E-commerce firms often face the decision on whether they should implement a word-of-mouth (WOM) system on their websites. An in-site WOM system can potentially boost customer conversion by conveying signals and information about product popularity and quality. However, implementing such a system might also have unintended consequences, hindering product sales because of the lack of control over WOM volume and content. This study examines how implementing a WOM system (through social media integration) on an e-commerce website affects customer conversion in the two stages of the consumer purchase funnel, namely, adding a product to the cart and placing an order. Identifying the causal effect of WOM system implementation is challenging because e-commerce websites usually make their decisions based on private information about the potential impact of the system and may have simultaneously implemented other initiatives that could confound the effect. As a result, we conducted a randomized field experiment in collaboration with a large e-commerce website in the United States by testing two versions of a web page: one with a WOM system (treatment) and one without (control). We find that the impact of a WOM system implementation on customer conversion is moderated by WOM volume such that its effect is positive above a threshold of volume (measured by the number of comments) and negative below the threshold. Additionally, we find that WOM valence reinforces the impact of a WOM system on customer conversion. These results suggest that a social-learning mechanism is in play. Furthermore, our results show that the impact of a WOM system mostly occurs at the consideration stage in the upper purchase funnel (adding a product to the cart) rather than at the evaluation stage in the lower funnel (placing an order). Our study not only contributes to research on online system design but also offers practical implications for implementing and managing WOM systems on e-commerce websites.