Abstract
Large Language Models (LLMs) offer significant potential for organizational decision support, but realizing this potential requires understanding where humans systematically struggle as decision makers. This paper develops a framework mapping LLM capabilities onto specific mismatches between evolved human cognition and the demands of modern complex systems. We identify six families of cognitive constraints and four clusters of system properties; their intersection produces predictable decision failures we term human-system mismatches. We focus on eight archetypal mismatches including temporal myopia, complexity compression, polarization cascades, and illusion of understanding, and derive eight corresponding LLM augmentation mechanisms: temporal bridging, complexity unfolding, signal amplification, counterfactual scenario generation, multi-perspective integration, risk and threshold scanning, silo bridging, and incentive transparency. The framework a more detailed roadmap for effective LLM deployment than generic adoption recommendations, while acknowledging that each mechanism inherits LLM limitations including hallucination and dependence on organizational data quality.
Recommended Citation
Keyhani, Mohammad; Vahidov, Andishe; and Jamshidi, Zahra, "A Human-System Mismatch Theory of LLM-Augmented Decision Making" (2026). ASAC 2026. 17.
https://aisel.aisnet.org/asac2026/17