents reported not seeing significant value from it.
Two things about that figure, because both matter. It comes from McKinsey’s own State of AI survey published on 5 November 2025, not from the April article that quotes it. And the 94% is 94% of everyone surveyed, not 94% of the nine in ten who deployed. Different denominator, smaller claim, still a striking one.
Has anything changed since then?
Barely, and that’s the part I’d sit with.
This year’s survey ran from 4 May to 8 June 2026, across 1,719 people in 97 countries. Nearly nine in ten still report regular AI use in at least one business function. Scaling across the whole enterprise climbed from 38% to 44%, so adoption did deepen. But the share of organisations McKinsey classes as high performers, meaning they attribute at least 5% of EBIT to AI and describe the impact as significant, stayed flat at about 6% for the second year running.
A year of deeper adoption, and the financial picture held still.
One number underneath it did move. Individual productivity improved for 80% of respondents, and half say AI helps them make better decisions. So the gain is real. It’s just sitting with the person doing the work rather than arriving anywhere the finance team can see it.
Why don’t productivity gains last?
Because your competitors get them too.
McKinsey are blunt about this. Productivity improvement is unlikely to expand profit pools or hand anyone a durable advantage, because competition erodes those gains and the benefit flows through to customers rather than to the businesses that did the work. What you save this year quietly becomes the baseline everyone is expected to hit next year.
Productivity resets the floor of an industry, not the ceiling.
They allow one exception and it’s worth knowing about. When productivity gains shift unit economics from variable cost to fixed cost at real scale, lower costs allow more volume, and more volume lowers costs again. That compounds into something defensible. For most small businesses it won’t apply, but it explains why a handful of companies pull away and stay away.
What does AI actually make possible?
The article sets out three waves. Productivity is the first and the shallowest. Differentiation is the second, meaning products, services and business models that weren’t feasible before. The third, and the one almost nobody is planning for, is what falling transaction costs do to the shape of an entire industry.
The electricity example is the clearest way in. When electricity first arrived in factories, most owners swapped the steam engine for an electric motor and left everything else exactly as it was. Same building, same line shafts, same workflow, slightly cleaner. The real change came later, when small motors let managers rearrange the machines around the work instead of around the power source. That’s when assembly lines became possible. That’s when mass production became possible.
Most businesses are at the motor-swap stage with AI, and that includes most of the AI use I see in marketing. Pasting a longer brief into ChatGPT is a motor swap. Rebuilding what you feed it, and cutting most of what you were feeding it, is the layout change. I wrote about that in why context beats prompts every time.
What does this mean for a small business in New Zealand?
More than you’d expect, and the reason sits in that third wave.
McKinsey run the argument through transaction-cost economics, which is Ronald Coase’s idea that industries take the shape they do because using a market is costly. Searching, comparing, negotiating, coordinating, switching. Where those costs run high, work gets pulled inside large companies or handed to intermediaries who manage the friction for a fee. Where those costs fall, markets fragment and specialise instead.
Their conclusion is worth reading twice. If AI agents handle search, negotiation, coordination and enforcement at close to zero marginal cost, the traditional advantages of scale weaken, and smaller specialised firms connected through AI-mediated networks could run on economics that used to belong only to the big players who own the whole chain.
Read that as a New Zealand business owner. Being small here has always carried a coordination penalty. That penalty is the thing under pressure.
There’s a catch in the same section, though, and it’s aimed straight at marketing. As buying decisions start moving through AI agents, competition shifts away from visibility and marketing spend and toward relevance and position within the systems those agents draw on. Discovery becomes continuous rather than occasional. Ranking starts to lean on structured product data rather than brand recall. Being findable by a person and being findable by an agent are becoming two different jobs.
So what should you do about it?
Change the question you’re asking. Not how can AI help me do this faster. What does AI make possible that wasn’t possible before?
For a NZ business that might be a completely different client onboarding model. It might be a level of service that used to require a much larger team. It might be a price point that never worked before, or something genuinely new for your market.
And before you buy anything sold to you as the answer, get clear on what the thing actually is. A great deal of what’s currently marketed as an agent is a workflow with better branding, and the two are priced very differently. The definition and a buyer’s test are in what is an AI agent, actually.
McKinsey’s closing line is the blunt version. AI is not a productivity revolution, it’s a competitive reset. In earlier technology shifts, the companies that mistook efficiency for advantage cut costs while others captured market share, and when it all settled the winners weren’t the fastest adopters. They were the ones who worked out earliest where the value was moving.
In a market the size of New Zealand, where reputation travels fast and the gap between businesses is usually narrower than it looks, the ones who work this out first won’t just save themselves time. They’ll change what they’re able to offer.
Which part of your business would look completely different if cost and complexity weren’t the constraints they used to be? If you want to think that through out loud, let’s talk.
Sources
- Montard, Diedrich and Catlin, Where AI will create value, and where it won’t, McKinsey Quarterly, 29 April 2026. Source of the Solow Paradox framing, the three waves, the electricity example, the productivity-erosion argument and the transaction-cost section.
- McKinsey, The state of AI in 2025: Agents, innovation, and transformation, 5 November 2025. Original source of the nine-in-ten deployment figure and the 94% of respondents figure, as of the end of 2025.
- Tinkoff, Van der Veken, Chui and Balakrishnan, The state of AI in 2026: On the road to ROI, 25 August 2026. Source of the 37% EBIT figure, the 6% high performers figure, the 44% scaling figure and the 80% individual productivity figure. Survey in field 4 May to 8 June 2026, 1,719 respondents across 97 nations.
- Robert Solow, We’d better watch out, New York Times Book Review, 12 July 1987. Origin of the Solow Paradox quip.
- Ronald Coase’s transaction-cost economics, as applied to AI agents in the McKinsey April 2026 article above.
