Tandem 2

Prof. Dr. Gunther Piller + Univ.-Prof. Dr. Franz Rothlauf Matthias Brunnbauer + Marc de Zoeten In 2006, Netflix announced an open competition aimed at improving its in-house movie recommendation system. The external team that succeeded in improving the accuracy of the recommendations by 10% compared to the state-of-the-art at the time would be rewarded with a prize of $1 million. With more than 40,000 participating teams, the response was overwhelming, and in 2009, a corresponding method was published and the prize money was awarded (Netflix, 2009). In retrospect, the Netflix competition was one of the most important milestones for the application of artificial intelligence methods (Chen, Chiang & Storey, 2012), which use observational data from companies (such as video views and user usage data, location data from mobile devices, and machine statuses) and utilize this data to improve business decisions (e.g., for recommendation systems for video providers or e-commerce, optimization of product offerings, or internal business processes). The availability of such data has increased dramatically with the growing prevalence of mobile devices and the interconnection of users and physical components. The application of these methods to decision-making problems in businesses forms the basis for current trends known by terms such as Big Data, Advanced Analytics, Deep Learning, Industry 4.0, etc. (e.g., McKinsey Global Institute, 2016; Gartner, 2017). A relevant area of artificial intelligence (which explicitly includes machine learning methods) deals with problems involving the recognition of patterns and structures in data and the (operational) decisions based on them. In recent years, both classical learning and optimization algorithms have been further developed, and new methods have been devised (see, e.g., Jordan & Mitchell, 2015). Examples include lifelong learning algorithms and deep learning concepts (e.g., LeCun, Bengio & Hinton, 2015; Liu, 2017; Liu et al., 2017). Consequently, the number of possible and usable methodological approaches for identifying structures in data is increasing significantly (Schmidhuber, 2015). A significant increase in the diversity of methods can also be observed in the field of optimization techniques, which find effective solutions to operational decision-making problems (Bottou, Curtis & Nocedal, 2018; Gendreau & Potvin, 2010). Although the volume of available observational data has increased sharply in recent years in companies and public institutions, and many organizations would like to use this data to improve decision-making at the operational, tactical, or strategic levels, they are often overwhelmed by the use of such methods (e.g., Günther, Mehrizi, Huysman & Feldberg, 2017). First, many organizations face the problem that it is unclear in which areas and for which decision-making problems the use of AI-based methods is both possible and appropriate (Salminen, Milenković & Jansen, 2017; Carlsson, 2018). Companies and public institutions often lack a sound assessment of which decision-making problems could be improved and what type and quality of data would be required for this. Second, once an organization has identified a relevant decision-making problem, selecting suitable methods is often difficult. In the scientific literature, practitioners are often confronted with a vast array of different approaches and must choose between them without having a detailed understanding of each one (Kotthoff, 2016; Rothlauf, 2011). The main reason for these two challenges is that only a very small portion of research in the field of artificial intelligence addresses the business perspective. Most research focuses on further developing methods to solve existing (standard) problems (see, e.g., LeCun, Bengio & Hinton, 2015; Najafabadi et al., 2015; Russell & Norvig, 2016) and leaves the application of these methods to organizations with extensive methodological and implementation expertise. Typical companies possessing such expertise are found among technology-oriented IT firms. In contrast, many companies and public institutions outside the IT sector, as well as small and medium-sized enterprises, generally have only limited knowledge of how to apply these methods (e.g., Derwisch & Iffert, 2017). This gives rise to the following research question: Which decision-making problems in companies and public institutions can be effectively supported by the use of AI? The central planned outcome is a catalog of standard decision-making problems in companies and public institutions for which AI methods are appropriate and create added value. An accompanying taxonomy will describe which of the possible approaches are appropriate for which problems in specific business or institutional contexts, and what requirements must be met for their use. Since the performance of these methods is subject to constant change due to technological progress, their suitability for specific decision-making problems also evolves. This dynamic will be illustrated by describing the performance limits of current methods and evaluating them from a technological perspective. Knowing the appropriate areas of application for artificial intelligence methods is a necessary prerequisite for their successful deployment. However, fully realizing the potential of complex analytical methods and tools in businesses and public institutions requires more. Findings from behavioral research on digital transformation efforts show that organizations respond to IT-driven innovations through a variety of activities. Current research includes, among other things (see, e.g., Vial, 2019): analyses of disruptive changes in organizational ecosystems; the development of strategic transformation strategies; investigations into how data-driven services can be used to change business models, value creation processes, products, and services; the establishment of data and technology platforms; and structural change processes to adapt organizational structures to new opportunities. Approaches to digital transformation projects in companies have been examined in several studies to date, mostly case studies (Vial, 2019). However, a systematic analysis of the success factors for introducing new AI-based methods, as well as a constructivist design of methods and reference processes for successfully implementing such projects, is still lacking. Both of these areas will therefore be investigated within the framework of the Research College. The corresponding research question is: How can companies and public institutions successfully introduce and utilize methods and tools from the field of AI? The planned key outcome is a set of reference processes for the introduction of tailored AI methods. These will include, among other things: technical, data-specific, and organizational prerequisites; necessary measures for building the relevant capabilities; methodological building blocks for the step-by-step testing and integration of AI methods into organizations, as well as for the resulting transformation processes required for business models and processes, products, or services.