D5.8 – Report on Demonstration – Mitakus

This deliverable reports on how the innovation Mitakus, provided by the German company Mitakus Analytics UG, which aims at reducing food waste in the food service sector, was applied in the LOWINFOOD project. Mitakus uses sales data provided by the users and combines this with external factors such as the weather or holidays to calculate a sales forecast. This can help users plan their production more accurately. Food waste can be reduced if the users avoid producing surpluses by utilizing the Mitakus forecast. In Germany, two users tested the Mitakus innovation as part of the LOWINFOOD project. Both users are university student canteens of differing sizes. This report presents the methodological approach to testing. The results of an initial food waste measurement in the businesses, an assessment of the forecasts’ performance, and the results of a survey on user-friendliness and potential improvements are presented. The innovation was highly affected by the COVID pandemic. Thus, the demonstration of Mitakus faced several challenges, such as acquiring businesses for testing or providing usable historical sales figures to create the forecasts. The opportunities offered by Mitakus included, for example, the company's responsiveness to customer requirements.The waste measurement in the businesses discovered that 46 and 41 grams of food were initially wasted per serving including plate, serving, and preparation waste. The user survey revealed that the implementation of the app was initially perceived as somewhat difficult. In contrast, use by management was described as rather simple. The dashboard and its features also received a good rating. The performance of Mitakus' short-term forecasts is better than those predicted by the users themselves. In contrast, the quality of the long-term forecast depends on the circumstances in the company and is strongly influenced by short-term changes to the menu.The current status of testing is that Mitakus is continuously improving the prediction models created.