← All case studies
Restaurants / hospitalityPredictive analytics

Prep forecasting across nine kitchens

Head chefs were prepping to memory. Daily covers and prep quantities are now forecast per site, cutting waste without running short on service.

−24%
Food waste
£96k
Annual saving
0
Sell-outs on core dishes
Problem

Nine kitchens, nine different guesses

The Larder Group ran nine sites with no shared view of demand. Each head chef prepped from experience, which worked until weather, a local event or a bank holiday moved covers by 30%. Waste was written off monthly and never traced back to a decision.

Approach

Covers first, then prep quantities

We forecast covers per site per service using two years of till data, local event calendars and weather. That feeds a prep sheet per kitchen: quantities by component, not by dish, so shared prep across the menu is counted once.

Adoption

The sheet had to beat the chef, on paper, first

For three weeks the forecast printed alongside the chef own numbers and nobody was asked to follow it. Chefs marked which was closer each service. Once the forecast was winning four services in five, adoption took care of itself.

Result

Less thrown away, nothing run out of

Food waste by weight fell 24% across the group, worth roughly £96,000 a year. Core-dish sell-outs, which the chefs feared would rise, went to zero because the forecast is per component rather than per dish.

Before / after
Food waste by weight
11.4%8.7%
Prep planning time
50 min/day10 min/day
Core-dish sell-outs weekly
60
Forecast accuracy on covers
n/a±7%

I argued with it for three weeks. Then I started printing it.

TLExecutive Chef, The Larder Group