Customer Targeting with Uplift Modelling using Gradient Boosting at Verivox

19.03. | 14:45 - 15:15 | tech.stage

Verivox is an online comparisor based in Heidelberg. One of our main products are ‚Loans‘. Customers apply online for a loan and obtain immediately offers from different banks. Credit Advisory can then support our customers by figuring out the most suitable loan offer as well as motivating them to actually apply for the loan. A crucial point is the time when the customer is contacted. Speed and Conversion show a high positive correlation due to the competitive environment.

The question is:
Which customers should be contacted first by Credit Advisory to increase Profitability?
The model is based on the idea of Uplift Modelling using Gradient Boosting. Uplift Models as opposed to Response Modelling focus on the incremental response due to the positive ‚Treatment‘ by Credit Advisory. The model’s idea is to distinguish customers whose positive response is in fact directly driven by Credit Advisory. Thus, the model should help to avoid marketing money to be spent on customers who do not want to be disturbed, the lost causes who will never convert but also the sure things, which will convert anyway.

The model was estimated in GNU R and is connected with a Rest API to the Salesforce Platform, which is distributing the Customers near real-time to the Credit Advisors using the tool Ortoo-Qassign.


Franz Eigner (Verivox GmbH)