E-commerce companies naturally need to cope with lots of data. Besides being used for the actual purchase, this data is also helpful for offering diverse (personalized) services like pinpoint search, chat bots, recommendations and so on. Due to the good results, quite a few of these services are based on AI systems that are known to be data and CPU power – and therefore also energy – hungry. Using as much data as possible seems to be helpful and is often said to optimize results until reaching a plateau, but did you ever check? In this video we will share the astonishing result that our experiments showed when purposefully applying the concept of data minimalism to a recommender system. In addition to optimizing results with using less data, computing time can also be decreased by 80%.

Dr. Niklas Pietsch studied physics and philosophy at the University of Hamburg. During his PhD studies at the European Organization for Nuclear Research CERN he worked on big data analysis, predictive modelling and searches for new elementary particles. For the past five years he has been developing and implementing machine learning applications for clients from the retail, e-commerce, and logistics industry. As Data Scientist at the Otto Group Solution Provider (OSP) GmbH Dr. Pietsch is also active in the research and development of data-minimalistic Artificial Intelligence.