Untitled

Kayaalp, Fatih

doi:10.7212/zkufbd.v7i2.875

Cited by 1 publication

(1 citation statement)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Conventional supervised learning models can be used to predict churn [16,20,25,24]. An extensive review of machine learning for churn prediction is given in [15]. The main drawback of this approach is the absence of causal insight: in fact, there is no indication that the campaign will be most effective on customers with a high probability of churn.…”

Section: Churn Prediction and Uplift Modelingmentioning

confidence: 99%

Predicting Reach to Find Persuadable Customers: Improving Uplift Models for Churn Prevention

et al. 2021

View full text Add to dashboard Cite

Customer churn is a major concern for large companies (notably telcos), even in a big data world. Customer retention campaigns are routinely used to prevent churn, but targeting the right customers on the basis of their historical profile is a difficult task. Companies usually have recourse to two data-driven approaches: churn prediction and uplift modeling. In churn prediction, customers are selected on the basis of their propensity to churn in a near future. In uplift modeling, only customers reacting positively to the campaign are considered. Though uplift is better suited to maximize the efficiency of the retention campaign because of its causal aspect, it suffers from several estimation issues. To improve the uplift accuracy, this paper proposes to leverage historical data about the reachability of customers during a campaign. We suggest several strategies to incorporate reach information in uplift models, and we show that most of them outperform the classical churn and uplift models. This is a promising perspective for churn prevention in the telecommunication sector, where uplift modeling has failed so far to provide a significant advantage over non-causal approaches.

show abstract