Discovering potential founders within academic institutions

Abstract

Technology transfer is central to the development of an iconic entrepreneurial university. To foster knowledge transfer, many universities undergo a scouting process by their innovation coaches. The goal is to find staff members and students, who have the knowledge, expertise, and the potential to found startups by transforming their research results into a product. Since there is no systematic approach to measure the innovation potential of university members based on their academic activities, the scouting process is typically subjective and relies heavily on the experience of the innovation coaches. In this paper, we study the discovery of potential founders to support the scouting process using a data-driven approach. We create a novel data set by integrating the founder profiles with the academic activities from 8 universities across 5 countries. We explain the process of data integration as well as feature engineering. By applying machine learning methods, we investigate the classification accuracy of founders based on their academic background. Our analysis shows that using a random forest (RF), it is possible to differentiate founders and non-founders with an average accuracy of 79%. This accuracy remains mostly stable when applying an RF trained on one university to another, suggesting the existence of a generic founder profile. The detailed analysis indicates a high significance of the career path as well as patent- and grant-related features among others. Furthermore, we show that using a RF, it is possible to exploit these features to predict the future founding probability up to 3 years in advance with an accuracy of 80%. Finally, by analyzing the academic disciplines of founders we show that the patent documents have more influence on the startup’s core orientation than the publications.