Improving Automotive Demand Predictions Using Online Activity Data

Brands’ website traffic, along with non-proprietary online resources, reflect consumers’ interests and intensions as part of their decision making process. We thus measure the relationship between traffic on the BMW website, and both Google and Wikipedia searches for the brand, to offline car sales. We then develop models for automobile market-level sales predictions, and test these models’ performance on recent sales numbers. We then proceed to construct prediction models for customer level purchase and churn probabilities, based on fine-grained CRM data, along with individual browsing patterns. These models assign purchase and churn scores for the brand’s existing and prospective customers, allowing for personalized marketing activities.