The bakery industry currently accounts for 31.5% of retail food waste in The Netherlands, according to a study by Wageningen University. One of the major challenges for bakeries is how to predict consumer demand so they can produce only what is needed at the right time and in the right place.
This is why Fuite Bakery initiated its ‘demand forecasting and smart processing’ project to reduce bakery waste by 5% in the retail sector – both from traditional and online outlets. Applied to all bakeries in The Netherlands, it has the potential to cut waste by nearly 25 million breads a year – saving 4 million kWh, 1.5 million m3 of gas and at least a million transport kilometres.
As online sales continue to grow, there is a pressing need to introduce a smart bakery system than can match supply and demand even more precisely.
Fuite Bakery is drawing on the expert knowledge of Silowacht BV to develop the optimum automation system for this purpose. Online retailer Picnic Technologies and supermarket chain Boni Markten are supporting the project by providing their product demand data.
Final report summary
Fuite Bakery used data from 2015 to 2020 to develop a first prototype of the demand forecast model. Due to changes in customer purchasing behaviour during COVID-19, additional data was collected to ensure the model’s validity. The model was also developed to account for the three types of bakery products identified: daily products such as bread; special products such as cake; and special occasion products such as Valentine’s cake.
Simulations were run in a realistic environment to test and validate the digital technologies. Silowacht was responsible for connecting the forecasting model with production/process sensors.
The outcome of this project was better than expected. Of the three product categories, daily products were easiest to forecast with an accuracy of +-2.5%. The overall accuracy across all three product types was +-3%.
Simulations also showed that waste can be reduced by 16%, which is 6% more than expected. Fuite now hopes to validate the model in an operational environment. If the method can be connected to all machines in the bakery, it will be possible to achieve close to real-time predictions of supply and demand.