AWS Transportation Execution Platform
Redesigned lululemon's shipping, rating, address-validation, and routing API platform for high-throughput guest-order fulfillment across 7 distribution centers and 500+ stores — reducing AWS spend by 60% and integrating 5 parcel carriers under standardized onboarding patterns.
Context
lululemon's guest-order fulfillment relied on legacy point-to-point integrations connecting 7 distribution centers and 500+ stores to 5 parcel carriers (UPS, USPS, FedEx, Canada Post, Australia Post). The architecture had accumulated significant technical debt, lacked operational observability, and was driving up AWS infrastructure costs — while peak-season demand left no margin for unplanned downtime.
Objective
Redesign the core transportation execution platform — shipping, rating, address-validation, and routing — to handle high-throughput order volume, reduce infrastructure spend, simplify carrier onboarding, and give operations teams real-time visibility into platform health.
Approach
Designed an AWS-based microservice and orchestration architecture (S3, SQS, EC2, API Gateway) optimized for transportation execution throughput. Ran targeted cloud cost-optimization initiatives against over-provisioned services and idle resources. Standardized carrier integration patterns across all 5 parcel carriers to eliminate bespoke onboarding work for each new carrier. Modernized the inbound logistics data architecture on AWS and Snowflake, designing a product-classification service that determines classification codes and duty rates by destination country — integrated via Kafka and Postgres into E2Open.
Outcome
Reduced AWS spend by up to 60% through cost-optimization. All 5 parcel carriers now operate under unified onboarding and monitoring patterns, reducing future time-to-integrate. Improved operational observability across distribution centers and stores. Inbound product classification runs automatically at order time, replacing a manual compliance process.
Lessons Learned
Cost optimization and performance optimization are often in tension — the discipline is knowing which trade-off a given service can tolerate at peak. Standardizing carrier onboarding patterns before the fourth carrier rather than after would have recovered significant engineering time.