During the “From BI to AI” roundtable at the Snowflake World Tour Paris, Marie-Jeanne Freiha (Director of Data & Analytics, Clarins), Philippe Thomas (Data Program Director, Saint-Gobain), and Nathalie Bachour (CIO, Devoteam) revealed the behind-the-scenes story of their data revolution.
Moderated by Laurent Letourmy (Devoteam), these three leaders shared a common observation: the rationalisation and modernisation of data platforms are no longer optional, but a necessary step toward new AI use cases.
Between drastic reduction of application portfolios, migration accelerated by generative AI, and the emergence of intelligent semantic layers, their feedback outlines a new paradigm. Data becomes an autonomous product, documented and directly accessible to business users with Snowflake Intelligence.
The Modern Data Stack and Data Products as a New Data Paradigm
The shift from traditional BI to data products represents far more than a simple change in terminology. Where traditional BI remains centralised and top-down, driven by data teams delivering dashboards on demand, the data product approach establishes a model of decentralised data assets, organised by functional domain and equipped with comprehensive documentation.
At Clarins, this transformation is embodied in the restructuring around data mesh and a Modern Data Stack approach. In this approach, autonomous domains (such as finance, supply chain, e-commerce, and physical stores) become producers and owners of their own data. Each data product is presented as a high-quality asset, governed, documented, and directly aligned with the business. For example, a “B2C transactions” asset that aggregates online and in-store sales.
This approach will fundamentally change the relationship between data teams and the business. We move from technical dependency to collaboration based on data contracts and shared business semantics.
Semantic Architecture and User Experience
Creating semantic layers designed for Snowflake Intelligence becomes the key element in activating this new approach. By directly connecting Snowflake data to business concepts understandable in natural language, the semantic layer enables unprecedented interaction. Users ask questions in natural language, and the AI automatically breaks down the query. It identifies relevant data domains, builds the SQL, and delivers the result in table or chart format.
At Clarins, testing this approach on in-store and e-commerce sales transactions revealed its transformative potential. Identifying the number of products sold in stores and via e-commerce becomes possible without SQL skills. The agent reformulates and breaks down the question, locates the right data products and tables, generates a query that can be validated by an analyst, and then proposes visualisation formats. This semantic layer is only possible if it relies on comprehensive upstream documentation.
Data Portfolio Rationalisation
The scalability of legacy systems hides dizzying complexity. Saint-Gobain initially had six different data platforms. In total: 32,000 reports, 5,000 users, massive data volumes, and more than 50 source applications.
The gradual migration to a modern stack (dbt, Fivetran, Snowflake) was radically accelerated thanks to generative AI. Workflows that previously took 2-3 hours to complete manually now require less than 5 minutes with AI assistance. As a result, Saint-Gobain reduced the number of reports from 32,000 to 20,000 in six months. At Devoteam, all exploration is done in “startup mode”: prototype, test, prove, then scale. This approach demonstrates that modernisation is not a barrier but an accelerator for migration.
User Adoption and Concrete Changes in Usage
The real test of success is measured by usage. At Clarins, concrete examples demonstrate a significant shift in usage. An analyst who would have requested a complex SQL query can now formulate it in natural language via Snowflake Intelligence’s semantic layer, explore the results, and visualise them without technical intervention.
At Devoteam, it’s the speed of deployment (sometimes 4 days for augmented code via Streamlit versus 2 to 3 weeks with a classic dashboard approach like Tableau) that becomes a clear productivity advantage, as well as the reusability of the semantic layer. Data refresh via streaming goes from 6-hour delays to near real-time. This opens up use cases that were previously impossible with the old architecture. Self-service based on a well-built semantic layer finally becomes possible. It frees data analysts from report production, allowing them to redeploy on higher-value tasks, such as business enrichment, indicator construction, and predictive analytics exploitation.
Tangible ROI and Technology Trade-offs
Investment in this modernisation generates immediate and measurable ROI. At Saint-Gobain, using AI to automate the testing and migration of MicroStrategy reports to Snowflake projects results in considerable time savings. The team has significantly reduced the number of reports. Full automation of some migration flows opens the prospect of completing 6 platforms by end of 2026 instead of 2027.
At Devoteam, the technology choices (Fivetran for ingestion, dbt Platform for transformation, the Streamlit & Tableau combination for consumption, and Snowflake Intelligence) create a more fluid ecosystem for massive analytical use cases. Specialised agents – one agent for financial controllers, another for operational managers – demonstrate how AI can capture business expertise (management rules, decision-making logic, corrective actions) and scale it across the organisation (10,000 employees, 20 countries).
The real trade-off is not technical but economic. Modernising the data portfolio initially costs more. However, this approach quickly pays off through increased data productivity, reduced licensing costs, and, most notably, the opening of new use cases (predictive analytics, recommendations, and autonomous self-service) that were previously impossible.