The power and limits of distributed representations in strategic decision-making

Abstract:

This paper develops a formal theory of distributed representations—collective cognitive models that emerge when organizations aggregate the simplified mental models of multiple members—and examines when they enhance or hinder strategic decision-making. We extend Brunswik’s lens model to multiple decision-makers and introduce “decision boundaries” from machine learning to explain how aggregation structures interact with individual internal representations across varying task environments. Using a mathematical model of project screening, we compare two prototypical aggregation rules (Averaging and Unanimity) against individual Specialists (single-cue experts) and Generalists (multi-cue learners) across various environments and levels of experience. Our analysis reveals that effectiveness depends critically on the three-way interaction between internal representations, aggregation structure, and environmental conditions: Specialists excel when one cue dominates; Unanimity guards against errors when good projects are rare and decision-makers lack experience; Averaging delivers robust performance across most settings; while only highly experienced Generalists outperform distributed representations, though such individuals are scarce in practice. These findings advance microfoundations by linking individual cognition and organizational aggregation, enrich the attention-based view by showing how cognitive processing and aggregation matter beyond attention allocation, and offer actionable guidance for designing decision processes under strategic uncertainty.

Other links:

• SSRN preprint