Honey Science Corporation provides Honey, an online service that automatically finds coupon codes for the website users are shopping at and applies them to their order when they check out, saving money and coupon searching time. Their Rewards program, HoneyGold, allows users to earn cash back when they are shopping online on various stores. The Los Angeles based company was incorporated in 2012.
Honey would like to identify and acquire new high value users. They wanted to leverage Machine Learning to intelligently comb through their millions of users and then identify high value ones. This will allow Honey to spend their marketing dollars efficiently and run specific marketing campaigns to keep and grow these users.
Google Cloud Implementation
SpringML partnered with Honey on exploring how data science and machine learning could be leveraged to improve its marketing efforts by understanding its customers and providing additional value to vendors and loyal users. SpringML implemented two use cases – a model that predicts the number of visits a user will make to the Honey website, as a proxy for customer lifetime value; and a model that segments users by the amount of times they complete code trials (stage 7 in Honey’s sales funnel) to better understand differences in Honey’s customer base.
Understanding which users are likely to generate more LTV would enable Honey to better target customers with advertising as well as determine how to optimally allocate marketing spend to different user LTV tiers.
Amid ad delivery issues, market saturation, etc., cost per user acquisition tends to fluctuate on the independent channel level. Due to limited demographic information, these models can be used for early engagement based on the lifetime value. The marketing channel owners will now be reviewing 30, 60, 90 day feature engagement for new acquisition on a monthly cadence to inform buying strategy.
Honey can use the customer segmentation model to identify which users land at a late stage more often than others and focus their attention on them. They can then weed out the less active users afterwards. This can then be expanded to other stages with similar patterns of usage.