Analyzing Purchasing Behavior to Customize Offers

The current project is aimed to analyse and create a targeting market model for the retail sector. In other words, the main objective is to develop analytical skills in customer’s analytics and categorise customers in different groups and segments based on their purchasing behaviour (Chen, 2014). In the current time there is a lot of data generated each day. This data can be used to make customized offers to the customers based on their purchasing behaviour(Akbar & Parvez, 2009; Linof & Berry, 2011; Ziafat & Shakeri, 2014).

The first task is devoted for the customer segmentation. For the analysis purpose the data given has been used in SAS Miner and the clusters of customers have been identified. In this task the clusters were prepared on the basis of the demographic variables given in the data set. In this research the SAS Enterpriser software has been used to conduct the clustering analysis. There are various clustering method which are used for different situation. For this case the K means clustering has been used. There are total 5 clusters which are run in this project. In the first section the demographic variables have been used for the cluster analysis. The target variable for the analysis is subscription (Iaci & Singh, 2012; Trebuna, Halcinova, & Fil’o, 2014)

The demographic variables have been distributed in the five different clusters.

First cluster: This cluster consists of the cluster on the basis of the age. This cluster comprises the respondents who are students as profession. In this cluster respondents from the other profession are not included as the cluster is only focused for the students. Respondents in this cluster are married and the level of the education of the people are not known(Akguc & Akg, 2010; Alibeli, 2014).

Second cluster: Respondents in this cluster are not only of the particular age and the profession of the respondents in this cluster also varies. The main differentiating point for this cluster is that the people with specific education level are included in this segment which was not the case in the previous cluster. In terms of the marital status most of the people are married and it was not possible to map this segment with any particular segment of the Roy Morgan.

Third Cluster: The cluster also consists of the people with different age and different profession, however none of the variables in this cluster are statistically significant. One point to notice in this segment is that most of the people have obtained the education level till secondary. Another distinguished factor is of marital status as all the people in this segment are divorced. As compare to the previous two segments, this cluster can be mapped as the “Something better” from the Roy Morgan. This conclusion has been derived from the fact that the people in this segment are divorced and educated. So it is expected that the people wants something better in life and get out of the current situation.

Customer Segmentation Based on Demographic Variables

Fourth Cluster: The fourth cluster also takes into consideration the age and the profession of the people included in the data set. As compared to the other cluster, this cluster is more of the young people as the people in this group are between the 27 and 37 years of age. So, this is the cluster of the young people. In terms of the marital status people are either single or married. In terms of the education level most of them have attained the secondary level of education. While analysing the cluster, it can be said that this cluster is the “young optimism” while mapping to the segments from the Roy Morgan. This is because are settled and want to improve the standard of living.

Fifth Cluster: The fifth and the final cluster of this section is also the mix of the people in different age with different education level and also different career options. In this section most of the people are married, which means that they are settled in life. There is no divorced people included in this cluster. However this segment could not be directly mapped to any other section of the Roy Morgan.

With Outcome Variable:

In this section the clustering has been conducted using the outcome variable also in the analysis. The graphical presentation of the clusters is shown in the figure below.

Distribution of each variables among the 5 clusters with target variable

As the figure suggest the outcome variable can be categorised in two different clusters. The first clusters is the people who are not the subscribers. People who have not subscribed are all together a different group which has been presented in the current case. Furthermore in the second cluster, the people who have subscribed has been included. The next step is to examine whether the demographic variable has any role to play in defining each cluster. However after analysing the results from the above figure, one cannot find any significant demographic variable which can be used for further clustering. So, it can be concluded that there is no significance difference in the groups of people who subscribed to the term deposit from those group who do not subscribed.

We have identified 3 key segments from the outcome variable. Segment 1 & 4 consist of only people who have not subscribed. The segment 2 consist of people who have subscribed. Now, we will examine whether any demographic variable have significant roles in defining the cluster.

Cluster Analysis Results Based on Demographic Variables

In the previous section the demographical variables were used for the clustering purpose and the results were discussed. However in this section the clustering has been performed only using the behavioural variables in the given data set. Similar to the previous section, in this section also the cluster have been run for the features and secondly for the target variable(Davis, Gunderson, & Brown, 2008; Seiler, Rudolf, & Krume, 2013).

In this case the major variables taken into consideration are the personal loan, mortagage and the default credit. As the figure shows there are five segments for each of the variable. In the first segment those people are included who have not defaulted and who do not have any personal loan. In other words they do not deal much in loan. The second segment is of those people who have mortgage loan and personal loan both but they have not defaulted in any of them. The fifth segment is of those people who do not have the mortgage loan also have not defaulted.

