Toward AI-Supported Decision Making in Innovation: Requirements and Design Implications for LLM-Based Multi-Agent Systems
- Published in
- Innovation and Product Development Management Conference (IPDMC 2026), full paper
- Year
- 2026
- AI-supported decision making
- LLM-based multi-agent systems
- Innovation management
- Product development
- Design requirements
- Human-AI collaboration
Abstract
Innovation decisions in product development are frequently made under conditions of incomplete information, fragmented documentation, and limited cross-functional coordination. These are also the conditions under which overreliance on AI outputs is most likely, particularly when the system’s underlying uncertainty is not clearly communicated. This study examines what decision makers in innovation contexts require from AI-based decision support and what this implies for the design of large language model (LLM) based multi-agent systems. An online survey (n = 33) of participants from twelve German organizations was combined with 16 semi-structured interviews across fourteen firms. The findings indicate that practitioners face recurring breakdowns in information integration, reasoning traceability, and cross-functional alignment, while consistently demanding transparency, human oversight, explicit handling of data gaps, and secure treatment of sensitive information. Based on these findings, seven design requirements and six design principles for AI-supported decision systems in innovation contexts are derived and their implications for LLM-based multi-agent architectures are discussed. The central contribution is an empirically grounded framework that positions AI as an analytical partner rather than an autonomous decision maker, identifies when multi-agent configurations are appropriate for coordinating distinct functional perspectives, and proposes that such systems are most likely to be effective when uncertainty is made explicit, adoption is introduced incrementally, and AI capabilities are treated as complementary to human expertise rather than as a substitute for it.
Cite this paper
Saini, B. K., Gürtler, L. I., Auspurg, D., Kumar, C., Peissner, M., & Hölzle, K. (2026). Toward AI-Supported Decision Making in Innovation: Requirements and Design Implications for LLM-Based Multi-Agent Systems. In IPDMC 2026.