Data Department built very short-term sales forecasting for a quick-service restaurant chain, so kitchens could prepare food ahead of demand instead of reacting to each order. Order waiting times became about 30% shorter on average. The result is an average, so individual orders and periods vary around it.
What decision were we asked to support?
In a quick-service kitchen, one decision repeats all day: what should we prepare right now?
Cook to order and nothing is wasted, but every customer waits for the full preparation time. Cook ahead and food is ready when the order arrives, but anything that does not sell is thrown away. Kitchen staff make this call constantly, usually by eye and by habit.
The chain wanted customers to wait less. That meant preparing the right items shortly before they were ordered, which requires knowing what will sell in the next few moments, not just today or this week.
How did the approach work?
We built a very short-term sales forecast: a prediction of what will be ordered in the near future, used as a signal for what the kitchen should prepare ahead.
A very short-term forecast is a different tool from the daily or weekly forecasts most businesses already have. The table below shows how the two typically differ.
| Daily or weekly forecast | Very short-term forecast | |
|---|---|---|
| Decision it supports | Ordering stock, planning rotas | What to prepare now |
| How often it is used | Once a day or once a week | Repeatedly during service |
| What matters most | Overall volume | Timing within the service |
| Cost of a wrong forecast | Excess stock or understaffing | Waiting customers or wasted food |
In projects of this kind, the forecast is only useful if it reaches the kitchen in time and in a form staff can act on during a busy service. Speed and simplicity of the output matter as much as the model behind it.
Results
- About 30% shorter order waiting times on average.
This is an average across orders. Some orders and periods improved more and others less. We report it as an average and do not publish other figures for this project.
Lessons
Measure the customer outcome, not only forecast accuracy. The goal was shorter waits. Waiting time was the measure that mattered, so it is the result we report. We explain why accuracy alone is not enough in our demand forecasting service.
The forecast horizon follows the decision. Preparing food is a decision made minutes ahead. A forecast built for that decision has to be short-term and refreshed during service.
Cooking ahead is a trade-off, so make it explicit. Preparing more reduces waiting and increases the risk of waste. A good system makes that trade-off a deliberate choice rather than a guess at the counter.
Kitchen staff must be able to use it at speed. A prediction that arrives late or needs interpretation during a rush will be ignored.
Waiting time connects to other restaurant decisions. See our 15-minute demand forecasting work for two quick-service chains, our restaurant staffing and shift optimisation project, and the restaurants and food service page.
Frequently asked questions
What is very short-term sales forecasting?
It predicts what customers will order in the near future, during the same service. In a kitchen, it tells staff what to prepare ahead so food is ready when orders arrive.
Why is the result reported as an average?
Waiting times differ from order to order and from one period to another. The about 30% reduction is the average across orders, so some improved more and some less.
Does preparing food ahead increase waste?
It can, which is why the forecast matters. The aim of this kind of system is to prepare ahead only what is likely to sell soon, so customers wait less without cooking blindly.
Is this the same as the 15-minute slot forecasting work?
No. This was a separate project, focused on preparing food ahead of demand to shorten waiting times.