Health Data and AI: Doctolib's Risky Bet

The context, in plain terms

On July 8, 2026, Doctolib informed its users by email about the launch of an artificial intelligence research laboratory, scheduled to start in August 2026. This ambitious project aims to leverage users' demographic and health data, including that of their relatives linked to the account, to develop clinical AI models.

To do this, Doctolib relies on the CNIL's MR-004 methodology, which authorizes the right to object (opt-out) without prior explicit consent for public interest projects. Users have the option to object to the use of their data via an online form accessible from the July 8 email or directly on the Doctolib website, before August 2026. The collected data will be pseudonymized, ensuring it cannot directly identify individuals, and will be retained for a period of five years for a three-year project.

This initiative follows an announcement made in February 2026 by Doctolib regarding the creation of a clinical AI laboratory with an investment of 20 million euros for the year. However, the exact date of the internal decision preceding the email's dispatch is not public, and the actual start of the laboratory, later than the announcement, remains to be confirmed.

A Divided Panel, Between Adherence and Hostility

The Kapari test bench engaged a simulated panel of 62 voices to evaluate Doctolib's decision. Reactions are significantly distributed: 30 voices express support for the project, seeing it as a potential health advancement. Ten voices are uncertain, raising questions about the modalities or implications. Finally, 22 voices declare hostility towards the initiative, citing ethical or confidentiality concerns. This distribution reveals a contrasting landscape, where enthusiasm for innovation coexists with substantial reservations.

Points of Tension Revealed by the Test Bench

The signals identified by the test bench highlight specific points of friction. A journalist specializing in digital health, although affiliated with a group largely favorable to the decision, personally declares opposition, expressing a disagreement in principle. This prompts the decision-maker to prioritize listening to this profile to understand the exact nature of their principled disagreement and to identify arguments that could reassure them.

Another major signal is the absence of the 'Psychological Contract' voice within the panel. This empty chair, signifying the absence of an expected reaction based on the implicit relationship between an organization and its users, points to a potential blind spot that the decision-maker must address. The absence of this voice encourages Doctolib to anticipate reactions related to the breach of an unstated trust contract with users before proceeding, and to proactively communicate on the ethical guarantees and benefits of this project to restore or strengthen this contract.

Doubt About Execution, The Main Obstacle to Disarm

The Kapari engine identifies 'doubt about execution' as the dominant friction hindering project adoption. This hesitation does not necessarily question the principle of AI in healthcare, but rather how Doctolib intends to implement it, data security, or process transparency. This is a crucial turning point: dispelling this doubt could shift some hesitant voices towards adoption.

The verdict calculated by the Kapari engine proved stable over three independent runs, meaning that the reactions of the simulated panel and the conclusions drawn did not vary significantly during repetitions of the process. This stability of the verdict across multiple independent runs indicates that the identified friction points, particularly doubt about execution, are robust and not due to random fluctuation, offering a solid basis to Adjust the decision and target communication and adjustment efforts.

Adjusting the Decision: The Kapari Pathway

Given this range of reactions and identified fault lines, the Kapari verdict is 'Adjust,' with a 'Moderate' reception risk. This means the decision should not be rejected as is, but requires strategic modifications to optimize its reception and minimize opposition. Adjustments must target the friction points revealed by the test bench to transform doubt into trust.

To defuse the principled disagreement expressed by the digital health journalist, it is essential to clarify the project's real added value for users and society, beyond the mere technological aspect, by highlighting the concrete benefits of this AI. To address the blind spot revealed by the empty chair concerning the 'Psychological Contract,' Doctolib should explicitly address the issue of trust and reciprocity, communicating on ethical guarantees and direct or indirect benefits for users who share their data. Finally, the stability of the verdict over multiple independent runs confirms the importance of dispelling doubt about execution. Doctolib should precisely detail security, pseudonymization, and data governance measures to reassure about the robustness and transparency of the process, thereby converting the voices of doubt.

Questions about this case

Related cases