Machine learning algorithms improve demand forecasts for groceriesback to overview
INFORM and PROGNOSIX start cooperation in sales planning for the food trade industry.
INFORM, specialist for optimization systems from Germany, and PROGNOSIX, the Swiss market leader in the field of factor-based sales forecasts, become partners in the field of sales planning for the retail food industry. This cooperation logically follows the collaborative research project in which the companies dealt with the influence of external factors on sales forecasts, particularly for fresh food. The software, PROGNOSIX Demand, which was developed as a result of this project, is now available.
The retail food industry is required to meet high standards with regard to quality and availability. However, the minimization of related costs represents a conflicting priority in this matter. Accurate forecasting methods for product sales can help significantly reduce these costs and thereby achieve higher profits. The aim of the research project "Comprehensive Sales Forecasting," sponsored by the Swiss Federal Commission for Technology and Innovation (CTI), was to identify the influence of various external factors such as weather, holidays and product similarities on the sales of fresh food and to integrate these findings into mathematical algorithms. Zurich University of Applied Sciences (ZHAW), the optimization specialist INFORM from Germany, the Swiss start-up PROGNOSIX and a variety of Swiss food retailers, all collaborated on this project.
A new generation of machine-learning algorithms for efficient and transparent support of human expertise was developed during the project. The case studies carried out with a software prototype show that, thanks to the inclusion of external factors, the new algorithms can predict the actual sales much more precisely than is possible with classical approaches (for example the time series analysis). For example, if the current warm weather is taken into account, instead of solely looking at the sales figures from the previous year, the sales of a number of weather-sensitive products (e.g. melons) can be more accurately forecasted. Food retailers can therefore work with much smaller stocks and subsequently significantly reduce food waste. These methods, combined with the possibility to avoid unnecessary costs resulting from lost opportunities, allow for a total increase in profits of up to five percent with very little effort. Thanks to the transformation of the project's prototype into a market-ready product, the retail industry can now achieve its full potential with the help of the PROGNOSIX Demand software.
Man and machine: working in harmony
Unlike with other machine-learning approaches, user-centering is one of the core ideas of PROGNOSIX Demand. "The interaction between man and machine, particularly in the case of a poor data basis, quickly provides users with valuable forecast-relevant insights," says Dr. Peter Kauf, CEO of PROGNOSIX. "It would just be fundamentally wrong to completely ignore the expertise that the planners gained over the years. Moreover, in case of purely machine-oriented approaches (e.g. neural networks), it takes a long time for the algorithms to achieve the desired quality so that they can finally be implemented in a company; not to mention the time it takes for an experienced user to build up the trust in a black box solution." The combined approach has a further advantage: Apart from a knowledge transfer from user to machine, a transfer of know-how from machine back to the user also takes place. "Supported by the transparent and objective feedback from the forecast algorithms, experienced planners can bring their intuition to a whole new level. At the same time, their young colleagues can benefit from the centralized information and quickly find their way and navigate through highly complex tasks. Last but not least, since the forecast results are comprehensible, the software achieves high approval rates among the employees," adds Kauf.
A strong partnership
"We believe this approach to be very promising for food retail and that a product such as PROGNOSIX Demand meets a real need in the industry," comments Peter Frerichs, VP Inventory & Supply Chain Division at INFORM, on the cooperation with PROGNOSIX. A solid basis for the optimization of the algorithms in the CTI research project was the add*ONE software from INFORM, which is already used for inventory optimization and sales planning by various well-known food manufacturers such as Deutsche See, Coppenrath & Wiese and Capri Sun. The young company, PROGNOSIX, hopes that the cooperation with a strong established partner in Germany will promote their successful entry into the international market. "Both companies have a university background, and INFORM also aims for the user-centered approach - just as we do. It was, therefore, quite natural that in the course of the research project we established connections that led to further cooperation," explains Kauf.
PROGNOSIX Demand can be integrated into any ERP system and flexibly adapted to the individual requirements of food retailers. The software is web-based and therefore allows for a quick and cost-effective implementation - either as a highly secure cloud version or as an on-premise solution.
About PrognosiX AG
PROGNOSIX AG was founded in July 2014 as a spin-off company of the ZHAW Zurich University of Applied Sciences. Their core competencies lie in the fields of mathematical forecasts (predictive analytics), forecast systems, data analysis and implementation of forecast systems. In its applications, PROGNOSIX builds upon the idea of people as competent decision-makers in a complex environment. By cooperating with selected competence centers at universities and development companies, PROGNOSIX is able to draw on a broad range of research and development resources. This creates a holistic view of our customers' innovation initiatives and a sustainable implementation of the projects and products defined by them. For more information on PROGNOSIX, please visit www.prognosix.ch
Do you have questions, comments or require more information? Please give us your feedback!