The 5% Club: how to tell if your AI rollout is building capability or just burning tokens

Everything.

And for the next three years, everything is exactly what it is worth.

Put a cello in my hands and I am holding a cello. Put the same cello in Yo-Yo Ma's hands and you will need to sit down.

Strap me into a Formula 1 car and I am a woman sitting in an expensive seat; even if I have adjusted the seat and mirrors to me, I am not an F1 driver set to grab first place.

The instrument, we see from the copious research, case studies and experience we now have with AI in business, was never the variable that mattered the most.

The hands though - the human hands - those turn out to be.

Almost every organisation on earth has now bought the instrument; licences are paid, logins issued, the lunch-and-learn delivered, tokens being measured, costed, found excessive in many cases (are people organising their fridges and planning their holidays on company tokens?).

But the reality after all this investment is stark: MIT's researchers join a slew of similar research that finds 95% of enterprise AI pilots to be delivering precisely ZERO measurable impact on the profit-and-loss account, while BCG finds the roughly 5 per cent of companies that are built for this are growing revenue 1.7 times faster than everyone else, with 3.6x the shareholder return, and pulling further away each quarter.

Same instruments, broadly. But in very different hands.

The question of what separates them, and how to cross from one group to the coveted other, is the whole subject of this piece. And I write it because it's an issue - THE issue, really - that every team, in every sector, in every industry seems to be grappling with right now. We've been doing this quite a few years now, and the pattern of who fits into the 95% failure rate vs. who fits into the coveted 5%, is clear.

Same instrument, opposite fortunes

The difference between AI access (that means licences to ChatGPT, Copilot, Claude, transcription tools, etc. issued), and capability (that means playing the cello, driving the F1 car, using the tools for something that creates something valuable) is wide, widening, and probably the most important distinction for leaders to understand right now.

In a classroom, access looks like assignments pasted into a chatbot, teachers marking work nobody did. Capability looks like a World Bank trial in Nigeria that delivered nearly two years of learning progress in six weeks with the same class of tool, guided by teachers. It also looks like my own daughter, who recently built herself an audiobook that weaves her favourite fiction series into the GCSE science curriculum so she can revise while she cycles; I told that story on video this week and you can watch it here.

One use of AI hollows out learning. The other deepens it, enhances it, accelerates it, on the learner's own terms.

In professional services, access looks like Deloitte's Australian arm last October, partially refunding a government report of around 440,000 Australian dollars after AI-invented citations and a fabricated court quote reached the final document. Weeks of remediation, a name in the press. Capability looks like an eight-person finance team I know well, who configured a Claude workspace that drafts their reports to a higher standard than before: built on their best previous templates, fed by live data, adjusted to each recipient's preferences. The hours that used to go into formatting now go into rooms with investors and clients, and the quality of their investor relations has visibly risen. Same class of tool. One firm budgeted for the tokens and issued the licences; the other thought about how this tool can help them do their jobs better and faster.

The pattern repeats at every scale. JPMorgan's contract system reviews in seconds agreements that once consumed some 360,000 lawyer-hours a year. Moderna's people built more than 750 tailored assistants in about two months, the legal team adopting fastest of all. And when IKEA's assistant began resolving nearly half of all customer enquiries, the owner did not thin out the call centre; it retrained 8,500 call-centre staff as remote interior-design advisers and built a sales channel now worth over a billion euros.

The people did not go. They were redirected in their roles to materially raise the bar on quality, speed, and service.

You can see - then - if you don't get on this AI capability thing, a competitor might.

We can all play Chopsticks, and we're proud of it

Sit any of us at a piano and we can pick out Chopsticks, and feel briefly, sincerely magnificent doing it. It is one of the most cheerful facts about being human that a small amount of skill feels like rather a lot from the inside; the psychologist Ola Svenson found back in 1981 that 93 per cent of drivers rate themselves above the median, which is mathematically impossible and completely understandable. I include myself in all of this. Everyone in AI is early, and every one of us has at some point mistaken saying hello to a chatbot for fluency. Chopsticks for Beethoven.

So when a leader tells me, warmly, "Oh, I'm a big AI power user," and it emerges that they mean tidying the odd email on a Friday, I hear Chopsticks, and I hear no shame in it whatsoever. Technology is moving faster than any working person with a full diary can reasonably track.

What I want to show them, and you, is what the concert-hall version of their own working day actually sounds like, because it is closer than most people think, and none of it requires being a techie, an expensive course, or a night shift as an AI researcher.

It sounds like sitting over coffee with an investor, fully present, listening to their goals and reservations, while a well-configured project space quietly turns those notes into ever more precise and helpful communication back to base. It sounds like twenty research agents working overnight so that everything you need to know before a client meeting arrives as an engaging fifteen-minute audio brief you listen to on the way there, key points pulled out because the system knows your client. That is what playing the violin well looks like in an ordinary job, and reaching it is a matter of coaching and practice, the same as it ever was for any craft. The dividends are significant, and this endeavour is worth pursuing.

Why I keep talking about orchestras

Because a business is a group, and the magic happens when the group plays together.

