What a pharmaceutical company just demonstrated on stage at VivaTech says more about an organization’s true level of maturity when it comes to AI than any press release from the company ever could.
On the VivaTech stage, a man pulls out his phone, opens an app called Concierge, and improvises. He plays a sales rep from Sanofi who is about to meet with a doctor, interrupts her three times, changes his approach, and asks for an email instead of a verbal response. The assistant remains unfazed: he already knows the patient’s profile, the doctor’s history, and the likely objection. Nothing about the scene is spectacular. That is precisely what makes it so instructive.
Because what the demonstration doesn’t show—and what its two protagonists, Emmanuel Frenera, Sanofi’s chief digital officer, and Love Kotari, head of AI at Snowflake, eventually revealed—is more interesting than the agent itself: it took five years before they were allowed to get to this point.
Infrastructure Before Action
The luxury industry is talking a lot about artificial intelligence right now. Pilot projects, partnerships, “digital transformation”: the terminology is spreading faster than its actual implementation, and that is precisely the subject of *The Silent Divide*—this gap, rarely acknowledged in public, between a fashion house’s rhetoric about its technological modernity and the actual state of its data.
Sanofi has nothing to sell to the luxury sector and does not target it. That is precisely what makes its story useful as a counterexample: the company has no reason to embellish its narrative. Frenera puts it bluntly: five years ago, the group’s data was stored in what he calls “data dumps”—fragmented repositories with no common governance. The decision was to consolidate everything onto a single platform before even considering anything more sophisticated. The first result of this investment, after several years of work, was not an agent or even a predictive model. It was a dashboard. A comprehensive view—finally possible—of data that had never been accessible that way before.
It wasn’t until 2025—after that foundation was laid—that Sanofi and Snowflake began building what is now Concierge. The agent didn’t come first. It came last—the reward for the behind-the-scenes work that no one cares about at conferences.
What the agent must keep quiet about
The temptation for a fashion house is to tell the story in reverse: to start with the feature that makes an impression and let the rest be left to the imagination. The risk isn’t just narrative. Love Kotari emphasizes a specific technical point that deserves to be highlighted.
Details. Each agent deployed at Sanofi is assigned a unique identity, a defined role, and an audited scope of access—distinct from that of any other agent or department. During the demonstration, Kotari points out that invoking the same assistant from his own account would produce a different response, since he does not have the same privileges. Governance is not something that can be negotiated after the fact: it is a prerequisite for access to the agent, not an additional layer on top of it.
This discipline comes at a cost that no House has any interest in downplaying if it decides to embrace it: transforming an entire team of “people who interrogate systems” into “people who supervise autonomous agents,” in Kotari’s words, first requires knowing who can see what—and being able to prove it. For an industry as strict about protecting its intangible assets as the luxury sector—recipes, techniques, customer files, design archives—the stakes should be clear without needing further explanation.
The number that doesn’t flatter anyone
The pharmaceutical industry operates with a disadvantage that the luxury sector does not face: Frenera points out that it takes ten to twelve years to get a molecule from the lab to the patient, with a 90% failure rate. It is this constraint—not a passing trend—that has driven Sanofi toward simulation and governed data—a matter of economic survival rather than image. The luxury sector, on the other hand, does not face this inherent pressure. It can afford to announce a pilot project without ever publishing the results. This is undoubtedly why strategic silence is more comfortable there than elsewhere: nothing compels a House to say whether its product is working—or even if it exists at all.
The only figure Kotari cites regarding Concierge’s actual use—a roughly 40% reduction in workload for field sales reps—is meaningful only because it’s tied to a specific tool that’s in production and was demonstrated live in front of an audience. Would a luxury brand announce a comparable figure under the same conditions—on stage—with the risk that the tool might malfunction in front of an audience? The question remains open, and that is precisely what distinguishes a demonstration from a statement of intent.
A lesson no one wants to hear
Frenera wraps up the discussion with a remark that should concern anyone who thinks the issue of data is settled once and for all: “AI-ready” data will never be a done deal, he says, but rather a “never-ending pursuit”—data will continue to exist in unconsolidated Excel files, regardless of how mature the organization is. This is not an admission of failure. It is a warning against the idea that an audit or a platform alone would be enough to close the book on the matter.
For a company that wants to follow a similar path, the question is therefore not whether it has “AI,” but whether it would be willing to spend five years organizing its data before it can boast about it. How many companies would publish such a timeline? How many, on the other hand, would prefer to skip straight to the demonstration—and leave people to guess what, if anything, lies behind the screen?
Cette publication est également disponible en :
