With 168 million passengers and 30 million customers a year, TGV-INTERCITES runs one of the world’s most complex rail operations. Data drives every pricing decision and performance analysis behind it, and until recently it was governed in silos.
Facing open competition, tighter regulation and growing AI ambitions, TGV-INTERCITES moved from a project mindset to a product mindset, drawing on Data Mesh principles and federated governance.
DataGalaxy is one of the tools supporting this shift, in charge of exposing data products and tracking access requests.
Challenges
TGV-INTERCITES, SNCF Voyageurs entity for high-speed and long-distance rail, faced 3 main challenges:
- Open competition. New operators are entering the high-speed market. Steering yield, customer knowledge and rail production now requires reliable data that’s comparable across teams.
- Growing regulatory complexity. PRR, GDPR, the EU AI Act and France’s Mobility Orientation Law (LOM) demand rigorous traceability, classification of personal and sensitive data, and access based on data minimization.
- A data organization that no longer scaled. Before 2022, each domain built its own datasets with no clear ownership, shared documentation or visibility. Data was reused without traceability, access requests were handled informally, and the central data team was overloaded.
In 2022, TGV-INTERCITES launched its multi-year Data Performance program (2022-2026) to break down silos and move to a data mesh model. A Data Office now sets the rules and shares best practices, while domain teams own their data.
Why DataGalaxy
DataGalaxy covers two building blocks the Data Office could no longer handle manually: a data product marketplace and access request tracking.
- Supporting federated governance. DataGalaxy centralizes data product documentation, keeps cataloging standards consistent across domains, and gives the Data Office a cross-domain view of listed products and granted access.
- Exposing data products on a marketplace. Instead of relying on word of mouth, data engineers and analysts will find every certified data product in one place, with its documentation, SLA level, regulatory classification, and consumption point.
- Managing access in compliance. The access request workflow traces every grant, involves the GDPR team for personal data, and ties access to data sensitivity and declared purpose.
Data as a Product
TGV-INTERCITES is moving from project logic, where each team builds what it needs, to product logic, where every dataset is a documented, governable and reusable asset.
A data product must be accessible to other teams, documented in DataGalaxy (owner, steward, classification and data contract) and published on the marketplace. Each one carries:
- General information: description, domain, consumption point, named data owner and data steward, plus a data contract covering data depth, SLA level, update frequency and file format.
- Regulatory information: data classification, personal data type and sensitive data type, which determine the access rights granted.
DataGalaxy Implementation
A model inspired by Data Mesh
TGV-INTERCITES’s data organization rests on three mutually reinforcing pillars.
- Data products and accountability. Each domain (Customer Relations, Distribution, Revenue & Finance, Production, Offer Management, HR, Rolling Stock and more) produces, documents and maintains its own data products, and is accountable for their quality, access and support.
- Platform services. The central data team provides shared building blocks so every domain works to the same standards: DataGalaxy for exposure and access tracking, Databricks for processing, Azure Data Lake for storage, Unity Catalog for access control, and Datadog for observability.
- Federated governance. The Data Office sets the rules, leads the community of data stewards and data owners, and relies on DataGalaxy for data product exposure and access management.
The technical ecosystem
Data moves through three zones of progressive qualification:
- Bronze: raw data ingested through Apache NiFi into Azure Data Lake
- Silver: standardized data products built on Databricks and Delta Lake
- Gold: enriched data products for business teams, consumed in Power BI
DataGalaxy sits across all three as the federated governance layer, alongside Microsoft Entra ID, Datadog, and Unity Catalog.
Access management: a decentralized, compliant process
Data access must serve business use cases while limiting regulatory and confidentiality risks. DataGalaxy’s access workflow involves four roles:
- The Business User creates the access request and follows it through.
- The Data Office approves the request or flags it as incomplete, and oversees the whole process.
- The GDPR team gives its opinion whenever the request involves personal data.
- The Data Steward / Data Owner makes the final call within their scope and approves the access grant.
Every grant will be traced, justified by a declared purpose and approved by the person who knows the data best, answering both GDPR and rail data security requirements.
Results
The transformation is ongoing, but foundations are already in place:
- 7+ business domains owning, documenting, and maintaining their own data products
- 5-tool unified stack: DataGalaxy, Databricks, Azure Data Lake, Unity Catalog and Datadog
- 3x fewer data requests with governed, discoverable data products
The direction is clear: from siloed datasets to governed, discoverable data products.
Key Takeaways
You can’t govern data with policy alone. It takes process, change management and accountability on the ground, backed by tools chosen for what they do well.
Data Mesh without tooling stays theoretical. Handing ownership to domains only works if each domain has the tools to take it on. DataGalaxy is one targeted building block: a marketplace for visibility, and traced access requests for compliance.
Compliance is built into the product, not bolted on later. With classification and sensitivity stored in each data product’s record, compliance becomes a native attribute, not a last-minute check before go-live.
A marketplace changes how people relate to data. When engineers and analysts can find documented products on their own, informal requests to producing teams drop and data use improves.
Access management is an act of governance. Tracking who has access to what, and why, is what makes it safe to share sensitive data across domains.
Structure first, then expose. Publishing undocumented products would only industrialize the mess. TGV-INTERCITES structures first, exposes next, and adds access management once the foundations are in place.
