Surveys without Questions: A Reinforcement Learning Approach

摘要: The ‘old world’ instrument, survey, remains a tool of choice for firms to obtain ratings satisfaction and experience that customers realize while interacting online with firms. While avenues survey have evolved from emails links pop-ups browsing, the deficiencies persist. These include - reliance on very few respondents infer about all customers’ interactions; failing capture customer’s interactions over time since rating is one-time snapshot; inability tie back specific because provided relate interactions. To overcome these we extract proxy clickstream data, typically collected every interactions, by developing an approach based Reinforcement Learning (RL). We introduce new way interpret values generated value function RL, as ratings. Our does not need any data training. Yet, validation against actual yield reasonable performance results. Additionally, offer draw insights function, which allow associating their two metrics represent one, customer-level other, aggregate-level click actions across customers. Both are defined around proportion pairwise, successive show increase in This intuitive metric enables gauging dynamics better predictor purchase than customer survey. allows pinpointing help or hurt experience. In sum, computed unobtrusively clickstream, action, each customer, session can interpretable more insightful alternative surveys.