As a destination for both avid collectors and casual fans alike, HobbyLink Japan (HLJ) boasts a sprawling catalog of over 700,000 products - some in limited release. Particularly when it came to new items, it was crucial that fans of specific characters, franchises, or brands be alerted to HLJ's inventory in a timely manner, even if the item had little-to-no interaction data.
In addition to being able to accurately interpret the on-site behavior (clicks, views, favorites, cart adds, etc.) from user to unique user, HLJ had two added layers to consider when serving recommendations. Firstly, HLJ offers a forgiving cancellation policy for pre-order items, allowing customers to cancel at any time before the product is shipped. That meant that a customer’s decision to buy a pre-order item needed to be weighed less favorably than a customer who purchased an item that was readily available to be shipped in terms of making further recommendations. Secondly, many of the items in HLJ's catalog are one-of-a-kind, meaning it had to avoid recommending sold-out items to users whose tastes matched that item.
Crossing Minds’ solution was able to account for "cold start" recommendations by interpreting catalog items’ user-generated content (UGC) and other attributes, ensuring that items were recommended to the right users more quickly. For the items that were one-of-a-kind or limited in inventory, the Crossing Minds’ algorithm was able to ensure that sold-out items were not being recommended to users whose tastes matched those products.
Additionally, because of the generous cancellation policy, an HLJ customer’s decision to buy a pre-order item needed to be weighed less favorably than a customer who purchased an item that was readily available to be shipped when it came to making further recommendations. With Crossing Minds’ innovative approach to Machine Learning, this capability was made possible.
HLJ previously deployed AWS as its recommendation engine; however, with Crossing Minds, it was able to increase clicks, transactions, and revenue dramatically. With a projected return on investment (ROI) of $106,000, HLJ witnessed firsthand the importance of partnering with a recommendation engine that was tailored to the nuances that make HLJ’s business model unique.