At the head of MAIF’s technology strategy, Guillaume Rincé is driving a deep transformation of the mutualist group’s information system. Between in-house software development, a strong commitment to open source, and reflection on digital sovereignty, he advocates a responsible and controlled vision of digital technology.
In this interview, he discusses how MAIF blends innovation, technological independence, and mutualist values, from cloud management to the prudent use of generative artificial intelligence.
Dawn Liphardt – What is your scope of activity as MAIF’s CTO?
Guillaume Rincé – I have two main activities. First, I define the information system strategy for the entire group and its subsidiaries. Second, I am responsible for the group’s technological activities. We operate in a matrix structure with squads that group developers, engineers, business analysts, designers, architects, etc. And we have “delivery” activities organized in tribes, according to our vocabulary, corresponding to MAIF’s different business domains: for example the “Channels and Flows” tribe or the “IARD Claims” tribe.
I lead the technological domains and my colleague Sébastien Agard handles all the functional deliverables. Then we mix our teams in these tribes, formed of people from the group’s various professions to build the applications we provide.
Is MAIF a publisher of its own software?
Yes, we develop most of our applications in-house. We have recruited several hundred employees, including many developers, in recent years. We operate as a software publisher organized to produce our own solutions and implement them. This gives us full command of the chain, in terms of skills, resources, and processes, including design, which is key to improving the user experience.
In this publishing activity, do you rely heavily on open source?
Open Source is a natural approach for MAIF, in line with our purpose to work for the common good. Building commons and sharing them aligns completely with the group’s values. When I say “open source,” I’m not talking about container tech dressed up by a vendor with a closed subscription policy. I’m talking about real open source distributions, genuinely free.
We rely a lot on technologies through frameworks like React or databases like PostgreSQL.
We have about a dozen products available on our GitHub platform (http://maif.github.io), which others can integrate into their information systems. For example, we share an API management solution that is state-of-the-art, which we use at group scale. We actively maintain it. We have users in media, sales, and other domains, not only in France but around the world.
We also share technologies such as “Feature Flipping” to enable hot code activation, or AI algorithm explainability, and we actively contribute to open source projects, notably to maintain some critical components. We have people who invest in different technologies. They are often contributions to “quick fixes.” We enjoy supporting projects we use, especially those that are important for our information systems but are led by a small team.
Each year, we try to support 2 to 3 projects through monetary donations or by helping fund a library. The idea is to support those who create these useful components and from whom we benefit, by returning part of the savings we achieve through Open Source.
How is the identification of needs, development, and production of applications carried out?
The objective is for this to be a company-wide topic, not just the IT department. We must be able to weave this transformation into the business lines that interact with us. In our organization, several elements structure this process. The first is what we call the strategic portfolio of initiatives. The simple idea: we have a number of strategic directions. Very often, behind these directions lie information system topics, but not exclusively. Each direction is led by what we call initiative leaders who work with the tribe “business owners” to build the product backlog and the evolutions needed to realize the strategy.
Decisions are made each year across the different strategic portfolios. Then the tribes organize the execution and coordinate the actions. We hold three or four “synchs at scale” per year, where all the collectives realign. We base this on strong principles of agility and trust-based management to empower all squad members, regardless of their role, so that everyone contributes to the project. The leaders of each roadmap are responsible for delivering the investments, the “business owners” of the tribes are responsible for arranging their collectives, and the tribe leaders ensure deliveries and proper coordination between product squads.
How do you maintain your application portfolio?
Keeping technology up-to-date is something we introduced five years ago. We no longer want a heritage that lingers unmaintained, versions that are not current, or obsolete libraries or components on our platforms.
Each year, our portfolio must benefit from a solid level of maintenance: upgrades, security, patches…
“Today, it is alive and it probably has 10 to 15 years of lifecycle ahead. I don’t want to launch programs or projects, deliver a great product, and then let it gradually wither for ten years after a heavy initial investment. We continually improve the products, both technically and functionally, taking user feedback into account. Not revolutions, but evolutions that enhance the experience. Yet, we must make choices, because we cannot do everything, and that requires a lot of work within the collectives. It is a big part of our run activity.”
What is your policy on cloud?
MAIF has had a Public Digital Charter since 2016, in which we explicitly commit to protecting our customers’ data. All our choices flow from this commitment.
We have built two data centers where we host the “core reactor” and our client databases to maintain control of our data. It’s a strong investment, a foundation we want to keep within our walls.
When we use the cloud, it is mainly for interaction flows, to create high-performing digital journeys, mobile experiences, or to interact with partners. We build stateless applications, which means that data is not stored in the cloud, especially if it is not sovereign. They merely transit through it.
For example, when you use our mobile app, you may pass through the cloud; but only for the duration of your interaction.
What is your approach to technological sovereignty for the cloud?
Five years ago, we chose to work with Microsoft Azure, in a context where hyperscalers’ cloud offerings were essentially American. But today that is no longer enough. We are thinking, like other major European companies, about turning to other European players. We are in the evaluation phase, and I cannot yet say who we will work with.
Two years ago there was still no credible large-scale offer, not in hosting, but in terms of a software stack to combine services. We now have real European sovereign cloud actors facing the American hyperscalers.
