Project - Machine Learning Propensity Model
Using machine learning to understand customer behaviour and identify high potential marketing leads, significantly increasing new customer acquisition.
Over 90% of customers who started the onboarding journey dropped out before successfully opening the product
Summary
Customer acquisition for niche investment products is hard with over 90% of customers who started the onboarding journey dropping out.
Our value goal was to significantly increase the customer acquisition and revenue using machine learning, enabling a far better understanding of a customer’s propensity to buy.
Impact
Applying the predictive model, we identified a large pool of previously unidentified leads, significantly increased the onboarding success rate and incremental revenue of £20m.
The average onboarding journey time was reduced by 25%, saving time and cost.
Our approach
Process and profile visualisation
We used data visualisation to understand the typical journeys and characteristics of customers who completed new business onboarding and those who did not. From this we were able to identify features and events in the process that indicated the likelihood of success of onboarding.
Transactional behaviour profiling
We applied Natural Language Processing and Principal Component Analysis to build aggregated behavioural profiles for customers based on transactions. What we discovered about the transaction behaviour of customers who tended to complete onboarding was significantly at odds with existing beliefs and resulted in fundamental changes to our client’s marketing approach.
Predictive machine learning models
Using what we’d learnt from our customer behaviour and transaction profiling, we built and tested different machine learning algorithms to predict which customers would successfully complete an onboarding journey.