Client Background
Our client is a mobile advertising startup providing customer retargeting solutions to clients across ecommerce, financial services, healthcare and various other domains. These retargeting solutions work by targeting end consumer with relevant display ads whenever they surface on publisher network.
Business Objective
Develop data science driven customer retargeting platform that should be able to identify product(s) a customer is most likely to buy based on his/her browsing behavior and generate recommendation scores for likely purchase. These scores will be used to target customer with display ads whenever he/she surfaces on publisher network using programmatic media buying through ad exchanges.
Solution
Different prediction algorithms were developed for each major product segment/category. This is because buying journey, browsing behavior and transaction volume is different across these segments (Mobile & Electronics, Travel, Real Estate, Financial Services etc).Components Included
Data processing engine: Converts customer’s browsing data into feature set used by Machine Learning algorithms for prediction. This engine is implemented on top of Spark.
Machine Learning Algorithms: ML algorithms were build using R initially and later on Spark. Spark allowed us to train algorithms on large scale datasets and deploy there itself. Product recommendation Scores generated by algorithms were stored in Cassandra database to enable quicker access to the data.
API’s: Allowed querying for recommended products and associated scores at customer level. Business rules were overlaid on these recommendations to decide on ad to be shown to the customer.
Outcome
- 100 % improvement in campaign performance within 6 months of implementation
- 2 x improvements in CTR performance.
- Campaign cost reduced upto 20% because of automation.