Siemens Healthineers, one of the world's largest medical technology companies, collects roughly 100 TB of Magnetic Resonance Imaging (MRI) scanner data each month for service, stability and compliance. After two decades, the limitations of its on-premises platform became evident. An early lift-and-shift to the cloud came back two to four times more expensive than staying put. Rebuilding on the Databricks Data + AI Platform with Delta Lake and Unity Catalog flipped that to around 50% lower costs, reduced constant copying between silos and set the foundation for Genie One self-service analytics, insights and data democratization.
Challenges faced by Siemens Healthineers in managing MRI scanner data
Since 2007, Siemens Healthineers has collected data from its MRI scanners over the Siemens Remote Service connection. Hundreds of log files feed three essential jobs: helping service teams stay ahead of scanner problems, monitoring field stability and meeting post-market surveillance requirements. Two decades on, the MRI install base alone generates billions of event log lines and roughly 100 TB of data each month. Every new scanner and software upgrade adds more.
Designed in 2007 with scalability in mind, XMART served the business for two decades and grew into one of the largest SQL deployments Microsoft had seen. But the economics and reach of an on-premises design eventually ran into three limits: cost, data sharing and access.
Cost challenges of scaling MRI data management
As volumes climbed, some Analysis Services cubes had to be taken offline because the data outgrew them. Scaling also meant provisioning servers years ahead of demand, sized for where the data would be rather than where it was. The approach kept working, but its cost curve continued to rise.
Data sharing and access bottlenecks across teams
Each business line ran its own server silo, so moving data to another team meant exporting it, encrypting it and physically transferring large files. Because the history dates back to 2007, requests for older records incurred significant computational overhead. The same data was often delivered and reworked two or three times. "If someone needs historical data, it could take weeks or months to get this data," said Georg Görtler, Product Manager for XMART at Siemens Healthineers.
More than 1,000 people across sales, service and headquarters functions relied on Qlik Sense dashboards, and most worked in Excel rather than SQL. Any questions the dashboards did not already answer were routed to a central team. At the same time, hospitals increasingly sought operational insight through teamplay to schedule patients and get more from their scanners. "There has always been a bottleneck in a central team, where users wait, and prioritization has to happen," said Michael Kelm, Cloud Data Platform Lead at Siemens Healthineers.
Rebuilding XMART on Delta Lake for improved scalability and governance
To modernize XMART, Siemens Healthineers needed a cloud migration approach that improved scalability and data governance without creating a cost overrun. The team found that simply moving the existing architecture to the cloud would not deliver the economics or access model they needed.
A rebuild on Delta Lake, not a lift and shift
An early proof of concept that lifted the SQL estate into the cloud unchanged came back two to four times the cost of staying on-premises. That was enough to stop the move before it started. A second proof of concept with Databricks changed the approach. The team rebuilt XMART on Delta Lake and split the monolithic database into discrete data products. This created room to process roughly 100 TB of MRI data each month while paying only for the compute it used.
Governed sharing through Unity Catalog: access by permission, not by copy
With that foundation, data no longer had to move to be useful. The team grants governed access to shared data products in a single environment. It is migrating to Unity Catalog one data product at a time. Because everything sits in a single metastore under a single account, sharing across business lines is now a permission grant rather than a copy. Customers and partner teams are shifting from centrally maintained SQL endpoints to querying with their own compute. This gives each team control over performance and cost. "In the cloud, we don't have to copy the data. It is just to share the data," Georg said.
Splitting the monolith into data products also made spending visible for the first time. Because each product is tracked individually, the team can measure exactly what the storage and processing costs are. This turns optimization into targeted work with results that show up in a dashboard.
Achieving cost efficiency and self-service analytics with Genie
With XMART rebuilt on Databricks, Siemens Healthineers is seeing a clearer path to cost control, broader access and self-service analytics. The shift also strengthens the governed foundation needed to expand AI capabilities over time.
Key outcomes include:
Costs are expected to be around 50% lower than the on-premises setup, reversing the earlier estimate that projected costs to be two to four times higher.
Historical data access is shifting from weeks to near instant through governed permission grants rather than exports and transfers.
Per-product optimization, reducing the cost of two data products by 80% and 90%.
Scanner insight reaching hospitals through teamplay as another governed data product.
A transition toward Genie One self-service for 1,000-plus analytics users.
"We can measure the cost of each data product individually, so one person can spend a month optimizing a process, and we see the savings in the dashboard," said Paweł Tajs, Lead Data Architect for XMART.
With the migration nearly complete, the company is turning to the self-service layer. Unity Catalog is the technical prerequisite for Genie One. The team is finishing that migration and describing its datasets so that it’s more than 1,000 analytics users can go straight to Genie One and ask questions in plain language rather than waiting on a central team. Working students are already prototyping cloud-based AI with promising early results. Siemens Healthineers treats this governed foundation as the base for a wider AI roadmap. "Every week, two to five people ask me what they can do with the data. Soon Genie will answer for them, and it might even stand in for me when I retire in a few months," Georg said.
FAQ: Siemens Healthineers and Genie on Databricks
Genie One is a natural-language interface on the Databricks Data + AI Platform that lets users ask questions about their data in plain language rather than writing SQL. It turns those questions into queries and returns answers directly, opening data to people without technical querying skills.