Observation of past usage can be informative about future usage. This paper develops an individual-level static learning model that incorporates a time-varying risk-averse coefficient that depends on consumer’s past experience to examine behavioral biases, and applies it to a dataset from one subscription service provider in the United States. We find strong evidence of be- havioral biases when consumers make choices under uncertainty: accuracy of their past choices does impact their self-confidence level, which in turn impacts their future choices. In particular, both "Hot-Hand Bias" and "Gambler’s Fallacy" hold in consumers’ decision making processes. Counterfactual simulations show that patients would have made more suitable choices if these behavioral biases can be overcome. This study yields a set of marketing implications for subscrip- tion services and other business modes such as loyalty programs. Optimal marketing strategies should take into account on how to help consumers to become better "forward-looking" person.