How Paris's Transit Giant Turned Metadata into a Strategic Asset
When most people think about RATP — the operator behind Paris's metro, bus, and RER networks — they think about trains running on time. But behind the scenes, the organization is asking the same question every AI-era enterprise eventually asks: how do we use AI to create sustainable competitive advantage reliably, without the AI hallucinating?
For RATP, the stakes are unusually concrete. Beginning in 2025, competition entered the Paris bus market for the first time in the network's history. Tramway follows in 2029. At the same time, RATP is deploying the MF19 — a new generation of hyperconnected trains across eight metro lines — and expanding the network by 200 kilometers and 68 stations. Service quality obligations are contractual. The margin for data error is narrow.
From Fragmented to Factory
In 2021, an internal audit surfaced what many large organizations eventually find: fragmented data centers, underused infrastructure, and technology choices that varied by team rather than by strategy. The data existed, but the shared understanding of what it meant and how it is best put to work for the business did not.
RATP's response was to build a Data Factory — a team of roughly 100 data professionals operating under a shared set of principles: value creation, fast time-to-market, and technological consistency. Led by Antoine Charpentier, the team works on structured timelines: five weeks to scope a project, three to four months to a minimally-viable product (MVP) to speed up innovation.
The underlying platform — internally called Diane — launched in December 2022. Built on AWS, Databricks, Power BI, and Collate, it now delivers 14 data products across 12 business domains, with advanced access controls and clear ownership throughout.
Metadata as a Strategic Asset
What's notable about RATP's approach isn't just the tooling or the team structure. It's the framing of how metadata and semantics support business operations and AI.
Alexandre Anquetil, Lead Data Product Management at RATP's Data Factory, puts it directly:
"At RATP's Data Factory, especially in the context of the rise of AI, we consider metadata as a strategic asset that contributes to serve people mobility through the performance and the compliance of our daily operations. Collate provides all the capabilities in one platform that allow us to carry out our metadata management activities efficiently to ensure consistent data usage and trust."
This is the shift that separates organizations building durable AI capabilities from those still running one-off experiments. Metadata isn't an administrative afterthought — a catalog you maintain because someone said you should. It's the foundation that determines whether your AI applications produce reliable outputs or confident-sounding ones.
At RATP, that foundation is already supporting real AI programs: predictive models for personnel assignment, GenAI-powered chatbots in stations that answer questions about equipment and fares, predictive maintenance for rolling stock, and driver coaching tools designed to optimize operations across the network.
None of that works reliably without consistent metadata. And consistent metadata doesn't happen without deliberate data and AI governance.
The Foundation That Makes It Possible
What RATP built is a blueprint for AI-ready data platforms. By establishing semantic foundations first to include clear ownership, consistent definitions, governed lineage, shared meaning across the organization, it created the conditions for AI to perform reliably at scale. The use cases followed from that foundation, not the other way around.
That sequencing matters. Most organizations approach AI adoption in the opposite order: they invest in models and infrastructure, then discover that the underlying data is too inconsistent to produce trustworthy outputs. The fix isn't more compute or a better model. It's the semantic intelligence layer that sits across your data estate, connecting what data exists to what it means, who owns it, how it flows, and what you're allowed to do with it.
That's what Collate was built to do. And it's what organizations like RATP are putting into production.
For a deeper look at the principles behind this architectural shift, we recommend reading The Value of Trusted, Secure Data and Unified Semantics in the Era of AI — an independent analysis by Mike Ferguson of Intelligent Business Strategies, one of Europe's (and the World's) leading voices on data management and AI readiness.
Source: "Avec sa Data Factory, la RATP consolide la gestion des données", Le Monde Informatique, Oct. 10, 2025 (article in French)