Customer segmentation has become a pivotal part of every organization seeking to improve decision making to enhance customer relationship management and service delivery within the value chain. The Traditional Recency, Frequency and Monetary value (RFM) concepts have been widely used for customer segmentation; notwithstanding this, they have not been able to capture contextual behaviour in the public sector service context. This study, however, extends the FRM model by integrating time and location variables to create a Recency, Frequency, Monetary value, Time and Location (RFMTL) model combined with K-means clustering for customer classification. A mixed method approach was employed through the use of transactional data from a public sector institution, which was normalized and analysed by the K-means clustering algorithms. Also, the Fuzzy Analytical Network Process (FANP) which was qualitative in nature, was used to determine the relative importance of the RFMTL variables based on judgement from experts. The results from all these analyses produced four distinct segments with moderate but acceptable clustering cohesion, as indicated by the silhouette and elbow analysis technique. A comparative analysis with regard to RFMTL and the traditional RFM was also conducted, showing modest numerical improvement but clearer behavioural interpretability when time and location were included. The results of this research suggest that there is incremental value in developing a classification of customers for service-based public organizations by incorporating both contextual behavioural data and clustering. The research finding also asserts that the results of this research should be understood as a decision-support model rather than a prescriptive classification approach. This research is important for the field in that it builds on RFM segmentation, which was itself conducted in a public sector setting. This research also improves on previous research by combining FANP with clustering. One of the key limitations of this research is the single organization data set with a limited observation period.
Azietaku et al. (Sun,) studied this question.