By Olivier Gomez (OG), CEO of IAC.ai · 16/08/2026 · First published in his newsletter on Substack
A travel brand built on human warmth just hit an 85 percent AI production rate.
The number most companies are quoted is closer to 5.
The gap is not the technology. It is how they built.
Club Med is about as human a business as you can find. French operator, founded in 1950, now part of Fosun, with roughly 2 billion euros in revenue and more than 20,000 employees across resorts on nearly every continent. The product is ambience, relationships, and hospitality. People pay to be looked after by other people. So when its leadership decided the company would become “AI first,” the obvious question was the one everyone keeps asking: how does a profoundly human company put AI at its core without losing the thing that makes it worth booking?
I sat down with Siddhartha Chatterjee, the Global Chief Data and AI Officer at Club Med, to get the answer. Not the technology answer. The organizational one. After twenty-five years of driving these transformations inside large enterprises, I am convinced the hard part was never the model. It is the operating system around it. What follows is the playbook he described, and why I think it is one of the more honest accounts of enterprise AI you will hear this year.
Success looks like discipline, not magic
The first thing worth noting is what Chatterjee did not do. He did not declare victory.
His own framing was deliberately humble: Club Med is on its way to being successful, not finished. That matters because most AI keynotes open with a trophy. This one opened with a caveat. And yet the number underneath the caveat is the one that should make every executive sit up.
Out of the AI use cases Club Med pushed through its pipeline, roughly 85 percent reached production. This is not a marketing figure invented for a podcast. Club Med’s own half-year results for 2025 report the same: twelve AI use cases piloted across the year with an 85 percent production rate, described internally as exceeding benchmarks.
Now hold that against the comparison Chatterjee drew. MIT research has been widely cited for a brutal statistic: that the vast majority of enterprise AI pilots, around 95 percent, never deliver measurable value. Gartner’s figures are kinder but still sobering. Pick whichever benchmark you trust. Club Med is operating several times above its capacity.
So the real question is not “what model did they use?” It is “what did they build around the model that lets nearly nine in ten ideas survive contact with the business.” That is the story.
The origin: a crisis that bought them time
The foundation was laid in the worst possible moment, which is exactly why it worked.
Chatterjee joined in 2022 with a mandate to lead a data transformation. COVID had emptied the resorts, and few live projects competed for attention. Most leaders treat a downturn as a reason to freeze. Club Med treated it as a window to fix the plumbing, rebuilding its data and AI infrastructure into something they could actually build on once travel returned.
The timing turned out to be uncanny. He finished the infrastructure work at the start of 2023, almost the exact moment generative AI broke into public consciousness. With a background in natural language processing stretching back well over a decade, he did not need convincing that the technology would reshape both how the company operated and how its customers behaved.
That is when the cultural pivot happened, and it is the part most companies skip. Club Med decided to stop being a fast follower. For a brand that had historically watched others adopt new technology first, the call was to become an early adopter, even a pioneer. They began experimenting early, connecting language models to their resort product data through retrieval-augmented generation, back when the available models still hallucinated badly, and the tooling barely existed.
Here is the lesson hiding in the failure. Those clumsy early iterations were not wasted. They forced the team to build a delivery methodology, a repeatable way to target the right problems and get use cases into production. The methodology is the asset. The early stumbles paid for it.
How you actually structure for scale
This is where most “AI strategy” decks fall apart, because they describe a vision and not an org chart. Chatterjee described both, and the structure is the most transferable part of the entire conversation.
The center is a data and AI team of roughly 40 to 45 people, organized into squads. The inspiration is the Spotify engineering model: small, lean, cross-functional units of four to five specialists each. Club Med runs six or seven of these squads, and each one can ship an AI product end-to-end, including the interface, the DevOps, and the security. Chatterjee called them Navy SEAL teams. The point is autonomy. A squad does not hand work across five departments to deliver something. It owns the whole thing.
Around the squads sit governance roles covering delivery, technical standards, data management, and data quality, so that the right problems get sourced, and the right products get built. That is the engine room.
Then comes the genuinely interesting move, the decentralization. Club Med created two roles that live out in the business, not in the tech team.
The first is the AI ambassador. There are around 150 of them worldwide, organized as a community, brought together this year for a seminar in Marbella. Their job is not just technical. They are trained to communicate, to animate, to create enthusiasm, and they work alongside HR to enable self-service. An ambassador helps colleagues write better prompts, build simple assistants, and stand up the lighter agents that can be handled locally.
