Transforming Inventory Forecasting with Production ML
summary
A production reference architecture on Azure ingests ERP/POS/IoT into a unified data lake with Azure Data Factory and runs models in Azure Machine Learning behind CI/CD, model registry and monitoring. Feature engineering explicitly uses lag features, promotion and weather signals; validation relies on backtesting with MAE and MASE plus bias/fairness checks. The deployed pipeline turns forecasts into automated replenishment orders — typical forecasts hit within 1–2 units of target (e.g., 30 → 28–29), cutting stockouts and excess inventory.
AI-written summary · about
chapters (11)
- 0:00Intro: Transforming Inventory Forecasting with Machine Learning
- 1:11Why Inventory Forecasting Matters (and What We’ll Cover)
- 2:22Why Forecasting Fails: Volatility, Manual Planning & Common Barriers
- 4:09Target Reference Architecture: Data → Models → Automated Decisions
- 5:21Data Ingestion & Governance on Azure (Pipelines, Lake, Quality)
- 6:55Feature Engineering: Building Predictive Signals (Seasonality, Promo, Weather)
- 8:11Model Validation: Backtesting, Accuracy Metrics, Bias & Fairness
- 9:39Deployment & MLOps: CI/CD, Model Registry, Monitoring
- 11:21From Prediction to Ordering: How the Model Drives Replenishment
- 12:06Business Impact + Real Scenarios (Healthcare & Logistics)
- 15:22Wrap-Up: Key Takeaways and Closing
related talks
35mBuilding a successful inventory planning solution with Amazon Forecast
26mRiddles in Supply Chain Management - How AI solves them?
37mForecasting using machine learning methods
32mProductionize Machine Learning on Azure
27mManufacturing optimisation & planning using ML platform
28mDemand Forecasting: Accuracy, Stability, Automation
28mHow Human Expertise and AI drive Scalable Forecasting Success at
32mAutomating Production Production Level Machine Learning on AWS