Can Databricks Genie and Amazon Quick actually automate retail replenishment?

Most retail planners are stuck in a loop of exporting demand forecasts and manually checking supplier stock, which usually means the fastest-moving SKUs are sold out before the order even hits the warehouse. The problem isn't the forecast anymore—foundation models have mostly solved that—it's the gap between the data platform and the actual purchase order. I've been looking at a setup that closes this loop by connecting Databricks and Amazon Quick to handle the "detect, decide, act" sequence without a human in the middle unless a rule breaks.

The architecture relies on four specific stages to move from a prediction to a purchase order.

How the replenishment loop functions

The process starts in Databricks with Many Model Forecasting (MMF) using Chronos-2. This handles the 7-day demand prediction for every SKU in the catalog. Once that's done, the system moves into a scheduled loop managed by Amazon Quick.

1. Detection: A Databricks Genie Agent identifies "surges." To keep out low-volume noise, the system only flags SKUs where the prior-14-day average is at least 1, and the next-7-day average is at least 1.5 times that prior average.
2. Decision: Amazon Quick checks these surging SKUs against live supplier availability stored in Amazon S3 Tables. It's programmed to pick the cheapest supplier capable of covering the demand.
3. Action: Amazon Quick Flows either triggers a purchase order via a Supplier Order API or, if no single supplier can fulfill the surge, it escalates the issue to a human review ticket.

Technical implementation details

If you're trying to reproduce this, you'll need the fresh_retail_net example to generate the forecast data. The logic for the Genie Agent to surface surges is the critical part of the filter.

To get the Genie Agent to identify the specific SKUs that meet the surge criteria, you would use a prompt similar to this:

Identify all SKUs where the average predicted demand for the next 7 days is at least 1.5 times the average demand of the prior 14 days, provided that the prior 14-day average is 1 or greater. Return the SKU IDs and the calculated surge ratio.

The integration relies on a Genie MCP connector to pull the forecast into Amazon Quick, while Amazon Quick Sight handles the supplier feed from Amazon S3 Tables as a live Direct Query dataset.

This setup essentially treats the forecast as a trigger rather than just a report. By the time a human planner would have noticed a trend in a spreadsheet, Amazon Quick Flows has already queried the S3 Tables and hit the Supplier Order API. The only risk here is the "floor" value; if you set that prior-14-day average too low, you'll get a flood of false positives from low-volume items. 1.0 seems to be the baseline for keeping the noise manageable.