The problem
BlueSprig Pediatrics ran its reporting on a legacy SQL Server and Power BI stack. Routine operational reporting took weeks of manual work. The platform carried heavy licensing and support costs, and every new question from Finance or Clinical Ops queued behind an analyst. For a KKR-backed provider reporting monthly to PE sponsors, weeks of latency is a business problem.
The approach
Instead of a conventional cloud-warehouse migration, my team rebuilt the estate as a DuckLake lakehouse on MotherDuck (DuckDB):
- MotherDuck (DuckDB) as the analytical engine, with serverless economics sized to the workload we actually have
- Prefect Cloud for orchestration, replacing hand-run refresh jobs with observable, scheduled pipelines
- Soda for continuous data validation, so quality failures surface before stakeholders see them
- Everything version-controlled in GitHub with CI/CD — the data platform ships like software, because it is software
The outcome
Platform and staffing costs fell by more than 75%. Reporting that took weeks became self-service. The board package draws from governed, continuously validated data, and the platform now underpins an AI-assisted analytics layer that lets non-technical stakeholders query the lakehouse directly.
Why it worked
Two decisions mattered. We sized boring, proven components to our actual workload and ignored the vendor reference architectures. And we ran the migration like a product launch for Finance and Clinical Ops, so adoption was built in from day one.