Business Process Management (BPM) is evolving rapidly. Building on a rich research portfolio that spans from early workflow automation to robotic process automation (RPA), large language models (LLMs) are the most recent disruptive technologies to influence all BPM capability areas. Particularly, these technologies fuel the emerging research fields of AI-enhanced business process management systems and autonomous process execution. In a design science research approach, we develop a modular LLM agent architecture that effectively enables the adaptation and execution of business processes. The architecture integrates a Frame Agent that can generate process descriptions and an Operational Agent that autonomously executes processes based on the Frame Agent’s process descriptions. It further enables a future Tactical Agent for autonomous process adaptations. We demonstrate the architecture’s applicability and utility with the real-world case of a meter-to-cash process, comparing it to the performance of an RPA bot. Our findings provide early insights into the challenges faced by autonomous process execution and its relation with RPA in terms of adaptability, flexibility, and complexity in autonomous process execution. While researchers can build on our findings to establish modular LLM-agent architectures for autonomous process execution, practitioners can derive early insights into the design of LLM agents for process automation. • Properties and perspectives of AI-enhanced business process management systems. • Modular LLM agent-based system design with process frame for adaptive and autonomous process-aware execution. • Evaluation based on real-world meter-to-cash process data.

Advancements in artificial intelligence (AI) technologies are rapidly changing the competitive landscape. In the search for an appropriate strategic response, firms are currently engaging in a large variety of AI projects. However, recent studies suggest that many companies are falling short in creating tangible business value through AI. As the current scientific body of knowledge lacks empirically-grounded research studies for explaining this phenomenon, we conducted an exploratory interview study focusing on 56 applications of machine learning (ML) in 29 different companies. Through an inductive qualitative analysis, we uncover three broad types and five subtypes of ML value creation mechanisms, identify necessary but not sufficient conditions for successfully leveraging them, and observe that organizations, in their efforts to create value, dynamically shift from one ML value creation mechanism to another by reconfiguring their ML applications (i.e., the shifting practice). We synthesize these findings into a process model of ML value creation, which illustrates how organizations engage in (resource) orchestration by shifting between ML value creation mechanisms as their capabilities evolve and business conditions change. Our model provides an alternative explanation for the current high failure rate of ML projects.

Historically, the field of financial forecasting almost exclusively relied on so-called hard information - i.e., numerical data with well-defined and unambiguous meaning. Over the last few decades, however, researchers and practitioners alike have, following the advances in natural language understanding, started recognizing the benefits of integrating soft information into financial modelling. In line with the above, this paper examines whether contemporary attention-based sequence-to-sequence models, known as Transformers, can help improve stock return volatility prediction when applied to corporate annual reports. Using a publicly available benchmark dataset, we show, in an empirical analysis, that out-of-the-box Transformer models have the ability to outmatch current state-of-the-art results and, more importantly, that our proposed feature-based Transformer approach can outperform a robust numerical baseline. To the best of our knowledge, this is the first empirical study focusing on stock return volatility prediction (1) to ever experiment with state-of-the-art Transformer architectures and (2) to demonstrate that a model based solely on soft information can surpass its numerical counterpart. Furthermore, we show that by including an additional numerical feature into our best text-only model, we can push the performance of our model even further, suggesting that soft and hard information contain different predictive signals.

This essay discusses the use of big data analytics (BDA) as a strategy of enquiry for advancing information systems (IS) research. In broad terms, we understand BDA as the statistical modelling of large, diverse, and dynamic data sets of user-generated content and digital traces. BDA, as a new paradigm for utilising big data sources and advanced analytics, has already found its way into some social science disciplines. Sociology and economics are two examples that have successfully harnessed BDA for scientific enquiry. Often, BDA draws on methodologies and tools that are unfamiliar for some IS researchers (e.g., predictive modelling, natural language processing). Following the phases of a typical research process, this article is set out to dissect BDA’s challenges and promises for IS research, and illustrates them by means of an exemplary study about predicting the helpfulness of 1.3 million online customer reviews. In order to assist IS researchers in planning, executing, and interpreting their own studies, and evaluating the studies of others, we propose an initial set of guidelines for conducting rigorous BDA studies in IS.