Case Studies · E-Commerce

Lakehouse migration that cut the bill 75–85%.

A global Amazon FBA aggregator ran ~900 transformation queries inside Redshift, driving compute to ~$40,000/month with no path to containment. We re-architected to a hybrid lakehouse on EMR, Iceberg and Airflow — and converted the hardest flows first to prove it.

Industry
E-Commerce / FBA Aggregation
Technology
EMR · Iceberg · Airflow
Duration
12 weeks
Scope
350+ of ~900 queries
The challenge

A monolithic ELT, growing more expensive by the acquisition.

Every transformation lived inside Redshift, powering 35+ Tableau dashboards, multiple ML models and write-back automations. As brand acquisitions grew data volumes, the cost and risk compounded.

Escalating compute costs

~$40,000/month Redshift bill driven by ~900 transformation queries, growing with every new brand acquisition.

Monolithic ELT bottleneck

All transformation logic coupled to Redshift — a single point of failure for dashboards, ML and automations alike.

Poor query performance

Critical queries taking 20+ minutes to complete, creating bottlenecks in reporting and operational workflows.

No cost observability

Zero per-query cost or performance tracking, making it impossible to identify expensive queries or optimize spend.

The solution

ELT → ETL: move the heavy lifting off Redshift.

We designed a hybrid lakehouse that decoupled heavy transformation from Redshift, retaining it only as a fast analytical store for the most latency-sensitive queries. The bulk of processing moved to cost-efficient Spark on EMR, with Iceberg on S3 as the storage layer and Airflow for orchestration — all integrated into the client's in-house platform with zero infrastructure changes.

  • Bronze / Silver / Gold medallion layers on S3 Iceberg, catalogued in Glue
  • Redshift-specific SQL refactored into ANSI-compliant Spark SQL
  • Per-query cost & performance observability via OpenTelemetry + Chronosphere
airflow — gold_layer_dag
$ airflow dags trigger fba_gold_layer
bronze iceberg load · schema enforced
silver spark-sql dedupe + joins (EMR)
gold aggregates → iceberg + redshift
validate parity vs redshift · 100%
cost $1.40 · 2m 51s (was 20m+)
# 350+ queries migrated · zero downtime
Solution architecture

A hybrid lakehouse, medallion-layered.

Ingestion

30+ PipelinesHourly & daily
Transient LandingS3 staging

Medallion Lakehouse — S3 Iceberg, Spark on EMR, Airflow-orchestrated

BronzeSchema enforce · partition
SilverCleanse · dedupe · join
GoldBusiness aggregates

Consumption & Governance

Redshift (Gold)Tableau · low-latency
ML ModelsRead Iceberg direct
Glue + Lake FormationCatalog · RBAC
OpenTelemetryPer-query cost

IaC with Terraform across Dev & Prod · GitHub CI/CD · encryption in transit (TLS 1.2+) and at rest (AES-256)

Outcomes & impact

Same dashboards. A fraction of the bill.

75–85%
Monthly platform cost reduction ($40K → $6–10K)
~90%
Faster queries (20 min → 2–3 min)
350+
Queries migrated of ~900 total
0
Downtime at cutover
BeforeAfter
~$40,000/month Redshift compute$6,000–10,000/month hybrid lakehouse
Critical queries at 20+ minutes2–3 minutes — up to 90% faster
All transformation coupled to RedshiftHeavy compute on Spark/EMR, Redshift for Gold only
No per-query cost visibilityPer-query cost & performance observability
No repeatable migration pathDocumented playbook for the remaining ~550 queries
Technology stack
Amazon EMR (Spark SQL)Amazon RedshiftApache Iceberg on S3Apache AirflowAWS Glue Data CatalogAWS Lake FormationAmazon SageMakerTerraformOpenTelemetry + ChronosphereGitHub CI/CD
Your data bill, contained

Is Redshift compute running away from you?

We'll profile your transformation workload, model the savings of a lakehouse split, and migrate the highest-impact flows first — exactly as we did here.

Book a consultation →