Florian Edouard warns that AI can turn nine months of consulting work into a three-day deck, but people do not read long documents. If another AI then summarizes and approves that output, AI ends up confirming itself. He says humans must know when their brain is needed in the process.
Florian Edouard notes vendors have promised to shrink MLR from six weeks to one day for years, yet cannot show customers who did it. He says the blocker is human: approvers ask who goes to jail if something goes wrong, so they demand near-perfect certainty before approving AI.
Florian Edouard says everyone calls data gold, yet people assume it belongs to someone else. He insists business and digital functions treat data as a company asset, alongside cash flow. Sanofi's move to Snowflake with one governed mesh and a responsible-AI framework made clean data the base for scaling.
Florian Edouard now hears 'AI agents' from many vendors and immediately asks what the backend does and whether it can plug into Sanofi's agent layer. He ignores widgets and shiny front ends; half the solutions disappear because there is little business value underneath. He wants specs and a live test.
Florian Edouard predicts rep-delivered marketing messages may have little impact in 24 months as physicians use LLMs. Old CRM mostly pushed message volume and monitored rep frequency; future CRM should help HCPs improve their practice. Natural-language interfaces may also replace customized screens and buttons.
Florian Edouard says his biggest AI worry is security and misuse, not the usual productivity debate. As models improve, scams, fraud, and attacks improve too, and small groups or governments can gain disproportionate power. Ethical frameworks help, but he is unsure how much control organizations really have.
Florian Edouard says working in Japan forced him to reconsider leadership. He arrived thinking he knew how to run a team, then saw a centuries-successful culture solve problems differently. He learned to watch and adapt, and now encourages others to take the same experience abroad.
Florian Edouard says 87% of AI pilots do not scale. Pilots succeed because teams carefully curate and maintain their data sets, but scaling exposes years of neglected structured and unstructured data. He argues organizations must fix the data layer before they expect enterprise-wide AI adoption.
Florian Edouard distinguishes giving people cloud or ChatGPT access from redesigning their work. Real transformation, he says, comes from sitting with business users and asking how an agent could take over parts of a workflow. Moving from typewriters to PowerPoint alone is just digitizing existing work.
Florian Edouard describes watching a strong organization he built slowly dismantle after he moved. People had joined for the vision and journey, and new leaders brought different ideas. He now says groups should be designed for three or four years, because business conditions change too fast for ten-year structures.
Florian Edouard says the fastest way for a vendor to destroy credibility is to claim they know the client's business better than the client. He accepts bold expertise, but says it must come with humility: bring your specific capability, not a claim to run pharma better.
Florian Edouard describes patients logging symptoms in ChatGPT for a year, then bringing a complete history to physicians who lack equivalent data. He expects patients to challenge diagnoses and asks whether pharma apps can beat the ChatGPT already on their phones; he doubts they can.