We observed that marketing teams of major companies tend to build their marketing campaigns based on study of client segmentation and market analysis which can be an heavy workload that directly deteriorate its ROI
Using our client Big Data (6 000 000 client informations on a Hadoop / Spark Cluster) we tried an other approach to dramaticaly increase the agility for building and tayloring new marketing campaigns.
We used Test&Learn methodolgies to increase the "uplift" of the campaigns inspired by the multi-armed bandit problem (https://en.wikipedia.org/wiki/Multi-armed_bandit) and implemented the whole approach in Apache Spark.
In this talk we want to share our feedbacks on the approach the success we had and probably more importantly the pitfalls we encountered.