AXA Japan is a leading life and property & casualty insurer where data is a core input for underwriting, operations, and customer experience. The business relies heavily on diverse datasets, from sensitive health and claims histories to complex risk modeling. Over time, AXA Japan has achieved high data maturity, with more than half of its workforce actively using analytics. While this widespread adoption forms a strong foundation for data-driven decisions, it has also raised internal expectations for speed, reliability, and data accessibility.
Following a broader organizational integration initiative, the company found itself managing information across multiple legacy systems. AXA Japan saw an opportunity to simplify that complexity: they chose Databricks to join siloed data estates into a single governed lakehouse, creating a common foundation for analytics today and AI-driven innovation in the future.
Unifying fragmented data with a trusted foundation
As AXA Japan joined its life and P&C businesses under a broader OneAXA transformation, it faced a familiar enterprise challenge: too many disconnected systems. Each business had built its own data lake, legacy applications, and operational processes over time.
The goal was not simply to modernize infrastructure. AXA wanted to create a single source of truth that would enable teams to access data more easily, improve data quality, and support a growing range of AI use cases. “We wanted to make sure that we put all the data in one place where we could access it very efficiently and where we could ensure that we have good data quality,” said Hebrant.
Before selecting a platform, the team evaluated several options. Ease of access, governance, and cost were important, but so was future readiness. AXA specifically wanted a platform that could support MLOps from the start, enabling production-grade AI systems that could be monitored, maintained, and improved over time.
“We decided to go with Databricks,” Hebrant said. “This is a choice that we have definitely not regretted so far.”
Migrating pragmatically from legacy systems to the Lakehouse
AXA Japan approached the migration as a phased modernization rather than a full rebuild. The team consolidated separate data lakes while gradually retiring legacy Redshift and Oracle-based systems that had accumulated over years of business-specific customization.
Initially, the team planned to refactor legacy workloads during migration, but quickly recognized the risks and complexity. “Those legacy systems have so much history,” said Yuka Inose, a Data Business Partner at AXA Japan. “The logic is really complex, which made it difficult to refactor entire systems. We decided to change our approach to lift and shift, which worked quite well.”
That pragmatic approach accelerated delivery. AXA used automated code conversion to move legacy ETL logic from Oracle and Informatica into modern SQL workflows, including dbt models running directly on Databricks SQL. “During our migration from Oracle to Databricks, we encountered several challenges related to both functional and non-functional requirements, migrating from OLTP to OLAP databases,” said Senior Data Engineer Keisuke Taniguchi. “However, features like SQL scripting in Databricks have been incredibly helpful, and the support from OSS such as dbt and Terrafrom has also been very valuable.”
By moving incrementally, the company was able to run existing and modernized systems in parallel, reducing business risk while validating performance.
The phased model also helped teams modernize while preserving user continuity. “We were able to develop by ourselves thanks to Databricks and dbt,” said Kento Yoshida, a Senior Data Engineer at AXA Life Japan. “I would say nearly 40% faster development after the migration.”
Lower costs, stronger governance, and a foundation for AI
By bringing workloads together on Databricks, AXA Japan reduced warehouse costs by up to 60% and cut ETL costs by 70% for certain parts of the migration effort. Retiring legacy Oracle servers also reduced infrastructure overhead and maintenance.
But cost savings were only part of the outcome. The bigger shift was operational. The team could now analyze data across previously separate systems, making cross-business insights easier while creating a stronger foundation for self-service analytics. Business teams that previously relied on static extracts are now accessing governed data directly, including for exploratory work in SQL and Python.
Governance became a major enabler. “Unity Catalog is definitely the right tool,” Hebrant said. “We want to really put all the access under control.” By centralizing governance with Unity Catalog, AXA can enforce access boundaries across business units while maintaining a single shared platform.
That unified platform is also shaping AXA’s future AI roadmap. The company is building a strong, well-governed foundation for long-term AI readiness, developing prototypes and operationalizing machine learning use cases spanning fraud detection to customer support automation. As Hebrant summarized, “Databricks is our core tool – not only for the data infrastructure, but also for the future deployment of AI use cases.”
As AXA Japan completes its broader data migration and modernization, the company is building toward a future where analytics, governance, and AI run on a single, trusted platform, making it easier to serve customers faster, operate more efficiently, and innovate at scale.