The second role is the one Chatterjee only half jokingly called the sexiest job in the company: the process designer. A process designer is a domain expert from finance, HR, legal, commercial, sales, or marketing. Someone who already understands how their function actually works. Club Med then teaches them AI, agents, and lean management techniques like value stream mapping, so they can map a process end-to-end and redesign it with an AI-first mindset. There were around 25 of them at the time of our conversation, with a target of 50 by year’s end, coordinated centrally by a head of process excellence and AI product strategy who animates the whole community.
The division of labor is clean. If a redesign is heavy, end-to-end, requiring RPA, automation, and multiple agents, the central squads build it. If a piece of the process can be improved locally through self-service, the ambassadors enable it. That is how you get to reach without bloating the center. Add the IT and digital teams on top, and the working community sits at close to 500 people.
Which leads to a point worth saying plainly to anyone who thinks AI transformation is a side project for two clever engineers. It is not. At enterprise scale, it demands real human resources. Today, Club Med has around 17 processes in active redesign, with six or seven already deployed.
Ownership and accountability, designed in from the start
Here is the test of any AI program. Not what happens when it works. What happens when it breaks?
Chatterjee tackled this before he scaled anything, and he started in a place most teams treat as an afterthought: ethics. The logic was specific to Club Med. If a deeply human company is going to champion AI, the program has to align with human values, or it will feel like a contradiction to employees and customers alike.
What I respected was how he set it up. Advised by a Sorbonne professor with a background in chairing a national research ethics committee, he was told something counterintuitive. You are not legitimate to define ethical principles alone. So do not. Instead, create an ethics committee, put every key stakeholder in the room, and reach a collective consensus on which principles to apply and how to implement them. Club Med stood up that committee fast and now runs it quarterly. Existing frameworks on transparency, personal data, and confidentiality feed in, but the decisions are shared, not decreed.
Sitting alongside it is a technical committee bringing together cybersecurity, AI infrastructure, data infrastructure, and digital product leads. Together, they maintain synchronized roadmaps across the governance pillars: FinOps, security, AI safety, the operating model, and the AI platform, with observability and monitoring layered in for anything mission-critical.
Chatterjee called the golden rule what it is. Do not deploy AI until those pillars are rock solid. Skip the foundations, and you can produce pilots, but you will not get them into production, and when something fails, you will have nothing to fall back on. The moment that happens, credibility evaporates, and once it is gone, you cannot keep hundreds of people motivated behind the effort.
The accountability mechanism that bridged the early gap was almost low-tech. After the first couple of use cases failed, the team started writing memorandums of understanding with internal stakeholders, spelling out accountability, metrics, KPIs, and expected outcomes. Then, symbolically, they had stakeholders sign them through DocuSign. A signature on a document that says, out loud, we agree on what good looks like and who owns it. The full governance structure matured later. In 2025, with first successes behind them, Club Med made its formal pivot to AI first, reorganized the teams, and stood up the technical committee with named leaders accountable for each stream.
Outcomes, not technology projects
This is the part I care about most, because it is where I spend half my advisory conversations. The trap is doing technology for the sake of technology. Falling in love with the tool. Chatterjee’s antidotes were concrete.
First, protect creativity without letting it run the business. Every team member gets to spend around 20 percent of their time sandboxing, prototyping, and testing, with no obligation to report or show results. If something works, they pitch it. That release valve keeps people taking risks while the rest of the operation stays disciplined.
Second, the three non-negotiables. You need the right skill sets because nothing happens without skills and discipline. You need to work on the right pain points, meaning processes that genuinely require reinvention, not an endless parade of low-hanging fruit. And you need serious KPIs tied to what the business actually counts: top line growth, cost savings, employee satisfaction, and customer satisfaction. Be explicit about which one you are chasing.
He made a sharp observation here. A commercial unit has revenue, return rate, loyalty, margins, and numbers no one can argue with. Technology departments often lack that precision, which means people can stay busy without ever being held to a result. The fix is a commercial mindset inside the tech team. Technical KPIs are not enough. You have to connect them to the P&L and to the strategic metrics that leadership recognizes.
