The finding rests on an uncomfortable piece of arithmetic. Moore's law doubled computing power every twelve to eighteen months; AI today doubles roughly every six. A company growing 10 % a year takes seven years to double. “There is a huge gap between the speed at which companies grow and the speed at which AI grows.”

The mechanism explains why the gap stays invisible. The danger is not dying, it is not noticing: dinosaurs did not go extinct overnight, they declined for a long time without knowing they were declining. “Nobody feels like a dinosaur, until the day you discover your company is dead, or far from where it should have been.” Three excuses sustain the illusion. “We're not in that business”: every extinct species was a specialist. “We're running a pilot, it's moving”: a POC that never reaches production is nothing but a cost line. “We'll wait for things to settle”: later, it is no longer an advantage, it is catching up.

Then there is the number everyone brandishes for comfort, the MIT one: 95 % of generative-AI pilots fail. He reads it the other way round. “It is not an AI problem. It is a problem of how you apply this technology to your business.” McKinsey says the same thing differently: 62 % experiment, 23 % scale, 6 % see real value. Where others read a collective failure, he reads a window: the remaining 94 % are the opportunity. The report in this issue (p. 25) breaks down what the 6 % who succeed actually do.

Then comes the most useful question of the keynote: what do you actually own? Models, you rent. Prompts, you rent: the one from three years ago has nothing to do with today's. Interfaces, you rent. Three things you own: your data and the pipeline that cleans it; the way you evaluate results, which is your culture; and your judgement. “Two of those three are not IT matters. The worst thing you can do (sorry to the IT teams in the room) is let IT drive the AI conversation.”

The rest comes down to three requirements. Data and process quality: “where are your SOPs?”, a question that usually produces an awkward silence. Security, because an agent is an employee with access to your systems. And integration, where he dismisses the usual objection: “even an old AS/400, even a COBOL system can connect. The problem is not the legacy, it is the processes around the legacy.” He closes on a question he refuses to answer for you: if the modern company was built in an era when thinking was expensive, and intelligence suddenly becomes cheap, where do you set the line between what you automate and what you keep in human hands?

The mechanism runs through two physiological signals. The first is unremarkable until you see the numbers: posture. At fifteen degrees of tilt, your neck carries the equivalent of twelve kilos; at thirty, eighteen; at sixty, twenty-seven. The second is more serious: the more we outsource our thinking, the fewer neural connections we build, and the prefrontal cortex, like a muscle, shrinks from disuse. “We become more like machines, instead of being human.” This is where he meets the morning speakers: the debate about schools is not pedagogical, it is a debate about what we protect.

01 · Patterns → purpose. Machines follow patterns; humans follow a purpose. His fill-in-the-blank formula: “I am the [metaphor] who [action verb] [positive impact].” He quotes Simon Sinek, “I am the optimist who inspires people to do what inspires them.” What all these statements share: service to others.

02 · Logic → values. Three values maximum, made visible in daily words and acts. “Values are not declared, they are shown. The parents in this room know it: our children follow what we do, not what we say.” The exercise to find them: describe yourself as an object, then name three qualities of that object.

03 · Rules → intuition. His favourite axis, and the most unexpected at an AI conference. He distinguishes the cognitive mind from the somatic one, the body's. Three pieces of advice for recovering intuition: stop and listen; feel it without forcing it; grow it through practice. “Rarely, very rarely, does your intuition mislead you.” Why we don't hear it: ambient noise, and our habit of running it straight into analysis until it dissolves.

04 · Code → conscience. “Code executes, conscience humanises. The machines' job is to make it work. Ours is to make it right.” From this comes the three-question test on the previous page, probably the most directly usable tool of the day.

05 · Efficiency → empathy. Three words we confuse: sympathy, feeling for you; empathy, feeling with you; compassion, empathy plus action. “People remember you for how you made them feel, not for your efficiency.” Four moves: listen without judging, see from their perspective, name the emotion, act. With a nuance many managers would gain from hearing: most of the time, people are not waiting for you to act. They want to feel seen and heard.

And he closes on the three fundamental emotional needs: to be seen, to be heard, to be recognised. This is where the line that silenced the room arrives. These systems were designed, with help from neuroscientists, to agree with you. A validation bias, built in. The result: they make you feel seen, heard and recognised, and they are starting to replace our capacity to get that from one another. “I'm afraid my children feel more connected to ChatGPT than to me. That they open up more easily to a machine than to their parents. That's a third alarm bell, for me too.”

Why the ratio is counter-intuitive. The model is the thing you buy, the line you budget, the demo you show the board. The remaining 70 % is invisible on a quote: redesigning a process, training by role, changing what you measure, holding the uncomfortable conversations. Organisations therefore naturally fund what is visible. That is exactly the inversion BCG measures among laggards. And it is the statistical version of what the Caudan stage repeated all day: Marc Israel's rewiring, Jenny Korten's fish, Manish Bundhun's mindset. Seven speakers, without coordinating, all described the 70 %.

The external gap: you against your competitors. On Rogers' adoption curve, profitability concentrates among early adopters; past the late majority the gap becomes hard to close, because leads compound: every quarter of usage improves the data, the processes and the judgement of the competitor who started before you. Your exact position on the curve matters less than its direction: if your competitors are operationalising while you are piloting, the gap widens every quarter, quietly.

The internal gap: your teams against you. While you discuss AI in the executive committee, your employees are already using it, in their browser, ungoverned. The blind spot is dangerous because it looks like good news: “our teams are getting on with it”. But ungoverned individual use produces personal productivity and nothing else: no shared process, no accumulated data, no collective learning. And leadership credibility erodes: according to Gartner, only 8 % of managers are genuinely AI-competent. Banning makes it worse (shadow AI, p. 14). Letting it run scatters it. The only way out: govern the usage that already exists, channel it into team-level cases, and train the managers first.

The local picture reproduces the global one (McKinsey: 88 % adoption, 6 % impact), on an independent measure. 83 % of respondents say they are comfortable with AI, and 88 % use it. But usage stays personal: asked “what do you use AI for”, the single most-selected answer in the whole questionnaire is “I don't use it for my work yet”, at 44 %. Management reporting tells the same story: 52 % still track their indicators in Excel spreadsheets, and only 15 % have automated dashboards. The tools are in people's hands; the value stayed on the table.

The gap is all the more striking because these leaders know what needs fixing. Their main difficulties: recruiting and keeping the right people (38 %), moving too slowly against competitors (24 %), too much manual copy-paste (21 %), teams that collaborate badly (20 %). And the decisions they find hardest are precisely the ones better data would improve: strategic expansion (34 %), budget and forecasting (33 %), resource allocation (31 %). Half the market still makes them on a spreadsheet.

The intent is there, the method is missing. 87 % have considered AI as a solution; but 61 % “dug into it on their own”. Only 21 % brought in outside expertise, and 18 % actually trained their teams. Among those who have not adopted, the brakes are not scepticism: company maturity (51 %) and lack of support (45 %). This is a market asking to be shown the way, and telling you how: concrete practice (59 %), real examples from its own sector (34 %), the fundamentals for the whole team (48 %).