“Squaring the circle: Decrease purchase price, increase satisfaction, and maintain revenues using a new machine-learning recommendation system” by Michael Harrison (USAA and the University of North Carolina) and Jan-Benedict E.M. Steenkamp (the University of North Carolina)

In this research the authors develop a new automobile recommendation algorithm using machine-learning methods and compare its impact to customers and to the firm, relative to that of its predecessor recommendation algorithm. Our algorithm is designed to reduce the average automobile purchase price so to as increase customer financial security, while maintaining firm revenues. The algorithm’s performance is evaluated in two ways including (1) a randomized field experiment with treatment and control groups matched based on characteristics relevant to automobile purchases; and (2) a comparison of firm’s automobile loan and insurance revenues in 2016 (old recommendation algorithm) versus 2017 (new recommendation algorithm). The recommendation algorithm we developed has increased the predictive accuracy of customer automobile purchases by 67 percentage points. The average automobile purchase price was 19.2% lower with our recommendation algorithm yet 91% of customers who had purchased an automobile using our recommendation algorithm said they would recommend the automobile they purchased to others. Revenues on auto loans and insurance policies were 16.7% higher, with our recommendation algorithm, in 2017 than in 2016. Lastly, with our recommendation algorithm, dollars in default declined by 89% and the number of delinquent loans decreased by 87%.