By Nick Ruffilo, CIO/CTO BookSwim.com
I’ve seen quite a bit of discussion around metadata and its power, but numbers that confirm its usefulness are rarely announced. Six months ago, BookSwim.com rebuilt its search and recommendation engine. The results from the upgrades were mind-boggling.
BookSwim is an online subscription book rental club. Members can rent unlimited titles from our catalog by paying a flat monthly fee. Because price is not a factor for our members, they select books based solely on their interests.
BookSwim made the choice to rebuild its search and recommendation engine because the acceptance rate (amount of times a book was added to a member’s wish list per search) was lower than the industry average. Our solution came by addressing the information we have on our readers’ habits as well as the information that determined our recommendations.
Our original engine only provided recommendations based on individual titles, judging by what past readers of that title had also rented. This is the methodology that most recommendation engines use today, such as Amazon’s suggestion section: “Those who read this book also read this book.” We analyzed different data attributes about books to determine what was available and what was most important to users. A book’s genre was considered highly important to readers in book selection and was readily available, so that was what we chose.
Our first step was to determine which genres were most important to individual members. By examining reader preferences and rental histories, we tagged each member as a specific genre reader. Many readers were split between two different genres, since few readers restrict themselves to only romance novels or only non-fiction books, etc., so we tagged them with both. Our last step was to adjust our recommendation engine to factor in book genres. No longer would a romance reader see a recommendation for a sci-fi book nor would a mystery reader receive a suggestion for a biography.
After our release, we monitored the traffic to our recommendations page and it held steady (which gave us to get a perfect comparison between the old and new versions). We found that members added 2.78 times more books to their rental queue with the new recommendation engine. The data shows that readers are open to exploring new titles as long as they are within their genres of choice.
While some book sellers may not know a reader’s book history, they can determine genre preferences with the simple question: “To help you find the best books, please let us know which genres you most prefer to read.” Similar results could also be found using any other piece of metadata, as long as you understood which specific attributes were important or preferred by the reader. The power of data to increase sales is great and surprisingly untapped.