Mahei Li
73 results
Now showing 1 - 10 of 73
- Some of the metrics are blocked by yourconsent settings
Item type:Publication, Conceptualizing hybrid intelligent service ecosystems(2025-07-08)Scopus© Citations 12 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, SynDEc: A Synthetic Data Ecosystem(2025-01-25); ; Given the critical role of data availability for growth and innovation in financial services, especially small and mid-sized banks lack the data volumes required to fully leverage AI advancements for enhancing fraud detection, operational efficiency, and risk management. With existing solutions facing challenges in scalability, inconsistent standards, and complex privacy regulations, we introduce a synthetic data sharing ecosystem (SynDEc) using generative AI. Employing design science research in collaboration with two banks, among them UnionBank of the Philippines, we developed and validated a synthetic data sharing ecosystem for financial institutions. The derived design principles highlight synthetic data setup, training configurations, and incentivization. Furthermore, our findings show that smaller banks benefit most from SynDEcs and our solution is viable even with limited participation. Thus, we advance data ecosystem design knowledge, show its viability for financial services, and offer practical guidance for privacy-resilient synthetic data sharing, laying groundwork for future applications of SynDEcs.Type:journal articleJournal:Electronic MarketsVolume:35Issue:1Scopus© Citations 13 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Enhancing IT Service Management Through Process Mining – A Digital Analytics Perspective on Documented Customer Interactions(Springer Fachmedien Wiesbaden, 2025); ; Our study explores the integration of text mining and process mining to enhance the understanding of IT support agents' problem-solving activities documented in service tickets. Despite the rise of AI-based self-service systems, the pressure on IT support to deliver high-quality service remains significant, necessitating advanced analytical approaches. While text mining has been used for classifying customer requests or predicting satisfaction, it falls short in revealing the actual processes agents follow. By conducting a systematic literature review and a case study, this research outlines a novel approach combining text and process mining. The findings provide practical guidance for extracting activity catalogs and generating event logs from service documentation, offering valuable insights into service processes and highlighting challenges related to data quality in digital analytics.Type:journal articleJournal:Forum DienstleistungsmanagementScopus© Citations 1 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Wiping out the limitations of Large Language Models - A Taxonomy for Retrieval Augmented Generation(2024); - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Augmentierung statt Automatisierung: Nachhaltigkeit in der Entwicklung von KI-Systemen aus der Mitarbeitendenperspektive(Springer Fachmedien Wiesbaden, 2024); ; Type:journal articleJournal:Forum Dienstleistungsmanagement - Some of the metrics are blocked by yourconsent settings
Item type:Publication, A Value Co-Creation Perspective on Data Labeling in Hybrid Intelligence Systems: A Design Study(2023); ; The adoption of innovative technologies confronts IT-Service-Management (ITSM) with an increasing volume and variety of requests. Artificial intelligence (AI) possesses the potential to augment customer service employees. However, the training data for AI systems are annotated by domain experts with little interest in labeling correctly due to their limited perceived value. Ultimately, insufficient labeled data leads to diminishing returns in AI performance. Following a design science research approach, we provide a novel human-in-the-loop (HIL) design for ITSM support ticket recommendations by incorporating a value co-creation perspective. The design incentivizes ITSM agents to provide labels during their everyday ticket-handling procedures. We develop a functional prototype based on 17,120 support tickets provided by a pilot partner as an instantiation and evaluate the design through accuracy metrics and user evaluations. Our evaluation revealed that recommendations after label improvement showed increased user ratings, and users are willing to contribute their domain knowledge. The improved labels can be utilized to continuously enhance the AI system as rewards showed increasing growth with decreasing marginal returns. Overall, our results emphasize agents' need for value-in-use by providing better results if they improve the labeling of support tickets pre-labeled by AI. Thus, we provide prescriptive knowledge of a novel HIL design that enables efficient and interactive labeling in the context of diverse applications of reinforcement learning systems.Type:journal articleJournal:Information Systems (IS)Scopus© Citations 21 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Design Features for Explainable Generative AI (GenXAI) Systems in Knowledge-Intensive Service Work(2026-01-06); ; The use of generative AI (GenAI) and large language models (LLMs) in knowledge-intensive fields like customer support is rapidly growing. While GenAI responses often appear persuasive, they carry the risk of inaccuracies and hallucinations. Hence, users must critically evaluate responses to reach appropriate reliance and knowledge utilization. Despite technological advancements, design knowledge for enhancing human-GenAI interaction from an explainable AI (XAI) perspective remains lacking. Thus, this study applies the design science research (DSR) approach to develop explanations that aid human interaction with GenAI systems. Drawing from XAI literature and human reasoning theories, we built and evaluated seven design features and instantiated a prototype that contributes to the development of reliable explainable GenAI (GenXAI).Type:conference paperJournal:Hawaii International Conference on System Sciences (HICSS) - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Evaluating and Improving Prompt Quality in LLM-Based Assistants: A Synthesis of Criteria and Indicators(2026-01-06); Generative AI (GenAI) assistants, particularly large language models (LLMs), are gaining increasing relevance across domains. The quality of outputs generated by these systems is highly contingent on the input prompts, giving rise to new professional roles such as prompt engineers. In this study, we systematically examine evaluation criteria and optimization methods that can improve prompt quality. Drawing on a systematic literature review, we identify key criteria, including clarity, accuracy, and precision, and initial measurement techniques. In addition, we synthesize common optimization methods such as iterative refinement and shot-based prompting. Our work contributes to the growing efforts to standardize the evaluation and improvement of prompts in interactions with LLM-based assistants, thereby fostering a more rigorous and coherent understanding of the prompt quality construct.Type:conference paperJournal:Hawaii International Conference on System Sciences (HICSS) - Some of the metrics are blocked by yourconsent settings
Item type:Publication, NAVIGATING THE LANDSCAPE OF IT OPERATING MODELS OR “HOW IT WORKS”(2026-06-15) - Some of the metrics are blocked by yourconsent settings
Item type:Publication, DEVELOPING A HYBRID VECTOR-GRAPH RETRIEVAL SYSTEM FOR ENTITY-PRESERVING AND INSPIRING STORYLINE CREATION OF PRESENTATION SLIDES(2025-06-12); ; Effective presentation slide creation is crucial for impactful communication, yet fully automating this task with AI is insufficient. Hybrid human-AI solutions often perform worse than pure AI or human creation due to overreliance on AI. To address this, we develop design principles for configuring human-AI hybrid systems in complex knowledge tasks using a design science research approach. Our prototype, NarrativeNet Weaver, leverages an underutilized corpus of existing presentation slides, applying generative AI advances in hybrid dense embedding and graph-based retrieval techniques. Evaluated through 15 think-aloud sessions and 73 user trials, users with NarrativeNet Weaver exhibit greater engagement and achieve equal or improved slide quality compared to those using a ChatGPT-based chatbot with a vector database. We contribute design knowledge for human-AI systems for complex multimodal content and offer a new approach to retrieving and visualizing existing slides, enhancing the utilization of valuable but underused resources.Type:conference paperJournal:European Conference on Information Systems (ECIS)Volume:2025