Mastery and sovereignty : Docaposte’s strategic choices in generative AI
In this interview, Olivier Senot (Group Innovation Director) and Stéphane Ingrassia (Strategic Marketing Director) share their vision : AI as a driver of productivity, but above all as a lever for transforming roles, use cases and organisations.
What, in your view, are the most structuring challenges of generative AI for organisations ?
Generative AI, which represents one of the three main branches of Artificial Intelligence, raises challenges that go far beyond the technology itself.
It is about transformation, adoption, understanding and, above all, integration into a broader strategic vision. It also involves value based choices regarding how this technology is deployed, balancing benefits and risks, particularly with respect to human impact.
It is a powerful productivity lever. However, as with any disruptive technology, organisational awareness often lags behind technological progress, which continues to accelerate with the emergence of new capabilities.
At its core, the real challenge lies in organisational transformation. Generative AI cannot be reduced to a simple technology project, it represents a deep transformation initiative. This is all the more true as the technology itself is relatively accessible to understand, develop and deploy, whether through foundation models or intelligent agents.
What is, in your opinion, the main difficulty in adopting AI within organisations ?
The technical aspect represents only a small fraction of the challenge. The real difficulties lie in human and organisational transformation, which account for the vast majority.
There is also a strong need for training : AI is not a magic tool, and its usage must be understood and managed responsibly by users. The objective is always to ensure responsible and informed use of AI.
To support the upskilling of teams, training and awareness sessions have been organised around the AI Act, as well as broader regulatory developments related to data and AI.
The AI projects deployed, both internally and externally, address concrete business needs. All solutions are already compliant with the AI Act, as Docaposte has committed to a proactive approach, ensuring early compliance, notably through the implementation of human oversight of algorithms.
In addition, each project involving an AI system must undergo a "trusted AI" process. This involves a detailed assessment of the project against ethical, regulatory and technical requirements, using tools that are continuously updated in line with evolving regulations and ethical guidelines. This process is essential to raise awareness among project managers, data managers, data scientists and business teams regarding regulatory and ethical challenges.
How do you approach concrete AI use cases at Docaposte ?
To support this transformation, around forty generative AI projects have been launched internally.
These initiatives are structured around two main approaches : assistance tools and intelligent agents.
For example, Docaposte has developed Dalvia Santé, an agent designed for hospital physicians, currently being deployed across several university hospitals.
What does Dalvia Santé specifically do ?
Dalvia Santé is capable of automatically reading and summarising a patient’s entire medical record, which typically consists of around 80 pages, in order to provide physicians with a clear, structured and immediately actionable overview during the first consultation following admission.
This agent enables healthcare professionals to save valuable time by automating the synthesis of medical data from electronic health records, while ensuring completeness, clarity and reliability, without hallucinations in the information provided.
This ethical solution has been supervised by healthcare professionals by design and continues to be monitored on an ongoing basis. It relies on sovereign technological components and secure hosting aligned with SecNumCloud standards.
The objective is not to replace physicians, but to allow them to focus on high value medical tasks. Dalvia is a strong illustration of ethical AI serving all stakeholders.
Are there other use cases currently being developed ?
Yes, several projects are underway. An internal monitoring agent has been developed, capable of analysing reliable information sources on a daily basis and producing relevant summaries.
Another initiative focuses on automated communication, where an agent generates LinkedIn and X posts, including visuals, aligned with the tone and communication style of an executive. Human validation is, of course, systematically integrated before publication.
Are you observing tangible results ?
Absolutely. With Dalvia Santé, processing time has been reduced by 80 %. Regarding the SEO agent, publication frequency has increased by 1.5x, without reducing headcount. This clearly reflects augmented productivity, rather than human substitution.
How do you prioritise all these AI projects ?
A comprehensive mapping of all AI systems is conducted, as required by the AI Act, to ensure appropriate measures are in place to address regulatory requirements. An internal unit, named Castor, has been established to prioritise projects based on two key criteria : business impact and productivity gains.
From a technical standpoint, what is your industrialisation strategy ?
As early as March 2023, the decision was made to use GPT 4 within a private and secure environment hosted in France. Today, an internal GPU cluster has been deployed within Docaposte’s private cloud.
The next step is to transition towards open source LLMs to strengthen autonomy, security and explainability. This approach will also enable higher levels of security, particularly for sensitive documents, while ensuring full data sovereignty.
Some clients are concerned about the proliferation of AI agents. How do you address these challenges related to technological structuring ?
This is indeed a complex topic, given the wide range of solutions available on the market. Structuring usage is essential.
This is why Docaposte has chosen to develop its own agent orchestrator, fully agnostic to LLMs. This approach ensures strategic independence, allowing the company to offer sovereign agents and orchestration capabilities to its clients.
This choice implies accepting a slight delay in terms of features, but it is fully aligned with Docaposte’s positioning, focused on sovereignty and technological control.
Another advantage of this orchestrator is its ability to leverage the model best suited to each use case, and no more. For example, if a small language model is sufficient, there is no need to rely on a large language model, which is more resource intensive and costly from an environmental perspective.
How do you position your AI projects, and which sectors do you prioritise ?
AI initiatives are structured around two main pillars : client facing and internal.
On the client side, the focus is on developing reusable use cases across sectors such as finance, healthcare and the public sector, domains that are particularly sensitive to data protection and extraterritoriality challenges.
Projects such as Dalvia Santé are designed to be scalable across all public hospitals.
Internally, the objective is twofold : improving business processes, particularly in monitoring and market analysis, and optimising back office operations to increase productivity at a national scale.
These priorities help structure initiatives and ensure alignment with concrete, sector specific needs.
In your view, what are the next challenges to address ?
One of the main challenges is scaling, supported by a clear and unified governance framework, currently being implemented.
It will also be necessary to pool efforts across other entities within Groupe La Poste, while ensuring that AI is truly embodied within the organisation : visible, understood and actively used.
This will also involve more engaging formats, such as AI driven avatars or podcasts, which are already being tested.
To learn more, visit the website of Docaposte
To explore these topics further and identify the impacts and opportunities relevant to your organisation, contact our partners and experts : Jean-Charles Ferreri (Senior Partner and Corporate and Efficiency Expert) and Simon Georges-Kot (Partner and AI Expert)
