The workflow before
Nu Skin had moved to AWS but carried its old patterns with it: 150+ overlapping ETL jobs that took 6 to 8 hours to run and failed 15 to 20% of the time. Engineers spent about 120 hours a week keeping them alive. Business data arrived 24 to 48 hours late, and analysts spent 60% of their time preparing data across 50+ markets.
What we built
We consolidated 150+ pipelines into 40. Kafka streams events from regional ERPs, e-commerce, supply chain and affiliate systems into an S3 data lake; Airflow orchestrates; Matillion reshapes the data and pushes work down into Snowflake.
We built one global data model for customers, products, affiliates and the chart of accounts, served through Tableau with role-based access, so analysts can answer cross-market questions without engineering.
We reviewed the whole AWS estate against the Well-Architected Framework: right-sized compute, Reserved and Spot Instances, scheduled shutdowns and cost allocation tags. Monthly spend fell from $185K to $128K.
Before and after
| Measure | Before | After |
|---|---|---|
| Monthly cloud cost | $185K | $128K |
| ETL pipelines | 150+ | 40 |
| Pipeline runtime | 6 to 8 hours | 1 to 2 hours |
| Pipeline maintenance | 120 hrs a week | 30 hrs a week |
What's next
With one global data model in Snowflake, Nu Skin is ready for L2 work: demand forecasting by market and an analytics assistant that answers finance and supply chain questions directly from governed data.
