Tracing a food cost overrun

The scenario and numbers in this demonstration project are simulated data.

Industry: Restaurant chain
What was built: Agents, skills, semantic models and write tools, built on demonstration data
Where the knowledge came from: Demonstration data from the finance, POS and SCM systems

Root cause before the board

When the month-end close finishes, the finance system raises an alert: some stores' food cost rates are above the group's alert line.
The CFO needs a root cause analysis before the board meeting. Finding the cause takes a comparison across the month-end figures, SCM purchasing records and actual usage from the POS.

This demonstration project has semantic models on the finance, POS and SCM systems.
The finance semantic model defines the food cost rate as food cost divided by food and beverage revenue, and shows the gap to the group's standard rate.
The agents use the semantic models to check each store's monthly food cost rate, find the stores over the line, and list the other stores in the same month as a comparison group.
The agents break out how much each store is over the standard, and find these stores' main suppliers and main ingredient categories.
The agents calculate actual usage as SCM receipts minus the month-end stock count. They then compare it with the theoretical usage from the POS standard BOM, and find the items used above standard.
Example: a buffet store uses more rice than the standard. The reason is that guests serve themselves and portions are not counted.
The agents check suppliers' price history for the last 6 months in the SCM, and confirm which ingredients went up.
The agents split each store's overrun into the share from supplier prices and the share from kitchen practice and waste.

The agents draft an improvement plan: recipe correction orders for the stores, plus a proposal to renegotiate a supplier contract.
Before the recipe correction orders and the supplier renegotiation are written to the POS and SCM, the agents stop for approval. The purchasing manager authorises the renegotiation.

At restaurant groups like this, recipes and supplier prices will keep being revised.
The agents will collect feedback on their answers from users. The feedback will go to Odin.
Wrong answers will become tests. A change will be kept only if scores improve.

Get started

AI feasibility review of a live workflow