Transfer Learning: An Approach for ROI Optimization


CTR is loosely correlated with the quality of users. As a result, training models to optimize for clicks can lead to achieving high CTR but poor ROI. Transfer learning is one approach to optimize for quality users.

The first step is mapping the users in our database to identify raw features that are most related to user quality. Features can be activity of users in similar apps, IAP purchase activity, etc. This data allow us to create different user segments. For example, active users can be put into active and very active user segments while spenders can be put into low, average, and high spender segments. If we want to target only very active users and high spenders, the created segments can be utilized to find audiences who share similar traits and conduct similar behaviors as the selected segments. This model is called lookalike audience targeting.

The search for lookalike audiences can be accomplished by leveraging a programmatic platform partner’s bidding algorithms and database. For example, once Aarki receives user segments, Aarki utilizes proprietary algorithms to find similar users and expand the list of device IDs in each segment. Aarki scores each device ID by how similar they are to each user segment. A low score means the user is not very similar to a selected user segment and high score means the user is very similar to the selected user segment. Aarki’s bidding algorithms then target users who have high scores in real-time bidding (RTB). In a recent campaign the transfer learning approach showed a 138% increase in purchase rate and 16% increase in retention rate.

Aarki’s data scientists and engineers are developing advanced machine learning algorithms to reach and acquire the best users and deliver strong ROI. Contact us at to learn how machine learning can help accelerate your marketing goals.

Topics: Machine Learning