Why Individual Gains Don’t Automatically Become Organisational Gains

McKinsey's latest State of AI survey found that:

• 80% of respondents said AI had improved their individual productivity.

• Only 37% said it had contributed positively to their organisation's bottom line, a share essentially unchanged from the year before.

• And only around 6% attribute 5% or more of their EBIT to it.
  

Those figures point to something important: individual gains don't necessarily become organisational gains.

The individual gains aren't really in question as half of the same respondents said AI helps them make better decisions, and there is little reason to doubt that. Some of the gap will also be timing as productivity takes a while to find its way into a financial statement. But when individual gains keep rising and the organisational number sits relatively still for a year, it's worth asking what happens in between.

And one way of thinking about that is as a network problem.

An organisation isn't simply a collection of capable people. It's a network, and its capability depends on how effectively those people share knowledge, build on one another's thinking and contribute what they learn back into the system. Historically, those nodes were largely people working in close proximity. Today they're people working across larger teams, hybrid environments and different jurisdictions. Increasingly, AI systems and agents are becoming participants in that same network.

So the question becomes: does the network get stronger every time work happens?

If I use AI to solve a difficult problem today, and through that process I make a judgement, challenge an assumption or learn something new, does that stay with me and my private AI context? Or does it feed back into the organisation, so that the next person, and increasingly the next AI, has more to work with? Even without AI, the same question holds.

This is where compounding starts to become possible: when one person's thinking and learning becomes the starting point for someone else's, rather than something they have to work out again. But it doesn't happen by default. You have to design for it. And the part we probably design for least is the people.

If we're investing heavily in making the technology more capable without being equally deliberate about making our people more capable, I think we're missing a significant part of the opportunity.

One reason the technology gets so much attention is that its returns are easier to see. Cost saved, hours saved and throughput are tangible, and what is easiest to measure can quietly become what we optimise for.

Giving people time back is the obvious example. The assumption is that with more time, people will naturally move into higher-value work. But more productive doesn't necessarily mean more capable.

I recently wrote an article on Capacity and Capability, and it's worth reading if you want further understanding of the differenc.

A simple way to think about it is that capacity is the room to think. Capability is what you're actually able to do with that room.

We often say AI will take away drafting, research, analysis and synthesis, a lot of the more mechanical work, and free people up for higher-value work. I think that's directionally right. But there's a catch, in that doing the work was also part of how we learned the work.

Think of the first proposal you wrote that came back covered in comments, or the numbers that wouldn't reconcile until you understood where they actually came from. Judgement was built by wrestling with something: drafting it, getting it wrong, comparing alternatives, receiving feedback, seeing what worked.

A lot of expertise has historically come from repeated exposure to fairly foundational work, particularly early in a career. If AI removes large parts of that experience, we can't just assume the judgement it used to develop will appear anyway. We need to become more intentional about how we create that experience for people.

None of this is an argument for preserving low-value work to keep people busy.

The better question is where the wrestling with the work happens now.

If AI does the first draft, perhaps the human role shifts towards framing the problem properly, challenging what comes back, spotting weak assumptions, applying context and making the judgement. Someone early in their career stops producing the first version and instead starts owning what comes back.

In other words, move the thinking rather than remove the thinking.

But that still requires the capability to do so effectively, and that design therefore needs to be deliberate. If we simply remove work and give people more capacity, capability doesn't automatically follow. Just like any other skill development, organisations need to think about what experience, support and training people need, where feedback happens, where judgement gets exercised and how responsibility increases over time.

There's also another practical risk. AI doesn't only remove work. It can create enormous amounts of new output.

If the future human role becomes sitting at the end of the process approving an endless stream of AI-generated analysis, recommendations and content, I'm not sure we've designed a particularly good job, or a particularly good way to develop capability.

And where that review is done properly, it's now where a lot of the judgement lives: what someone challenged, what they rejected and why, what they knew that the AI didn't.

That is also the part most likely to disappear.

The output moves on, usually without any of the reasoning that shaped it, so we have to become much more intentional about how that learning travels beyond the person who did the work.

So the question isn't really what work AI can take away. It's what people should be getting better at as AI takes more of that work.

And the risk isn't that we change where in the process learning happens. The risk is that we remove it without deliberately creating a new place for it to happen.

And that new place can't sit with the individual alone, or we're back where we started.

As people get better at framing, challenging and judging, they need to be able to start from what others have already worked out, rather than from a blank page and their own private AI context. And what they work out in turn needs to feed back into the shared environment, so that the next person, and the next AI, can inherit it, build on it and leave the network stronger again.

That is how an individual gain starts to become an organisational one.