Having been knee-deep in these transformations for years, across public companies, government bodies and businesses large and small, I can tell you an AI rollout is never one thing.

It is individual skill, hard-won and personal, distinct in every role.

It is leadership, which in our client work has alone marked the difference between 30 per cent of a team using AI and 90.

It is middle management, the layer where value compounds and the layer almost everyone forgets; of 250 organisations examined in one study, only three specifically trained their middle managers on AI.

It is organisational scaffolding, culture, change, governance and technology, all moving in relation to each other.

Deloitte asked 3,235 senior leaders what most limits the value they get from AI, and the answer was none of the machinery: it was insufficient skills in the workforce, ranked above cost and above regulation.

That is a symphony's worth of parts. Get the strands right and you get the Mozart the studies keep promising, the multiples of productivity, the happier workforce, the higher-quality output; Goldman Sachs sizes the full score at almost seven trillion dollars of added global output over a decade, and the next three years or so, before this becomes table stakes, are the movement in which positions get taken. Let one strand drop, and you can be the firm writing the refund cheque and answering questions in the press.

And this is the heart of the misunderstanding I meet most often.

Learning to play the violin means coming to know the instrument intimately: how it responds, what it can carry, what your part asks of it in this particular piece at this particular time. It is nothing like what's happening at the moment - which is everyone turning on the radio and complaining about the music that's coming out. We each need to create our music with our instruments. Real capability is each person learning their own instrument, for their own chair, in the orchestra of their organisation.

The score is what sets the players free

There is a persistent belief that governance is the brake on all this, and I have come to believe the opposite. I have just finished Daniel Dobrygowski's book, Technology Governance: Build Trust in Digital Innovation, and I would put it in the hands of anyone who sits on a board, and any leader thinking about AI right now. Which is all of them. Dobrygowski, formerly head of governance and trust at the World Economic Forum and now teaching at Columbia, makes a point I see confirmed in every engagement: the right amount of governance propels innovation, and too little stalls it, because nobody plays boldly when they can't see the score. A score and a conductor are precisely what allow a section to improvise without the piece falling apart. People do their bravest work when they know where the lines are; take the lines away and we get a nervous silence, which is a very expensive sound in business.

The baton thrust into busy hands

Here is the tender spot, and I will say it with all the sympathy it deserves, because part of my job in this newsletter is to spot what's working in AI rollouts and what's not. In most organisations, the person conducting this is a capable, busy leader who already had a full-time job when the baton was pressed into their hands between meetings. They were given the instruments budget and the deadline, and not the depth of training that conducting actually takes. That is nobody's failing, but it explains a great deal of the 95 per cent. Nobody stands in front of the Berlin Philharmonic for the first time and conducts like Sir Simon Rattle, who spent sixteen years earning that podium. If we want a world-class orchestra, we bring in someone who knows the art and science of it, or we make concerted efforts to set aside time and budget to train our brilliant, busy leader to become a conductor.

Imagine the concert hall

So imagine, for a moment, a world in which your leadership team sets the score once: the strategy, the usage policy, the communications plan, the goals and guardrails and constraints. And that score does not sit in a drive; it diffuses into every individual's daily use of AI, so that each person's adoption is guided along the path the organisation has actually chosen.

Imagine every individual receiving private coaching on their particular instrument: a daily lesson tuned to them, to their real work, and to your rules, and a weekly mission that is a genuine task in the flow of their job. Hey James, you have a transcription tool; try running it in tomorrow's team meeting and sending everyone their follow-ups, so nobody leaves unsure what they owe. Hey Jane, someone across the organisation is building something remarkably close to yours; go and find each other - meet, talk, collaborate. Hey Joe - that thing you're doing could also be done like this, using AI. Try it, share the results with your team.

These nudges build private talent, but they also build capability the institution owns.

And imagine the person at the front no longer conducting deaf. They can see each player's level on one screen, hear the quality of the sound as it develops, watch return on investment, sentiment and tool use move week by week, and step aside after the performance with the player who lost time, kindly, to help them refine before the next one.

(We built Atheni.ai to do exactly this: a successful AI transformation in a box: The organisational score at the top, the Atheni coaching in every individual's flow of work, and Mission Control above it all: a window, at last, onto the thing leadership teams and boards have been funding blind. Across our programmes, clients average 91 per cent adoption inside 90 days, and 43 per cent of their people reach the top few per cent of most sophisticated AI users globally. Every programme has hit its 90-day targets, in regulated finance, in pharma, in manufacturing, in education, and it is why I now find myself saying this from stages as often as boardrooms, from London to the FT's stage in New York, in the IEEE's international work on AI and human wellbeing, and in the US from this autumn. The method is the moat.)

The question to tape to your boardroom table

We have the access; we might even be spending a fortune on tokens. But where, exactly, is our capability being built? It is a generous question, because for most organisations the honest answer today is "nowhere in particular," and the distance from there to a workforce that can truly play is closable, and measurable, in 90 days.

What is the difference between AI access and AI capability? Everything.

The instruments are bought and the hall is full.

The 5% Club is simply everyone who decided to learn to play, together, on purpose, and - allow me one final metaphor - the overture has already started.