Moovit Cuts Analytics Cloud Spend 33% via Multi-Engine Modernization

In an architectural case study published on September 3, 2026, AWS detailed how mobility platform Moovit overhauled its analytical platform to cut cloud pipeline costs by 33%. Facing scaling bottlenecks and ballooning expenses from an expanding provisioned Amazon Redshift RA3 cluster, Moovit transitioned from a monolithic data warehouse to a multi-engine lakehouse model. The team segregated workloads by function: heavy daily and weekly data transformations were offloaded to Spark on Amazon EMR Serverless, exploratory and ad-hoc analyst workloads were routed to Amazon Redshift Serverless via data sharing, and the core provisioned Redshift cluster was reserved strictly for serving curated business intelligence queries. This restructuring reduced the provisioned Redshift compute footprint by 50% while achieving more than a 40% reduction in transformation runtimes. As enterprise data volumes scale, treating a data warehouse as a one-size-fits-all compute layer creates severe financial and operational inefficiencies. Provisioned clusters are frequently over-scaled simply to handle periodic compute spikes during batch transformation windows, leaving costly compute capacity idle during quiet periods. By moving variable, compute-heavy transformations to ephemeral serverless processing engines, data platform engineering and FinOps teams can align compute expenses directly with actual execution times. The Moovit case demonstrates that meaningful cost optimization does not require degrading query latency or platform capabilities; rather, deliberate workload segmentation simultaneously lowers baseline costs and protects interactive query performance. This modernization reflects a broader shift across cloud data architectures toward modular, multi-engine lakehouse architectures. Over the past several years, monolithic data platforms have increasingly given way to open data formats residing on object storage like Amazon S3, accessed by specialized engines tailored to distinct workloads. Rather than locking all processing within an expensive, always-on data warehouse, modern FinOps-aligned engineering organizations increasingly structure compute into three distinct tiers: transient batch processors (such as Spark on EMR) for ETL, on-demand serverless engines for intermittent discovery, and dedicated provisioned clusters for steady, low-latency BI dashboards. For practitioners looking to optimize data infrastructure costs, the primary takeaway is to profile and decouple compute workloads rather than default to continuous vertical scaling. Engineering teams should audit their data warehouse query logs to isolate heavy batch transformation pipelines and intermittent ad-hoc querying from critical production reporting. While orchestrating multiple purpose-built engines introduces slight pipeline complexity, validating data-quality parity across engines and leveraging data sharing yields compounding cost dividends as analytical demand continues to grow.