Oracle Cloud

Data Warehouse

Modernizing a Legacy Oracle Exadata Data Warehouse for a Global Retailer

4 min read

Mar 5, 2026 1:46:22 PM

How Pythian turned aging Exadata into a cloud-native analytics and AI platform

A global multi-brand retailer with 2,000+ stores across North America and Europe ran its data warehouse on aging Oracle Exadata hardware approaching end of support. Holiday traffic exposed critical bottlenecks, Oracle licensing consumed over 40 percent of data infrastructure spend, and business teams waited hours for reports. Pythian migrated analytics workloads to Google BigQuery, stabilized OLTP on Oracle Cloud, and delivered real-time insights, production AI, and seven-figure annual savings.

90%

Reduction in peak-load query time

$1.8M

Annual savings

99.99%

Peak-season availability

Account profile

Industry: Retail / E-commerce

Organization scale: Global enterprise, $2B+ revenue, 2,000+ stores

Tech stack:

Oracle Exadata X5 (on-prem)
Oracle RAC / PL/SQL (500K+ lines)
OBIEE / ODI / Informatica
Google BigQuery / Looker
Airflow / Fivetran / dbt
Vertex AI

When peak season breaks the data warehouse

The retailer's Exadata environment served the business for over a decade. But omnichannel complexity — e-commerce traffic doubling year over year, multiplying marketplace integrations, and data-hungry demand models — pushed the platform past its limits during the most recent holiday season.

A $3M hardware ultimatum

The retailer's Exadata X5 hardware was 18 months from end of support. Oracle quoted a $3M+ rack refresh on top of $4M+ annual licensing. The CFO demanded a business case for continued Oracle investment versus cloud-native alternatives.

500K lines of PL/SQL, zero workload separation

A decade of development left 500,000+ lines of PL/SQL across stored procedures, materialized views, and ODI/SQL*Loader pipelines. Analytics and transactional workloads shared one Exadata cluster. Smart Scan degraded under holiday concurrency, and IORM tuning had reached diminishing returns.

Black Friday data blackout

Demand-planning queries ballooned from 20 minutes to three hours, forcing merchandising to replenish on stale data. E-commerce conversion analysis lagged six hours. OBIEE dashboards timed out. Finance missed two reporting deadlines. The warehouse wasn't slow — it was costing revenue.

From projection-era complexity to cloud-native advantage

Pythian followed a clear principle: stabilize what's running, map every dependency, then move — without disrupting production analytics. The engagement spanned Foundation Layer stabilization through Platform modernization to Production AI enablement, and we delivered it as a single coordinated program.

Discovery phase

We conducted a four-week assessment covering performance baselines (AWR/ASH/ADDM), storage, RAC health, licensing, and workload classification. The assessment revealed that over 60 percent of licensing cost came from analytics workloads that didn't need Oracle.

The audit uncovered $800K in immediate savings from unused options (Advanced Security, In-Memory, Partitioning) and over-provisioned cores. When combined with projected post-migration savings, the total business case exceeded $1.2M per year and funded the modernization from day one.

Strategic architecture

We designed a workload-separation architecture with two parallel tracks.

A previous vendor had attempted automated PL/SQL conversion and abandoned the project at 58 percent. Pythian automated 70 percent using Google Batch SQL Translator and Ispirer, then manually refactored the complex 30 percent — the advanced packages, autonomous transactions, and DBMS_* dependencies where generalist firms hit a wall. We also redesigned Smart Scan-dependent query patterns into BigQuery-native columnar strategies.

Track A — Analytics exit to BigQuery

We migrated all OLAP workloads (demand planning, inventory analytics, financial reporting) to BigQuery, replaced OBIEE with Looker for self-service BI, and replaced ODI with Airflow, Fivetran, and dbt for modern orchestration.

Track B — Oracle OLTP to OCI

We kept mission-critical transactional workloads (POS, order management, inventory writes) on Oracle and migrated them to Exadata Cloud Service on OCI, capturing cloud elasticity without disrupting PL/SQL-dependent applications.

Implementation roadmap

Phase 1: Assessment and licensing optimization

We assessed the full Exadata environment, optimized licensing immediately by deactivating unused options and right-sizing cores, and delivered the migration business case to the CFO.

Phase 2: Migration and PL/SQL refactoring

We migrated analytics to BigQuery by business priority, refactored PL/SQL (70 percent automated, 30 percent manual), replaced ODI and SQL*Loader with Airflow, Fivetran, and dbt, migrated OLTP to Exadata Cloud Service on OCI, and maintained dual-run synchronization for six months with data validation and rollback strategies at every stage.

Phase 3: Analytics, AI, and managed services

We deployed Looker dashboards to replace OBIEE, trained a Vertex AI demand-forecasting model on three years of transaction data integrated into the replenishment workflow, and provided 24/7 managed services for the OCI Oracle environment.

From bottleneck to real-time retail intelligence

The modernization changed how the retailer uses data. Merchandising runs on real-time demand signals. Finance closes faster. The AI demand-forecasting model paid for the entire migration in its first production quarter.

90% reduction in peak-load query time

Demand-planning queries dropped from three hours to under 12 minutes on BigQuery. Concurrent dashboard users scaled from 50 to 500+ with no degradation.

$1.8M in annual savings

Licensing dropped $1.2M after deactivating unused options, right-sizing cores, and migrating analytics off Oracle. The remaining $600K came from decommissioning hardware and shifting to consumption-based BigQuery pricing.

99.99% peak-season availability

Zero unplanned downtime through the first post-migration holiday peak.

AI-driven demand forecasting

Overstock fell 18 percent and stockouts fell 12 percent in the first holiday season. Month-end close reporting went from five days to two.

Oracle Exadata consulting services

Ready to solve your data challenges?

Speak with a Pythian Oracle Exadata expert ->

On this page

Ready to unlock value from your data?

With Pythian, you can accomplish your data transformation goals and more.

Speak with Pythian consultants now →

Services