Abstract: We investigate the moderating effect of product attributes and review ratings on two different impacts of a purchase-based collaborative filtering recommender system on an e-commerce site: number of product views and conversion to purchase conditional on a product view. We run a randomized field experiment with a top North American retailer’s website with 184,375 users split into a recommender-treated and a control group. We tag the attributes of 37,125 unique products via Amazon Mechanical Turk and augment the data with other sources of product information on the site (e.g., review data, product descriptions, etc.).
Our study confirms that the use of a recommender increases both views and conversion rate among treated users for all products, but this increase is moderated by product attributes and review ratings. We find that a recommender’s positive impact on product views is greater for utilitarian products compared to hedonic products and for experience products compared to search products. In contrast, a recommender’s positive impact on conversion rate is greater for hedonic product compared to utilitarian product. Furthermore, we find that recommenders’ positive impact on conversion rate is greater for products with lower average review ratings, suggesting that a recommender acts as a substitute to high review ratings. While the opposite is true for product views – recommender and high review ratings are complements in increasing views. We discuss the potential mechanisms behind our results as well as their managerial implications.