How the Breakaway Pulls Away

I was up in Durham last week, guest lecturing on the MBA programme. Before I went, I had my trusty research agent run across my workspaces and pull together the latest data, stats, and case studies on AI capability and the gap between AI leaders and laggards; the kind of stats boards might be interested in as they think through what to ask their C-suite leaders, their own exposure, and what prudent action looks like in the era of AI and agentic AI. I wanted more recent numbers for my slides. Boy did I get them.

The picture that came back:

A small group of organisations is pulling away from the rest of the field on AI. Boston Consulting Group (BCG)'s Build for the Future 2025 work, sampling one thousand two hundred and fifty senior executives, finds that about 5% of large enterprises now qualify as what they call future-built. These organisations are delivering 1.7 times the revenue growth, 1.6 times the EBIT margin, and 3.6 times the three-year total shareholder return of their slower-moving peers. 60% remain in the laggard group, with substantial spend and very little or no return on their AI investments. McKinsey & Company's State of AI 2025 reaches the same place from a different angle: while 88% of organisations now use AI, only around 6% are getting genuine bottom-line value.

Different surveys, different methodology, identical conclusion - which is that a small subset of organisations are winning at AI, and the rest are not. And the gap between them is widening every quarter rather than closing.

It feels a bit like the Tour de France to me.

A breakaway forms in the first hour of a stage. Six or seven riders pull clear of the main pack. They take it in turns at the front, each shielding the others from the wind, dropping back to recover, swapping the lead every couple of minutes. The drafting saves them roughly a third of the energy they would burn riding alone. The peloton chases. It is much larger and on paper much stronger. But in the early kilometres the gap holds, and then it grows.

The riders in the breakaway are not materially better than the riders in the pack. They are not necessarily a lot stronger. But they are working together in a way the pack is not. Each turn at the front is shorter because there are six others to share it. Each rider's energy lasts longer because of the draft. Each kilometre is being ridden, on average, at lower effort by the leaders than by the chasers. Beyond a certain point the gap stops being tactical and becomes structural. By the time the pack realises the breakaway is going to stay clear, the maths has turned against them.

This is what is happening to organisations on AI, and most of the commentary I read misses it.

The default explanation for the value gap is that the leaders have better technology, or smarter people, or deeper pockets. None of that holds up. The technology is largely the same for everyone. The vendors are the same. The leaders are not, on the whole, employing dramatically smarter people. They are not even necessarily spending more in absolute terms.

What they are doing differently is, almost entirely, organisational. And it is in the part of the work that does not look like a procurement decision.

The way I think about it is an iceberg. The part above the waterline is everything the boardroom procures. The platforms. The vendor contracts. The pilots. The Copilot rollout, the ChatGPT licences, the bespoke build with whichever consultancy has the warmest relationship. Everything with a known price and a known shape, everything procurement teams know how to handle.

This is the visible portion of the value at stake - around 30%.

The real value though, the 70%, sits underneath. BCG put numbers on this a few years ago and they have held up; 10% of AI value comes from algorithms, 20% from technology, 70% from people and process. The vast majority of AI spending is pointed at the first thirty per cent. The bit that actually moves the dashboard is below the waterline, waiting for someone to fund it.

The plain reading of the data is that the difference between the leaders and the laggards is not which tools they bought. It is what they did, organisationally, with the tools they bought.

And the reason that work tends to go unfunded is that it does not look like a procurement decision. It does not have a vendor. It is slow. It is messy. It is mostly leadership work. So to make this concrete, having worked with these winners, guided companies on successful ai adoption and capability building, and sharing this methodology at scale, here is what that elusive recipe looks like up close:

Picture a farmer. He has spent thirty years swinging a scythe. His shoulders are built around the swing. His back knows the rhythm. His morning starts with the smell of grass and the small ache that means his body is ready. He is, in the proper sense of the word, an expert. The scythe is more or less an extension of him. He can do a day's work and feel it was good work.

One morning a combine harvester arrives at the gate.

The first question is whether it is the right combine harvester. Wrong machine for the wrong field and nothing else matters; the tool is wrong before the conversation begins. That is the procurement question. It is the part most boards are reasonable at, because there are vendors and reviews and references and a sensible-looking spreadsheet.

Assume, for the sake of the argument, that the combine harvester is well-chosen.

What happens next is the bit that's a bit harder to navigate. Because the combine harvester, however brilliant, sits on the doorstep doing nothing until the farmer is prepared to climb up into it. And the farmer is not climbing up into it.

He is, in fact, terrified of it. He is wondering whether it will kill him. He is wondering whether it will take his job. He is wondering whether his children, who have grown up watching him swing the scythe, will respect a man who pushes buttons in a glass cab. He is wondering whether his data, his fields, his thirty years of knowing exactly which patch needs cutting first in August and which in September, will be quietly extracted and used somewhere he does not control. He is wondering whether the very skills that have defined his working life are about to be made obsolete, and whether the village will still need him afterwards. He is, in short, doing what any reasonable person would do when confronted with a machine that promises to change their working life beyond recognition. He is hesitating.

If you do not address that hesitation, nothing else happens. The combine harvester rusts in the yard; the farmer goes back to the scythe.

