Consider or Choose? The Role and Power of Consideration Sets

TL;DR

This paper emphasizes the importance of consideration set formation in customer choice modeling, demonstrating its critical role in model identifiability, assortment planning, and predictive performance, even with a simplified second-stage choice mechanism.

Contribution

The paper introduces a consideration set model characterized by symmetric demand cannibalization, highlighting its identifiability and implications for assortment optimization and computational complexity.

Findings

Model is fully identifiable due to symmetric demand cannibalization.

Optimal assortment is revenue-ordered within consideration set partitions.

Despite simplicity, assortment optimization remains NP-hard to approximate.

Abstract

Consideration sets play a crucial role in discrete choice modeling, where customers often form consideration sets in the first stage and then use a second-stage choice mechanism to select the product with the highest utility. While many recent studies aim to improve choice models by incorporating more sophisticated second-stage choice mechanisms, this paper takes a step back and goes into the opposite extreme. We simplify the second-stage choice mechanism to its most basic form and instead focus on modeling customer choice by emphasizing the role and power of the first-stage consideration set formation. To this end, we study a model that is parameterized solely by a distribution over consideration sets with a bounded rationality interpretation. Intriguingly, we show that this model is characterized by the axiom of symmetric demand cannibalization, enabling complete statistical identification. The latter finding highlights the critical role of consideration sets in the identifiability of two-stage choice models. We also examine the model's implications for assortment planning, proving that the optimal assortment is revenue-ordered within each partition block created by consideration sets. Despite this compelling structure, we establish that the assortment problem under this model is NP-hard even to approximate, highlighting how consideration sets contribute to nontractability, even under the simplest uniform second-stage choice mechanism. Finally, using real-world data, we show that the model achieves prediction performance comparable to other advanced choice models. Given the simplicity of the model's second-stage phase, this result showcases the enormous power of first-stage consideration set formation in capturing customers' decision-making processes.