RE-RFME: Real-Estate RFME Model for customer segmentation

TL;DR

The paper addresses scalable customer segmentation for large, dynamic user bases on online real-estate platforms. It proposes RE-RFME, an end-to-end pipeline that computes Recency, Frequency, Monetary, and Engagement features to cluster users into four segments. Monetary is defined as $($PDP$)$ × 1 + $($Leads$)$ × 7, Engagement is the sum of per-session actions, Recency is days since last visit, and Frequency uses the last $45$ days' sessions. Using K-means with $k=4$ on Housing.com's web and app data yields interpretable segments (High Value, Promising, Needs Attention, Needs Activation) and supports targeted marketing strategies, validated by elbow-method evidence.

Abstract

Marketing is one of the high-cost activities for any online platform. With the increase in the number of customers, it is crucial to understand customers based on their dynamic behaviors to design effective marketing strategies. Customer segmentation is a widely used approach to group customers into different categories and design the marketing strategy targeting each group individually. Therefore, in this paper, we propose an end-to-end pipeline RE-RFME for segmenting customers into 4 groups: high value, promising, need attention, and need activation. Concretely, we propose a novel RFME (Recency, Frequency, Monetary and Engagement) model to track behavioral features of customers and segment them into different categories. Finally, we train the K-means clustering algorithm to cluster the user into one of the 4 categories. We show the effectiveness of the proposed approach on real-world Housing.com datasets for both website and mobile application users.

Figures (4)

• Figure 1: An End-to-end pipeline for proposed $\mathtt{RE\hbox{-}RFME}$ system
• Figure 2: Number of Clusters for App and Web
• Figure 3: Cluster analysis for mobile application test data users
• Figure 4: Cluster analysis for website application test data users