Algorithmic Discrimination in Service

Abstract

This research investigates conditions under which algorithmic discrimination can impact long-term demand and profits. We employ experiments and an agent-based model to demonstrate that algorithmic discrimination can be profitable in the short-run but can erode profits in the long-run. This research shows that discriminatory algorithms have short-term profit advantages, but non-discriminatory algorithms earn higher long-term profits when factoring in consumer word-of-mouth and competition. Large error in measuring consumer quality (value or profitability to the firm) exacerbates algorithmic discrimination, while large consumer heterogeneity attenuates it. This research emphasizes the long-term benefits of using non-discriminatory algorithms, as well as incorporation of word-of-mouth considerations in the algorithm’s design. However, for firms that must engage in using discriminatory algorithms, this research recommends increasing investment in methods of measurement error reduction and increasing exposure to consumers of different populations. By doing so, a firm could reduce algorithmic discrimination while improving both long-term profits and societal well-being.