Cost Optimization Framework·2022–2024 Client-confidential
Cost & Credit Diagnostics
Finding the leverage point that makes the whole system cheaper.
My role
Diagnostic lead
Outcome
Reduced Snowflake credit spend by roughly 80% across multiple enterprise clients.
01
The problem
- Multiple enterprise clients faced Snowflake bills that grew faster than the value they were getting. The instinct in these situations is to squeeze every warehouse a little — which rarely moves the number and annoys everyone.
- The real problem was almost always a handful of structural decisions: oversized warehouses left running, expensive query patterns repeated at scale, workloads that had never been right-sized to their actual demand.
02
The approach
- I built diagnostic tooling to attribute spend precisely — down to the users, workloads, and query patterns responsible for the majority of cost.
- Rather than optimizing uniformly, the method isolates the small number of high-leverage changes: the warehouse that should auto-suspend, the query that should be materialized, the workload that was scaled for a peak that never comes.
- Each recommendation was tied to a specific, reversible change with a clear expected impact — so clients could act with confidence rather than fear of breaking something.
03
The outcome
- Across multiple clients, credit spend fell by roughly 80% — not through austerity, but because the underlying design got better.
- The patterns proved general enough to productize, and became part of the thinking behind CloudPAHL.
Technologies
SnowflakeSQLPythonCost AttributionQuery Optimization