AWS re:Invent 2025 - Automotive Supply Chain Optimization using AI (PEX305)

Automotive Supply Chain Optimization using AI

Challenges in the Automotive Industry

• Increased product complexity: More vehicle configurations and accessories, putting pressure on supply chains
• Higher customer expectations: Demand for Amazon-like one-click buying and transparency
• Tariffs: Significant impact on profits, with top manufacturers reducing estimates by over $25 billion
• Supply chain disruptions: Strikes, chip shortages, natural disasters affecting P&L and customer experience

Impact of Supply Chain Disruptions

• 94% of companies affected by disruptions
• Increased logistics/transportation costs, mis-deliveries, delayed launches, lost revenue
• All-time high average transaction price of new vehicles ($50,000+), creating affordability issues

Causes of Supply Chain Complexity

• Disconnected processes: Different systems and manual steps between order and delivery
• Multiple data challenges: Inconsistent data, outdated systems, lack of integration
• Legacy technology: Old systems not connected to new technologies or cloud
• Processing bottlenecks: Batch processes and slow data updates

Solving the Challenges

Key principles:

1. Real-time processing: Event-driven platforms to act on changes immediately
2. Unified end-to-end: Seamless integration of processes and data
3. AI-powered: Automation of tasks, "human in the loop" for exceptions
4. Customer-centric: Translating value to the customer through better experiences

Conceptual Architecture

1. Data ingestion:
   • CDC (Change Data Capture) from mainframe
   • SFTP batch process
   • Messaging system (e.g., Amazon Kinesis) to store events as a stream
2. Data storage:
   • Polyglot storage (raw files, structured/semi-structured data)
   • SQL and NoSQL databases (e.g., Amazon Aurora, MongoDB)
3. Data transformation:
   • Data mesh layer for business logic separation
   • Event processing and retry logic
   • Canonical data models (e.g., vehicle, dealer, parts)
4. Data exposure:
   • REST APIs and pub/sub model for downstream applications
   • Enterprise search capabilities
5. Downstream applications:
   • Internal line-of-business apps consuming event data
   • Data science and data engineering use cases
   • Exception management and alerting

Physical Architecture

1. Data migration:
   • AWS Database Migration Service for CDC from mainframe
   • SFTP process for batch data ingestion
   • Data landing in Amazon Aurora database
2. Real-time processing:
   • Amazon Kinesis for event streaming
   • Apache Flink on Amazon ECS for stateful transformations
   • Separate rules engine for maintainability
3. Data storage:
   • MongoDB for current vehicle events, with Elasticsearch for performance
   • Amazon DynamoDB for historical event data (data lineage)
   • Amazon Aurora for reporting and downstream consumption
   • Amazon S3 for data lifecycle management
4. Data access:
   • AWS AppSync for graph query language access to MongoDB
   • Amazon API Gateway for B2B API exposure
   • Amazon QuickSight and Tableau for dashboards
   • Mobile apps and dealer applications consuming APIs

Machine Learning and Predictive Analytics

1. Feature engineering:
   • Data stored in Amazon S3 and databases for feature extraction
   • Amazon SageMaker Wrangler for data cleaning and preparation
   • Principal Component Analysis (PCA) for feature selection
2. Model training:
   • Use of regression and time series algorithms (e.g., XGBoost, Random Forest)
   • Amazon SageMaker for model training, hyperparameter tuning, and batch inference
   • Instance recommender to optimize training cycles
3. Model deployment and monitoring:
   • MLOps practices for model versioning, access control, and drift monitoring
   • Batch transform jobs for pre-computing predictions every 4 hours
   • Integration with real-time inferencing for latest vehicle events

Toyota's Supply Chain Visibility and Predictive ETA

• Visibility into key milestones: Manufacturing, quality, yard, transportation, dealer delivery
• Factors affecting delivery: Vehicle volumes, transit status, day of week, route stability, carrier capacity
• Target outcomes: Accurate ETAs, tight delivery windows, replacement vehicle options
• Customer-centric delivery pipeline: Personalized views, reusable components, optimized schedules

Key Takeaways

• Importance of people and collaboration, not just technology, for successful transformations
• Comprehensive, end-to-end visibility and predictive analytics for automotive supply chains
• Leveraging AWS services (e.g., Kinesis, SageMaker, AppSync, API Gateway) to build scalable, event-driven solutions
• Emphasis on customer-centricity and delivering seamless experiences throughout the vehicle delivery process
• Continuous improvement mindset to stay ahead of evolving customer expectations and industry challenges