Algorithmic Accuracy in Context: Assessing the Impact of Financially Motivated Dishonesty

Abstract

Abstract Algorithms are expected to radically improve the way businesses operate. To fully realize these benefits, customers have to accept the algorithmic implementations. Accuracy is a critical component of algorithmic acceptance models. However, thus far, research into accuracy has been limited to user ratings of the accuracy of the algorithmic output despite strong evidence for customer (user) acts of embellishment and other moral hazards. This article aims to deepen the understanding of accuracy ratings by assessing the impact of variation in user input accuracy on the accuracy ratings of the algorithm’s ultimate response. Drawing on honesty, a survey was presented to 500 individuals on the Cloud Research platform. The quantitative analysis showed that the more inaccurately a user felt and behaved, the more accurately they rated the algorithmic response of ‘decline’, with contextual accuracy factors explaining up to 35% of the variation in ratings of the algorithm’s accuracy. This finding raises important implications for researchers and practitioners who want to improve algorithmic acceptance but may have limited their improvement focus to explainability or model accuracy without considering the user behavior. Algorithmic accuracy ratings and acceptance should be evaluated in the context of the user and their likelihood to provide inaccurate inputs.