Research Policy

• Using Artificial Intelligence (AI) technologies is associated with significantly higher product and process innovation results. • The strongest link between AI and innovation output is for world-first innovation. • Applying AI broadly (using different methods for different applications areas) increases innovation output. • Firms with a longer experience in using AI yield higher innovation results. • In 2019, only 5.8% of all firms in Germany were actively using AI. The contribution of AI-based innovation to total innovation output is significantly higher for world-first innovation (18% of total world-first sales). • AI-using firms more often report innovations that exceeded expectations, but less frequently innovations that did only partially or not at all meet the firm's expectations. This paper analyses the link between the use of Artificial Intelligence (AI) and innovation performance in firms. Based on firm-level data from the German part of the Community Innovation Survey (CIS) 2018, we examine the role of different AI methods and application areas in innovation. The results show that 5.8% of firms in Germany were actively using AI in their business operations or products and services in 2019. We find that the use of AI is associated with annual sales with world-first product innovations in these firms of about €16 billion (i.e. 18% of total annual sales of world-first innovations). In addition, AI technologies have been used in process innovation that contributed to about 6% of total annual cost savings of the German business sector. Firms that apply AI broadly (using different methods for different applications areas) and that have already several years of experience in using AI obtain significantly higher innovation results. These positive findings on the role of AI for innovation have to be interpreted with caution as they refer to a specific country (Germany) in a situation where AI started to diffuse rapidly.

Artificial intelligence and industrial innovation: Evidence from German firm-level data