Artificial intelligence (AI) is high on the political agenda in many countries. In recent years, several countries have implemented national AI strategies with the purpose of taking the lead in developing and using AI (OECD, 2021:1). However, the emergence and expansion of AI have raised questions on the topics of structural change and economic growth.

The purpose of this report is twofold. First, to investigate the characteristics of Swedish firms that use AI, and which factors that affect the possibility of firms to implement AI into their operation. Second, to analyze if there is a relationship between AI usage and firm-level productivity. Thereby, we want to contribute with new knowledge on the importance of AI for the growth of the Swedish business sector. This report is part of a large project that has the general purpose of increasing knowledge within the field of AI.

This report utilizes microdata on Swedish firms and regression analyses to empirically assess the characteristics of firms that use AI and the relationship between AI usage and firm-level productivity. Our primary data source is the survey “ICT usage in enterprises 2020”, which is provided by Statistics Sweden and contains information about AI usage among a randomized sample of firms that have a minimum of 10 employees. Estimating a linear probability model, we identify firm-level factors that have a statistically significant relationship with the likelihood of AI usage. To quantify the link between AI usage in 2019 and firm-level productivity in 2020 (value added per employee), we estimate a regression model that accounts for, e.g., firms’ lagged productivity levels.

The main findings of this report are:

• The probability of AI usage increases with firm size (number of employees). Our results indicate that large firms (minimum 250 employees) have a probability of AI usage that is 24 percentage point higher compared to small firms (10-49 employees).
• Estimates for each sector indicate that firms within the ICT sector are most likely to use AI. Corresponding estimates for each region indicate that firms that operate in the vicinity of Stockholm are more likely to use AI compared to some other regions.
• The probability of AI usage seems to increase with a firm’s wage and education level, respectively.
• Our main model accounts for firms’ 2014 productivity levels. It indicates that firms with AI usage in 2019 have a productivity that is 7 percent higher in 2020 compared to firms without AI usage.
• In the main model, the positive link between AI usage and productivity is explained by firms having a high productivity in 2019, i.e., parallel to their AI usage. We find no positive and statistically significant link among firms having a low or medium-high productivity level in 2019.
• To account for a potentially reverse causality between AI usage and firm-level productivity, we include control variables for firms’ lagged productivity levels in other years than 2014. The link between AI usage and productivity remains positive and statistically significant when we account for lagged productivity in 2015-2018.

In summary, we identify multiple factors that have a statistically significant relationship with the probability of AI usage. Firms that use AI appear to be characterized by a large number of employees, and by high wage and education levels. Furthermore, AI usage among Swedish firms does also appear to be most prominent within the ICT sector and the area of Stockholm. Considering AI usage and firm-level productivity, we find a positive and statistically significant link between AI usage in 2019 and firms’ productivity levels in 2020. This positive relationship remains statistically significant when we account for a number of factors, including lagged firm-level productivity in 2014-2018. Also, our findings indicate that the positive relationship is more prominent for firms that adopt AI in multiple usage areas.

An empirical challenge is the fact that we only have access to data covering firms’ AI usage in the year of 2019 specifically. Thus, we are not able to identify the extent nor the duration of the AI usage. Consequently, we are unable to identify whether the productivity changes when a firm begins to adopt AI. In turn, this means that we are unable to sort out whether AI usage results in higher productivity, or vice versa. However, the positive link between AI usage and firm-level productivity remains statistically significant when we account for firms’ lagged productivity levels.

Our analysis only includes Swedish firms, but the findings are in line with previous studies, regarding both a positive relationship between AI usage and firm-level productivity (Cznarnitzki et al., 2022; OECD, 2022) and the characteristics of firms that use AI (OECD, 2022).

It should be highlighted that even though the analysis is based on a randomized sample of firms, our results are not necessarily applicable for the entire population of Swedish firms. Firms that have filled out the questionnaire may differ from those firms that did not. However, the response rate regarding the questions related to AI usage is high – approx. 80 percent.

Even though our results should be considered indicative rather than causal, this report – to the best of our knowledge – constitutes the first study with the purpose of empirically assessing the link between AI usage and firm-level productivity in Sweden. Internationally, the previous research does also seem to be relatively sparse. To facilitate the opportunity for future studies to estimate causal effects and provide important knowledge on this topic, more comprehensive and detailed data on firms’ AI usage needs to be collected. For instance, information regarding when firms started using AI, how comprehensive the usage is, and the firms’ purpose of using AI are key.