Data Department built load forecasting and shift optimisation for premium restaurant groups. The forecasts used weather, holiday and event features to predict how busy each location would be, and an optimiser turned that forecast into shift plans. Labour cost fell by 5% to 20% at the same service level. The range reflects differences across locations, not a single average.
What decision were we asked to support?
In a premium restaurant, service is the product and labour is one of the largest costs. Too few people on the floor and guests wait, which they notice. Too many and the business pays for idle hours, every day.
Shift plans are usually built from a manager's experience and last year's calendar. That works on ordinary days. It works less well when the weather turns, a holiday falls midweek or an event nearby fills the street. The question was simple: how many people should be working, and when, so that service holds and labour cost falls?
How did the approach work?
The work had two connected parts: a forecast of the load, and an optimiser that turns that load into shifts.
| Step | What it does | What it uses |
|---|---|---|
| Load forecasting | Predicts how busy each location will be across the day | Historical demand, plus weather, holiday and event features |
| Shift optimisation | Builds the lowest-cost shift plan that still covers the forecast load | The load forecast and the agreed service level |
Why weather, holidays and events. Demand at a premium restaurant reacts to things outside the restaurant. A warm evening, a public holiday or a concert nearby can change the shape of a whole service. Features for these give the forecast a reason for unusual days, which are the days a fixed rota gets wrong.
How shift optimisation typically works. In projects of this kind, the optimiser works inside the rules managers already follow: minimum staffing, allowed shift lengths, rest periods and labour regulations. Within those rules it searches for the set of shifts with the lowest labour cost that still meets the service level. We describe the method in more detail on our optimisation and simulation page.
Results
- Labour cost 5% to 20% lower at the same service level.
- The figure is a range across locations. Some locations improved more than others, so we report the range rather than one number.
Holding the service level fixed is what makes the result meaningful. Cutting labour cost by letting service slip is easy. Cutting it while guests are served just as well is the point of the project.
Lessons
Fix the service level first, then measure cost. Every comparison in this project kept the service level constant. Without that rule, any saving could simply be a worse experience for guests.
Forecast the load the staff will feel. Staffing follows workload across the day, not a daily sales total. The forecast had to describe when the pressure arrives, not only how much is sold.
An optimiser must respect the rules managers live by. A plan that breaks labour rules or ignores minimum staffing will not be used. Building those rules into the model from the start is what makes the output a plan rather than a suggestion.
Report ranges honestly. Locations differ. A range of 5% to 20% tells a reader more than a single headline figure would.
Staffing is one of several connected decisions in a restaurant. For quick-service chains, see how we forecast demand in 15-minute slots and how very short-term forecasting cut order waiting times. More on the sector is on our restaurants and food service page.
Frequently asked questions
What does "at the same service level" mean?
It means the new shift plans were held to the same standard of service as before. The saving in labour cost did not come from making guests wait longer.
Why is the result a range and not one number?
The project covered several locations, and the saving differed between them. Labour cost was 5% to 20% lower depending on the location, so we report the full range.
Why use weather, holiday and event features in a staffing forecast?
These factors change how many guests arrive and when. Including them helps the forecast anticipate unusual days, which are the days a rota based on habit tends to get wrong.
Does the optimiser replace the restaurant manager?
No. In our work, shift optimisation proposes a plan within the rules managers already follow. Managers still make the final call, with a better starting point.