Data Warehouse Migration
Snowflake → S3/Athena migration with 80% cost reduction and zero-downtime dual-write cutover
80% cost reduction
1.1 TB migrated
The Problem
Snowflake costs consuming 80% of data budget, split-region infrastructure, 1,294 BI queries locked to legacy warehouse
Snowflake costs consuming 80% of data budget, split-region infrastructure, 1,294 BI queries locked to legacy warehouse. A mobile gaming studio was running its analytics on Snowflake across two AWS regions at $3,500/month — with 90% of spend driven by a single BI warehouse. Revenue pipelines, attribution data, and financial reporting were split between regions with no clear path to consolidation. External data-sharing dependencies and 1,294 Metabase queries created migration risk that made a lift-and-shift impossible.
We designed a 9-phase migration plan that replaced Snowflake with S3 + Glue + Athena in a single region. The key architectural decision was a dual-write strategy: every production pipeline writes to both Snowflake and the new S3/Iceberg stack in parallel, enabling row-by-row reconciliation before cutover with zero data loss risk.
13 revenue ETL pipelines were audited, containerized on ECS Fargate, and extended to write Apache Iceberg tables alongside their existing Snowflake loads. 223 Glue Data Catalog tables were registered across three databases (revenue, attribution, financials). A daily pLTV data pipeline delivers cohort-level attribution data to an external modeling partner at $5/month operating cost — replacing a vendor data-share dependency.
The migration was sequenced to minimize business disruption: revenue pipelines first (highest cost, simplest schema), then attribution data (most complex, external dependencies), then BI cutover (1,294 Metabase queries validated against Athena). Each phase includes automated reconciliation before the Snowflake writer is disabled.
Cost reduction
BI queries migrated
pLTV pipeline cost
Editorial notes
Mandate
Cut warehouse costs by 80% without disrupting 1,294 live BI queries or losing a single row of revenue data during the transition.
Signal
The dual-write pattern turned a high-risk migration into a series of verifiable, reversible steps — each phase provably correct before the next begins.
Context
A mobile gaming studio was spending $3,500/month on Snowflake across two AWS regions. 90% of warehouse cost came from BI queries. Revenue and attribution pipelines were split between regions with no consolidation path.
Constraint
1,294 Metabase queries depended on Snowflake. External data-sharing dependencies had no direct Athena equivalent. Zero data loss was required during migration.
Intervention
Designed a 9-phase dual-write migration: Snowflake → S3/Iceberg/Athena. 13 pipelines writing to both systems in parallel, 223 Glue tables registered, automated reconciliation per phase, and a $5/mo pLTV delivery pipeline replacing a vendor data share.
Outcome
80% projected cost reduction, 7 pipelines dual-writing in production, single-region consolidation, 223 Glue tables operational, $5/mo pLTV pipeline live with 91–100% data accuracy.
Architecture
Dual-write migration from Snowflake to S3/Iceberg with zero-downtime cutover
Dual-Write Pipelines
Iceberg Data Lake
pLTV Delivery Pipeline
BI Migration
Tech Stack
Target Stack
S3 + Glue + Athena (Apache Iceberg)
Source Stack
Snowflake (1.1 TB, 5 databases)
Pipelines
ECS Fargate + Lambda (Python 3.12)
Orchestration
EventBridge + Step Functions
BI
Metabase (1,294 queries migrated)
Delivery
Cross-account S3 pLTV export
Results
Planning a Similar Mandate?
A direct working session about the problem, the constraints, and the fastest credible path forward.
We respond within 4 hours during business hours