To enforce that, Club Med built the business case discipline into the methodology. It starts with an audit. Does the pain point even exist, or are you inventing a solution for a problem nobody has? Is it serious? Is it big enough to be worth the time? Only then do you define the business case and the KPI you are targeting.
The role that makes this real is one Chatterjee created and named the AI value tracking manager. This person builds the P&L of each initiative and links it to the company’s P&L, use case by use case. They also help shape the business case up front. From there, the team maps every opportunity on a value versus complexity grid. A case can be valuable and still be too complex, blocked by legacy systems, change resistance or competing priorities. Chase those, and you set yourself up to fail. The output is a prioritization engine constantly steering effort toward the points best optimized on both axes.
And then the discipline I wish more teams had the nerve to enforce: a time box. Club Med tries not to take on any use case that needs more than three months to reach a value showing pilot, and no more than six months to scale. Chatterjee was candid that the lessons came from scar tissue. They have run projects for three years, or eighteen months to partial delivery, undone by alignment problems, legacy debt, and data quality. That is precisely where their roughly 15 percent failure rate comes from. Short timelines force you toward the problems that actually matter.
Build or buy, hire or rent
The question I get asked most often is whether to staff internally or bring in outside help. Chatterjee’s answer was the one I gave too, and it comes down to one word: ownership.
The core intelligence and the core platforms have to be internal. This technology is too business-critical to outsource. But the skills are rare, so the practical model is a balance. Club Med has run with roughly half its workforce on these topics as external contractors, deliberately using those experts to retrain internal talent, then gradually increasing the internal ratio over time. The mindset is augmentation, not dependency. An internal team, augmented where the skill or experience gaps are, never the other way around. And whatever the mix, document everything, so that no one is indispensable.
On the technology itself, his framing was the cleanest I have heard. In AI, the moat is the data you put inside the model. When you own the data, the expertise and the business rules, build your agents yourself, because you can control and supervise them. Where high-quality data is expensive or slow to collect, and an outside company has already spent years amassing it, in domains like legal, contractual, or specialized economic knowledge, there is little point in burning resources to recreate it. Buy or rent instead.
He put it bluntly. The agentic engineering is not the hard part in 2026. Most teams can stand up the architecture with a little effort. The hard part is the data. Think of agents as digital collaborators and draw their org chart. Some are built in-house because they are part of the company DNA. Others you hire from outside, the same way you would hire a lawyer or a consultant, as long as the third party can offer the service level agreement you need to trust them.
In the end, it is a leadership posture
When I asked what advice he would give a leader who knows they have to move faster and build this capability properly, he tied it back to the external recognition Club Med has received for its AI work, and he gave three things.
One, strategy first. AI first is powerful, but it is not for everyone. It depends on your business model and your maturity. Understand your strengths and weaknesses, decide how forward-looking or defensive you want to be, and define your North Star before anything else.
Two, demonstrate value. The field is still a black box to most business leaders, so start small, with low investment, and put your energy into governance, change, and conversation. Once you launch an initiative, do not abandon it. Take it all the way to the end. Then sell it internally, because internal selling earns the enthusiasm and the proof that lets you scale. And when you scale, scale properly: every country, every market, every language, every targeted employee, supported by manuals, communication, and relentless follow-through.
Three, share externally. Take what you have learned, minus anything confidential, to forums and conferences. The feedback you get back sharpens what you are doing and teaches you what is working and what is not.
His final point is the one I would underline twice. AI first is, before anything else, a leadership topic. You can run every play above, but if the leaders, from the CEO down to the managers, lack the right posture, the transformation stalls. Club Med worked with HR to define that posture around four qualities: a strategic mindset that projects the future rather than getting stuck in the present, because a model that disappoints today may be transformed in three months; authenticity, the honesty that creates real enthusiasm; the ability to influence, using your network to find the real bottlenecks; and the one he values most, the mindset of a coach, permanently training, explaining and bringing the team along.
The takeaway
Strip away the brand and the resorts, and you are left with a transferable blueprint. Fix the foundations first. Build small autonomous teams that own end-to-end delivery and push capabilities out to domain experts who understand the work. Govern ethics and risk before you scale. Track value in the language of the P&L. Time-box ruthlessly. Own your core, rent the rest, and never forget the moat is the data.
Club Med did not win by buying the best technology. It won by building the best operating system around. That is the part you can copy, whatever business you are in.
More from OG: all articles · subscribe on Substack