Cloud-native data lake replacing costly legacy Oracle infrastructure

Challenge

Real-time telemetry from thousands of vehicles, point-of-sale events from hundreds of locations, scheduling data from a 30-year-old planning system, and external feeds (weather, traffic, fuel prices) — all flowing into an Oracle warehouse that was struggling at 12TB and would not scale to the projected 50TB within three years. Oracle licensing alone was approaching seven figures annually. Real-time use cases (fleet rerouting, dynamic pricing) were technically blocked by batch-only access patterns.

Approach

Designed a lakehouse architecture on AWS — S3 as storage, Glue and EMR for batch processing, Kinesis for streaming, Athena and Redshift Serverless for query. Built a strict zoning model (raw → cleaned → modeled → mart) with full lineage. Migrated workloads in waves: dashboards first, then ad-hoc analytics, then real-time pipelines. Decommissioned the Oracle analytical instance after a 6-month parallel-run validation.

Outcomes

• Storage costs down ~70% versus the prior Oracle warehouse
• Scaled past 30TB without re-architecting
• First real-time fleet-rerouting use case shipped within six months of platform go-live
• Self-service analytics for 40+ business users where previously only the data team had access