AWS, Databricks, and Amazon Quick have published a joint technical walkthrough for automating retail replenishment across a four-stage pipeline (AWS Machine Learning Blog). Databricks Many Model Forecasting (MMF) serves Amazon's Chronos-2 foundation model to predict 7-day demand for every SKU, while a Databricks Genie Agent detects demand surges—defined as a next-7-day average at least 1.5 times the prior-14-day average (AWS Machine Learning Blog). Amazon Quick then reconciles each surging SKU against live supplier availability in Amazon S3 Tables and automatically places purchase orders through a Supplier Order API or escalates to human review when no single supplier can cover the demand (AWS Machine Learning Blog).
The solution addresses a persistent bottleneck in retail operations: the gap between a good forecast and an executed order. For years, the hard part was the forecast itself, but foundation models have shifted the constraint downstream to the manual work of checking supplier availability and placing orders across tens of thousands of SKUs (AWS Machine Learning Blog). By automating this loop, commerce teams eliminate delays that cause stockouts on fast movers and reduce the need for planners to manually export, validate, and action each forecast. The solution runs on Databricks for intelligence and Amazon Quick for orchestration and action, with a companion GitHub repository providing end-to-end deployment scripts and example notebooks.
The architecture uses Amazon Quick as the integration layer, reaching the Databricks Genie Agent over the Model Context Protocol (MCP) with user-delegated OAuth, pulling the supplier feed from S3 Tables as a live Direct Query dataset, and calling the Order API through an OpenAPI connector (AWS Machine Learning Blog). Commerce practitioners can deploy the full stack using CLI scripts and console setup steps, with deployment constrained to AWS regions that support Amazon Quick's agentic capabilities.