low level of reliability of AI tools is mitigated by applying bias reduction techniques or improving the tool based on the employee’s feedback.
Both the benefits of working together with AI and the challenges related to the collaboration are the foundation of answering the research question. AI supports the perceived team productivity by working together with employees. This collaboration contributes to the perceived team productivity if provided that the challenges related to collaboration are mitigated. The contribution can have multiped forms: focused on quality improvement, improvement of the job experience, and the reduction of process time. Factors benefiting the collaboration are training (non-technological factor) and the low usage threshold of an AI tool (technological factor).
5.2. Discussion
The findings of this research are partly aligned with the existing literature. The positive contribution of AI to the perceived team productivity through collaboration matches the expectations based on the paper of Rust and Huang (2012). Further, the different AI applications in the banking sector, as described by Hassani et al. (2018), are reflected in the case studies. The multiple ways an AI tool assists employees in the different case studies correspond to the idea of Wilson and Daugherty (2018) that AI could offer many different types of assistance. Also, the strength of AI in processing large amounts of data mentioned in the article of Du Croo De Jongh et al. (2018) is present in the case studies. Another link to the existing literature is workplace innovation. The assistance of AI leading to an improved job experience is a good example of AI’s role as supportive technology in workplace innovation (Oeij et al., 2012). The challenges of resistance and reliability - as mentioned in the papers of Huang et al. (2019), Ali et al. (2016), and Fountaine et al. (2019) – also occurring at the different banks. The fear of replacement is also present in the case studies of ING and Rabobank. This observation matches the findings in the research of Liu and Zhan (2020) regarding the feeling
of job insecurity among employees. The mitigation action of using training to convince employees of the benefits of working with AI is corresponding with the advice of Amershi et al. (2019). The advice of Amershi et al. (2019) stresses the importance of convincing the employees of the possibilities of AI tools. Also, the advice of Amershi et al. (2019) about providing feedback on the collaboration frequently is reflected in the case of ABN AMRO. The importance of training addressed by Kolbjørnsrud et al. (2017) is supported by all the three banks participating in this research.
However, some findings differ from the observations in the various academic papers.
For example, the observation about the relation between the type of setup and the complexity of work differs from the relation described in the paper of the Deutsche Bank (2019). In this paper, the researcher describes customer contact as the place for AI to take over tasks. The more complex jobs are more suitable for working together on the same tasks. Another finding that differs from the literature is the low usage threshold. The paper of Kolbjørnsrud et al. (2017) stresses the importance of training but is not discussing the alternative of a low usage threshold.
The main theoretical contribution of this research project on the productivity of human-AI teams in the banking sector is to respond to the call for research on collaboration in specific fields (Seeber et al., 2020) and the request for more research covering both the technological and social aspects of human-AI collaboration (Dahlin, 2021). This research adds one more collaboration research in a specific field, the Dutch banking sector, to the existing AI collaboration literature. The request of Dahlin (2021) is reflected in the challenges. For instance, the challenge of resistance is a social aspect, and the low level of reliability is a technological aspect. Further, the findings of this research can be used as starting point for further research on the ways AI can assist. In this research, three different ways AI can assist employees are observed: increasing the (perceived) quality of their work, reducing the time, of processes or improving the employee's job experience. A next step could be to focus more on quantitative contribution to (team) productivity of the different types of AI assistance.
The practical relevance of this research is to share insights into the possible benefits of human-AI collaboration and the related challenges. Moreover, this research shows how the three largest Dutch banks mitigate the effects of these challenges. The different approaches of the banks to the application of AI and the setup of the collaboration with the employees could be used as a best practice for other companies leveraging AI. The same applies to the ways the banks mitigate the challenges related to the collaboration. Organizations could also consider using AI as a supportive technology for workplace innovation. Finally, the findings in this research emphasize the importance of training and guiding the employees in working with AI and the supporting effect of a low usage threshold on this human-AI collaboration.
5.3. Limitations and Future Research
In this paragraph, the strengths and limitations of this research are discussed. The main strength of this research is the identification of ways AI could support team productivity by conducting case studies at the three largest banks of the Netherlands. Furthermore, the multiple case study could provide insights for other companies working with AI or companies considering implementing AI. For example, this research also shows how the banks deal with collaboration-related challenges or how the banks set up the collaboration between humans and AI.
Besides the strengths, this research also has its limitations. The qualitative measurement of team productivity is debatable. The academic field disagrees about the best way to measure productivity in a qualitative way (see 2.4). In this research, team productivity is measured as
“experienced” productivity. In the future, other researchers perhaps pick a different measurement method to measure productivity. Another limitation is the difference in the role of the interviewees. At ING, the interviewees were the employees. At Rabobank and ABN AMRO, the interviewees were developers responsible for developing and guiding/training the users. The different roles have their own vision on the collaboration between humans and AI, which difficult the comparison of the different opinions/experiences.
In the ideal situation, the different interviewees have the same role and work in a similar department. Further, there is a limitation related to the case study of Rabobank. This case study is only based on a one-hour-long spoken interview with a manager. The interview itself was very useful, but there was no possibility to cross-check the observations. The last limitation concerns the written interview. In written interviews, the interviewer cannot observe interviewees' emotions and ask immediate follow-up questions. Unfortunately, this availability of participants was restricted due to the COVID-19 guidelines and company policies. For that reason, the follow-up questions on the written interviews are partly included in the spoken interviews.
Finally, the future research directions are discussed. The first suggestion is to continue conducting research on human-AI collaboration in different fields, as requested by Seeber et al.
(2020). The next suggestion for a research direction is to conduct research on AI involving both social and technological aspects to respond to the call of Dahlin (2021). Another interesting research direction is to conduct this research again, but then a quantitative analysis. The last suggestion is to conduct multiple case studies at other Dutch banks to identify the similarities and differences within the Dutch banking sector. Future research could also focus on foreign banks instead of Dutch banks to determine the possible differences between countries.