Data Department built demand forecasting and delivery-schedule optimisation for a quick-service restaurant chain. The forecasts estimated what each restaurant would need; the optimiser decided when deliveries should arrive to meet that need at the lowest combined cost. Together they lowered inventory and logistics cost by about 10%.
What decision were we asked to support?
A restaurant supply chain makes two linked decisions over and over. How much does each restaurant need? And when should it be delivered?
The two pull against each other. Frequent, small deliveries keep stock low and food fresh, but every trip costs money. Fewer, larger deliveries save on transport, but they tie up stock, fill limited storage space in the restaurant and raise the risk of perishable items expiring.
Planning stock and planning deliveries separately tends to optimise one cost at the expense of the other. The question was how to plan both together.
How did the approach work?
The project combined two pieces. A forecast estimated each restaurant's demand. A delivery-schedule optimiser then used that forecast to decide when to deliver, balancing the cost of holding stock against the cost of each delivery.
In projects of this kind, the optimiser needs more than the forecast. A typical checklist of inputs looks like this:
- [ ] Consumption or sales history per restaurant and item
- [ ] Shelf life of each item, especially perishables
- [ ] Storage capacity at each restaurant (dry, chilled, frozen)
- [ ] Delivery constraints: vehicle capacity, routes and the hours restaurants can receive goods
- [ ] The cost of a delivery and the cost of holding stock
- [ ] Rules the operation will not break, such as minimum stock of key items
The list is general. Every operation has its own constraints, and an early part of any such project is writing them down with the people who run deliveries every day.
Results
- About 10% lower inventory and logistics cost.
The result is measured on the combined cost of holding stock and making deliveries. That is the number the business pays, so it is the number we optimised.
Lessons
Optimise the combined cost, not each piece. Lower stock that requires twice as many trips is not a saving. Treating inventory and logistics as one problem is where the result came from.
A forecast is an input, not the end product. The forecast mattered because it fed the schedule. We judge forecasting work by the decision it improves, a point we make in more detail in our demand forecasting and inventory optimisation service.
The schedule has to work in a real restaurant. A delivery that arrives during the lunch rush, or that will not fit in the cold room, creates new problems. Practical constraints belong in the model, not in a footnote.
Start with the constraints people already know. Delivery teams and restaurant managers know the hard limits. Capturing them early avoids rework later.
This project sits alongside our other restaurant work: 15-minute demand forecasting for two quick-service chains, and the wider picture on our restaurants and food service page. The scheduling side uses the same methods as our optimisation and simulation work.
Frequently asked questions
What is delivery-schedule optimisation?
It is a method that decides when, and how often, each location should receive deliveries. It balances the cost of each delivery against the cost of holding stock, within limits such as storage space and receiving hours.
Why combine inventory cost and logistics cost?
Because they move in opposite directions. Reducing one on its own usually raises the other, so the only honest measure is the combined cost.
Is this the same project as the restaurant stock forecasting work?
No. This was a separate project. The stock forecasting work is described on our restaurant demand forecasting page.