Group Data Platform: from 2 weeks to 1 day

The context

When I joined, Data Science teams had fragmented Data capabilities and were heavily dependent on IT processes.

A simple analysis could take more than two weeks, notably because of manual processes, IT-dependent deployments and an environment poorly suited to the full Data Science project lifecycle.

The existing platform also relied on a costly, poorly scalable licensing model, limiting access to Data Science to a restricted number of users.

The challenge was therefore broader than creating a new technical environment: it was about transforming the Data Science delivery model.

From a Data environment to a Group Data capability

I defined the vision and target architecture for a new Data platform covering the full Data Science lifecycle: ingestion, transformation, data exposure, development, deployment and operations.

The objective was to let teams manage their delivery cycle much more autonomously while retaining Group requirements around security, architecture and governance.

The platform was designed with a cloud-native and Security by Design approach, using a modern architecture to industrialise processing and deploy different types of Data and AI applications.

Designing the transformation, not just the technology

The project involved as many organisational and economic trade-offs as architectural choices.

I notably:

• defined the vision, architecture principles and target trajectory;
• selected and structured the technology foundation;
• designed the platform to cover the full lifecycle of Data Science projects;
• coordinated contributions from Architecture, Security and external partners;
• supported changes to development and deployment practices;
• led the progressive exit from an existing solution that had become too costly and restrictive;
• helped evolve the delivery model towards greater autonomy for Data Science teams.

The platform was designed around reusability and industrialisation, so different projects could be hosted without rebuilding their technical foundation each time.

A change in the operating model

Before the transformation, a change could follow a cycle such as:

Data Science → IT request → deployment → wait

The new platform gives teams much more direct control of their cycle:

Code → pipeline → deployment → operations

This reduced operational dependencies while bringing Data Science practices closer to software-engineering standards: version control, automation, separate environments and reproducible deployments.

The goal was not to remove IT or Architecture teams, but to clarify responsibilities and give Data Science teams the autonomy required for their work.

Balancing autonomy, security and cost

The platform had to meet three constraints simultaneously:

• accelerate delivery significantly;
• comply with Group security and architecture requirements;
• remain economically sustainable.

Security was therefore integrated into the architecture from the start rather than treated as a later step.

This approach notably enabled the platform to meet security requirements assessed through controls conducted on the Data Science environment.

Results

>2 weeks → 1 day

The time required to produce and make available a Data Science analysis or POC fell from more than two weeks to around one day.

~€60k / year

Exiting the existing solution reduced annual platform costs by around €60k while removing user-count limitations.

End-to-end autonomy

Data Science teams can now directly manage a much larger part of their development and deployment cycle without systematically depending on IT intervention.

A foundation for Data & AI

The platform provides a shared foundation for industrialising Data Science and AI projects, from data preparation through deployment and operations.

What I take from it

A Data platform is not just an assembly of technologies.

Its value comes from its ability to transform how an organisation works with data: reduce dependencies, accelerate delivery, secure usage and make industrialisation repeatable.

The role of a Data leader is therefore not only to choose an architecture. It is to align technology, organisation, governance and economics to create a lasting capability.

This combination is what turns a Data Science team into a true Data & AI capability at Group scale.