The addressing of it is the leadership work. It is communication, repeated and clear and patient, about what the machine is for and what it is not. It is incentivisation, so the farmer knows that using the machine well is rewarded and not punished. It is strategy, so he understands where the farm is going and why. It is governance, so he knows the boundaries of what he is allowed to try, and policy, so he knows what is being done with his data and his work. None of this lives in the combine harvester. All of it lives in the leadership around the combine harvester.

Assume you have done that work, well enough. The farmer has at least agreed to consider the thing.

Now you have the next problem. Nobody really knows how to drive the combine harvester yet. The instruction manual is out of date by the time it is printed. There are no real trainers, because the technology has moved faster than the training. The honest position, if you are the farm owner, is that you have to admit to the farmer that you do not entirely know either, and that you and he and the other farmers are going to have to work it out together. Learning out loud. Sharing what works. Codifying what you find.

So what you can offer him, as a leader, is an environment. Not certainty; an environment. A safe-to-fail environment, where the blades are not yet engaged on the first three runs around the field. Where it is acceptable to look stupid for a fortnight. Where the question is not "have you cut the field" but "what have you learned today that we should know tomorrow". Where the metric is not square metres per hour, because nobody knows yet what the right number is. Where the farmer is allowed, encouraged, expected, to think bigger than he used to be allowed to think.

And then, if you have done all of that, one day the farmer comes back. He has done the whole field in under an hour. He is half laughing and half a little bit shaken. He cannot quite believe it.

This is the moment most leadership conversations end. We assume the value has been captured. We have what we wanted. Tick.

It is also the moment where the real value begins.

Because the question that follows is not "good, now do another field tomorrow at the same pace". The question is what is the farmer going to do with the day he no longer needs to spend on cutting the field. The honest answer is that he is going to do the things nobody had time for before. He is going to look at the soil and notice patterns he has been too busy to notice. He is going to teach the younger farmers. He is going to design next year's planting around what he has just learned. He is going to take on the field on the next farm because he has the capacity now. He is going to think.

That, in one farmer in one day, is what the seventy per cent looks like.

Now multiply that across thousands of farmers, ten thousand workflows, sustained for two years. That is roughly what the future-built organisations have been doing, and it is where the answer to the question we started with lies.

The seventy per cent compounds, in a way the procurement-led work simply does not.

BCG's analysis describes the mechanism precisely, and this is the bit my agent's research had not pulled together for me before in quite such a sharp way until the trainride up to Durham University Business School - The Durham MBA . Future-built organisations are three times as likely to operate centralised AI platforms, where each new use case strengthens the platform and the next deployment is faster. The second farmer learns from the first, the tenth from the second. They plan to upskill more than fifty per cent of their workforce in AI in 2025, compared with about twenty per cent at the laggard firms. Their population of farmers capable of climbing into the cab is widening every quarter. They are six times as likely to dedicate structured time for AI learning, so the work actually happens rather than being squeezed into the margins of someone's diary. And they reinvest their gains into the next round; their AI budgets in 2025 are running at more than twice the laggards', funded substantially by the value they captured last year.

Add those four together and you have a flywheel. More platform reuse. More capable people. More structured learning time. More margin to invest. More workflows in motion. More data on what works. Better next deployment. Each turn of the wheel makes the next turn faster.

This is what is going on in the breakaway. The leaders are not riding faster than the chasers. They have built a structure that lets every kilometre cost them less than it costs the pack. Once you see that, the widening gap stops being mysterious.

A note on security, because it is part of the same picture. Every new AI capability above the waterline is, in the same act of procurement, a new surface for attack. Every new agent you deploy is a new privilege boundary. Every workflow you hand to a model is a new point at which an adversary, or simply an accident, can divert what the workflow does. The organisations doing this work properly have started treating AI security as a continuous, dynamic measurement discipline rather than a one-off audit. More like the way a treasury team thinks about FX exposure than the way an IT team used to think about penetration testing. The shape of the security work mirrors the shape of the value work; a continuous, recalibrated picture of where capability has been built, where surfaces have been opened, and where the next adjustment is required.

What senior leaders can take from all this

First, the gap between the leaders and the rest of the field is widening because the work that creates the lead compounds. Each piece of organisational work the leaders do makes the next piece easier and the next piece after that easier still. That is not a small advantage. That is the structural advantage that decides which organisations end up in the breakaway and which spend the next three years working hard and getting nowhere.

Second, the work is available to everyone. The breakaway did not get there because the riders were faster, or because they had better bikes, or because the wind was kinder to them. They got there because they organised themselves to share the load, properly, before the pack did. Anyone can do this. The leaders in your sector are not gifted. They are organised.

The most useful question to put on your next strategy conversation is probably not "what is our AI strategy". It is "where is the seventy per cent work happening in our organisation, who is being explicitly resourced to do it, and what is the floor we have built underneath them". Those are questions the people you already have can answer.

The breakaway has pulled away. The good news, if you are in the chasing pack, is that the same work is available to you. It just has to be funded properly, sustained for long enough to start compounding, and treated as a leadership project rather than a procurement decision.