What we want is to be able to do programmable cloud in all its complexity to benefit from a true richness of services. It’s not just a VM or a “big disk.” We know how to do that ourselves. The real question is to have advanced features to develop, orchestrate, and run our systems with fine control.
Today there are actors delivering technologies built by real European engineers, notably in France. That changes the game. We hope to integrate this capability by the end of the year, thus obtaining sovereign features in addition to what we already do. This is even more important with generative AI, which involves processing with capabilities we may not be able to host in our data centers, due to cost and rapid evolution.
To do generative AI, we will need cloud infrastructure, but always in environments where we can guarantee sovereignty, with a security level equivalent to that of our data centers. Equipping our infrastructure with this capability will allow us to implement generative AI much more confidently, while fully honoring our commitments. And that is essential.
The Cigref regularly denounces the inflation of cloud services and software costs. What is your view on this?
Regarding cloud costs, I’m fairly calm. American players are in strong competition outside the United States, notably in Europe, which ensures relatively stable pricing. For me, there is no major cost difference between cloud and a well-managed internal data center. It’s the only market, along with generative AI, where there is real competition.
However, what we are highly concerned about are the commercial policies of American software publishers. The list is long… We face commercial policies that make little sense, with price increases justified by marketing rhetoric but which in reality reflect a pure financial strategy. The aim? To create a lever to push customers to migrate to the cloud, with new subscriptions across various scopes. Behind it, the math is simple: I double, or even triple, my prices. Customers who haven’t invested much yet can leave easily. But 70% are locked in, because it would take five years to exit. Yet they have other priorities and are consumed by their projects, so they stay.
This shocks us deeply at MAIF: we are a mutual, what we pay comes directly from our members’ money.
For me, the real threat today for European companies is not so much technological sovereignty in terms of infrastructure, but rather dependence on publishers. We are clearly being squeezed. Sometimes, it’s almost racketeering, to be honest.
Furthermore, as a mutual company, we aim to support the European economy. Our European purchases circulate money within the European ecosystem. We seek to make responsible choices that develop our ecosystem’s economy and create wealth in Europe, which in fine benefits our customers and fellow citizens. Beyond geopolitical concerns, companies must also make responsible choices to support the economy.
So you will push further your internal publishing strategy?
Yes. It is a strategic choice to invest in people who have the skills or can acquire them. I prefer paying salaries and strengthening my teams, rather than paying monthly licenses with the vague promise that “it runs by itself.” We are not at all in the mindset of “cloud as magical service.”
The cloud is technology. And technology can fail. We do the same job with the same tools. They are neither better nor worse than us. I think this deserves to be demystified.
What we are trying to do is operate in the same way, because there is a lot to learn from their operational models. One question we ask ourselves is: “Are we professionalizing even more our internal publisher logic? With a team that builds the software, one that puts it into production, and one that operates it?” We could imagine going that far.
How do you approach generative AI? What use cases have you identified?
We try not to get swept up by the hype, even though the media attention is strong in both directions. We wanted to take the subject head-on, understand what it was, and, above all, see what it could change for our businesses. We began working over a year ago. At this stage, we have identified two priorities. Our objective was to experiment with them, then start putting them into production.
The first topic, not very original, is supporting business activity, notably access to knowledge in natural language. These use cases work fairly well, but they are not easy to implement if you want relevance. Because in every company, we have knowledge bases of varying quality, often with a lot of history that is rarely cleaned. If you put all that in without sorting, AI mixes everything and produces unreliable results.
So the main challenge of these early use cases is investment in data grooming. And when the data is clean, you get very good results. Today, we have deployed this across several business areas via a natural language assistant made available to users. We have two major production use cases: assistance to the claims management metrics and assistance to digital workplace users, including information around migrating to Windows 11.
In addition, we provide all developers who want it with Copilot licenses so they can code with AI and see what changes in daily work. What matters is to maintain a strong dialogue between what AI offers and the practices expected within the company.
Today, uses are mainly related to supporting certain métiers, such as soon the legal teams, where the stakes are high, with a lot of documentation and case law, thus with high added value. Ultimately, our objective is to give back time to the business lines so they can refocus on their real value proposition.
What are your points of attention?
There are many questions about energy consumption and the efficiency of models; this is a topic we are attentive to and which will gain significance for use cases that will find a lasting place in production.
The other major topic is change management. It’s exactly the same phenomenon as in the general public: have you managed to stop using your favorite search engine and begin with a generative AI first? Often we realize we are so conditioned that we start with our traditional search engine and then think we could still try AI. It’s the same inside the company: colleagues tend to go to their knowledge bases first. Adoption takes 6 to 12 months because you must deconstruct deeply rooted practices.
Products aren’t complex, but they aren’t easy to design. We have concluded that real ground-level support from field teams is essential.
Another topic: we also talk about technologies that are less European. That is a real concern because you must interact with clients in different forms, and that is shaped by culture. European culture, and French language even more, are not well represented in training data: 99% of the data used come from Anglo-Saxon cultures, with their political or ideological biases. We want to support and encourage initiatives to train models on European culture and European languages, especially French, so that letters reflect our cultural elements rather than simply being translations. We are very attentive to this.
