The MAIF, a major French insurance company, is working to integrate artificial intelligence into its operations through partnerships with three tech firms: Mistral AI, Scaleway, and Diabolocom. The goal is to develop a sustainable AI strategy that maintains internal control over data, rules, and skills. These collaborations are part of a long-term plan to move from experimental AI projects to a fully integrated system. Scaleway provides computing infrastructure, Mistral AI supplies AI models and support for generating content, and Diabolocom offers a platform to manage interactions with customers. Since 2017, the MAIF has been building AI capabilities with tools like Mélusine, an internal system for handling customer emails, and ZELROS, which was first tested in 2017 and later used to help insurance advisors create recommendations, especially around electric vehicle coverage. In 2022, the company created a dedicated Data and AI department, which grew to 240 employees by 2024. This department focuses on aligning business needs with technical and legal requirements before rolling out AI assistants company-wide. The MAIF’s AI tools are used for tasks such as categorizing customer requests, generating recommendations, searching documents, creating reports, and assisting with writing. A major challenge is ensuring these tools work cohesively—using the same version of contractual rules and maintaining consistent access rights. The company plans to implement a hybrid IT system by 2025, combining internal data centers, Clever Cloud, Azure, and containerized applications on OpenShift. The core system, still partly built in Cobol, is being modernized to use Java and microservices. The MAIF is also focusing on internal skill development, aiming to increase its internal expertise in AI from 50% to 75%. The company wants to keep control over data, interfaces, and validation processes, turning these into proprietary assets. Every AI contract includes specifications about what remains under internal control, such as managing document collections, handling user identities, tracking activity logs, and ensuring the ability to replace AI models if needed. The MAIF is working on three main generative AI projects: researching internal knowledge, generating call reports, and helping respond to customer complaints. By 2025, the company expects to have around 50 AI experiments and 10,000 employees with access to Microsoft Copilot. Document search is a key test, combining knowledge base queries with response generation using the RAG (Retrieval-Augmented Generation) method. Managing historical data is important for governance, compliance, and building trust within teams. The introduction of Diabolocom is part of a larger modernization effort, addressing challenges related to the lifecycle of service providers. The platform aims to maintain consistency across customer contact channels, help qualify customer needs, assist advisors, and provide a better understanding of past interactions. This involves multiple departments, including marketing, customer relations, and IT. The MAIF sees this as a transformation journey, with metrics covering first-contact resolution, complaint management, and time spent on complex cases. Sovereignty in technology is a key factor in the MAIF’s purchasing decisions, alongside performance, security, and reliability. Choosing Scaleway supports the company’s push for a European cloud platform. The MAIF also considers the difficulty of switching vendors or bringing services in-house as a critical factor in outsourcing decisions, in line with the DORA regulations on IT service resilience. The MAIF has reached an agreement with its six trade unions to avoid layoffs due to AI implementation. The agreement includes training programs, reinvesting efficiency gains into customer service, and establishing an AI commission linked to the CSE (Works Council). Human validation is emphasized, requiring clear definitions of what needs to be checked, access to necessary data, time for review, and follow-up on corrections. The MAIF’s AI strategy is centered on aligning AI with business rules, historical documentation, obligations, and internal teams. For companies aiming to scale, AI becomes strategic when treated as a full operational transformation rather than a series of tools. The next step is to measure AI’s long-term impact, linking investments to service quality, processing speed, retained skills, and real gains.