To save costs, marketers often utilize traditional response modelling to target only those customers that are likely to respond to the marketing campaign. However, these models fail to differentiate between customers who respond favourably because of the campaign and customers that respond favourably on their own accord, regardless of the campaign. Uplift modelling aims to establish the net difference in customer behaviour resulting of a specific treatment that is given to the customer.
Several state-of-the-art techniques and methodologies are benchmarked on multiple private and public datasets, representing both cross/up-selling and retention campaigns. In evaluating the results, we highlight a couple of issues. One of those regards the evaluation of different uplift modelling techniques. Although different evaluation techniques exist to evaluate the performance of uplift models, the interpretability of the metrics is not so intuitive. Therefore, we propose a profit-driven approach towards uplift modelling which considers the costs of the campaign and the expected benefits. Our profit-driven approach allows us to identify the customers who are both highly influential and beneficial for a campaign. This approach allows for clear and interpretative knowledge to be used in future business decisions when setting up a new campaign.
Original languageEnglish
Publication statusUnpublished - 15 Jun 2018
EventBusiness Analytics Research Day - KU Leuven, Leuven, Belgium
Duration: 15 Jun 2018 → …
Conference number: 2018
https://data-lab.be/bard

Seminar

SeminarBusiness Analytics Research Day
Abbreviated titleBARD
CountryBelgium
CityLeuven
Period15/06/18 → …
Internet address

ID: 38822507