Study uncovers three fundamental mechanisms and their necessary conditions that enable machine learning to deliver different value propositions over time.
Advances in artificial intelligence (AI) are changing the competitive landscape. In search of an appropriate strategic response, companies are currently engaging in a variety of AI projects. However, companies are struggling to create tangible business value through AI. In a study, researchers from Copenhagen Business School, the Universities of Bamberg and Paderborn, and the IT University of Copenhagen examined how companies can create sustainable value through applications of machine learning (ML) - a core AI technology.
Study helps better target and monitor AI initiatives
During the four-year exploratory qualitative study, the researchers looked at 56 operational ML applications at 29 companies. To do so, they conducted 40 interviews with data scientists and executives involved in developing ML applications. In the process, they uncovered three fundamental mechanisms by which ML can add value to the enterprise. "Each mechanism requires certain circumstances to be used successfully, but also brings a unique value contribution to the company," explains Dr. Konstantin Hopf from the Chair of Business Informatics, in particular Energy Efficient Systems, at the University of Bamberg. "The systematic comparison of the three types of ML use and the respective necessary conditions helps companies to plan and monitor their AI initiatives in a more targeted manner," explains Prof. Arisa Shollo from Copenhagen Business School, who initiated the study. The study was recently recognized as the best paper in the Journal of Stra-tegic Information Systems 2022.
Three mechanisms and selected conditions for creating enterprise value through ML:
Knowledge creation: The first mechanism that the researchers were able to uncover is the use of ML for knowledge generation. This means that companies use algorithms to recognize patterns in data and thus gain new insights into their business field or customer behavior, for example. For this kind of value creation, companies need data science skills and domain knowledge in particular.
Supporting people in decision-making: In the second mechanism, companies use ML to assist or guide people in performing tasks. Applications range from small extensions of existing software programs to specialized systems that support decisions or provide specific recommendations for action. A good data infrastructure and expertise in the design of user interfaces, for example, are essential for this type of value creation.
Autonomous agents: In the third type of value creation, ML is either directly integrated into new products or services and thus made available to the end customer, or parts of business processes are executed by AI itself instead of just supporting them. To realize such deployment scenarios, companies depend on the integration of ML applications into operational processes and IT systems, a stable business environment, and few legal and ethical issues with their ML applications.
Value creation mechanisms change over time
During the analysis, the researchers found that companies change the value creation mechanism of ML projects over time. This can be illustrated by the example of a jewelry retailer that had to stop its fully automated ad placement in online media at the beginning of the Covid 19 pandemic due to very poor advertising performance. The company reverted to human decisions about ad placement and then began using ML to gather knowledge about the new situation. The knowledge gained was then used to create automated dashboards to support routine human decision making and, finally, the company trained new predictive models tailored to the pandemic situation. The jewelry retailer returned to the old mode of ML-based process automation. "Knowing that ML value creation mechanisms can change over time allows companies to take an evolutionary perspective on their AI initiatives. So, if the envisioned value of a single project has not been met, there may be opportunities for further development of the AI initiative. These should be exploited instead of considering a project a failure too early," says Konstantin Hopf.
"Many companies' ML projects fail because the company's resources and capabilities or the environmental conditions do not match the selected ML mechanisms. In their AI activities, for example, companies often overlook the fact that the various ML mechanisms build on each other and that upstream mechanisms can create the necessary resources and conditions for more extensive ML projects. It is therefore advisable for managers and data scientists to keep an eye on the capabilities and limitations of the technologies as well as the conditions of the company and to dynamically adapt the AI strategy to them," says Prof. Dr. Oliver Müller, Chair of Information Systems, especially Data Analytics at Paderborn University, providing further insights into the findings of the study.
The article, which was named "Best Pa-per 2022" by "The Journal of Strategic Information Systems", is freely available online at: https://doi.org/10.1016/j.jsis.2022.101734.
This article has been translated automatically.
