Purchase-intent scoring

Summary

Predictive models for a CPA network’s sales workflow.

The business needed to distinguish buying intent from low-value traffic and prioritise follow-up effort.

A behavioural-data pipeline and predictive model that use interaction patterns to score purchase intent and prioritise a call queue.

Tech Challenge

• By using deep learning, our team first aimed to understand customer behavior and then model the probability of purchase based on user’s web-surfing experience.
• Those recommendations had to be revenue driven, maximizing profits of the service, while providing high quality services to a customer.
• Integration of our Apache Spark + Tensorflow architecture with clients Elastic + PostgreSQL + RabbitMQ.

Solution

• We’ve developed a detailed events tracking plan of customers activity and then tested the implementation against the historical data.
• The above helped to measure the most important behavioral patterns, correlating with customers purchase activity.
• Deep learning technologies helped us find the unintuitive correlations between users activity and purchasing outcomes, which are hard to find and interpret.

Impact

A predictive scoring workflow connected to sales follow-up prioritisation.