While the clustering is analysed for the response variable, one of the segment shows that this cluster consists of the customers who have not subscribed. Also these people have not taken any personal loan and also they have not defaulted. So it can be concluded that there is no significant difference in the segment of customers subscribed to long term deposit from those customers who did not subscribed to the long term deposit.

The third task is devoted to the cross cluster analysis which consists of both the behavioural and the demographic variables. For this task Microsoft excel have been used and the results are shown in the table below.

To construct the cross section clusters, the clusters from the first section ( both the demographical and the behavioural clusters) have been used. Each respondent in the data set was assigned to one particular clusters and then the Microsoft excel was used to create a cross tabulation of the clusters. On the basis of the results from the cross tabulation, none of the clusters show and significant associations.

To analyse how the outcome variable and the combined clusters are related researcher have examine the number of the subscribed clusters in the each cells. After that researcher have examined the number of subscriber who lies in the combined section.

In the above cross tab, we are observing the percentage of subscribers falling into each clusters. For example, in cluster 1 of behavioural clusters & cluster 3 of demographic clusters there are 113.44% subscriber. The lift ratio of 13% in the segment. Similarly, the cluster formed from 2 in demography & 5 in behavioural consists 117.86% subscriber as compared to average of each selected segments. All the segments marked with green are key segments for the business.

Cross-Cluster Analysis with Behavioural and Demographic Variables

This can be explained using one example. Take the third cluster of the demographic clusters and the first cluster of the behavioural clusters, this shows that there are 113.44% subscribers. In this case the lift ration is 13 %.  Furthermore the cluster number 2 in the demography and the cluster 5 in the behavioural segment consist 117.85 % more subscribers as compared to the average of each selected segments. It should be noted that all the values marked in green are the segments which are important from business prospective.

In this section the combined clustering has been conducted using the behavioural data and the demographic data. As discussed, there are 5 clusters at max. Among the 5 clusters the significant variables is the default credit. This is because people in all the 5 clusters have either defaulted the loan or not defaulted. Among the 5 clusters the fifth clusters which only consists of students shows that they do not have personal loan, which was expected as the student do not have stable income so they do not take personal loan. The third segment can be of interest from business point of view as this section consists of people who have not defaulted on the mortgage loan.

Furthermore the cluster of the outcome variable has been shown in the figure below:

As the results shows, in the first segment only those customers are included who have subscribed. This segment consists of the people in different age group and different profession and people who have not defaulted. On the basis of the analysis it can be concluded that default credit and the personal loan are the most important variables in the current case.

Now running the cluster analysis with the target variables. In segment 1, only customer with subscribed people are included. In segment 1 there are mix age of people, also consist of people with mix career & who has not defaulted yet. People who have subscribed have no personal loan. So, the most important variable is personal loan & default credit.

Conclusion

The results from the clustering can be used by the managers to know which section of customers will require which types of the loan. Also the behaviour of the customers who have higher probability to default can be predicted using the results from the cluster analysis. People who are working and have higher level of education are less likely to default.

References

Akbar, M. M., & Parvez, N. (2009). Impact of Service Quality, Trust, and Customer Satisfaction on Customers Loyalty. ABAC Journal, 29(1), 24–38.

Akguc, M., & Akg, M. (2010). The Effects of Different Stages of Education on Income across Countries. Toulouse.

Alibeli, M. (2014). The Effect of Education, Employment, and Income on Reproductive Attitudes and Behavior of United Arab Emirates Married Women. PERSPECTIVES ON GLOBAL DEVELOPMENT AND TECHNOLOGY, 13, 227–2458.

Chen, J. (2014). Retail Customer Segmentation using SAS. Retrieved from https://www.sas.com/content/dam/SAS/en_ca/User Group Presentations/Calgary-User-Group/Chen-RetailSegmentation-Apr2014.pdf

Davis, A., Gunderson, M., & Brown, M. (2008). THE EFFECT DEMOGRAPHICS HAVE ON THE DEMAND. Florida.

Iaci, R., & Singh, A. K. (2012). Clustering high dimensional sparse casino player tracking datasets. UNLV Gaming Research & Review Journa, 16(1), 21–43.

Linof, G. S., & Berry, M. A. (2011). Data Mining Techniques: For Marketing, Sales, and Customer Relationship Management. New Jearsy: Wiley Publishing, Inc.

Seiler, V., Rudolf, M., & Krume, T. (2013). The influence of socio?demographic variables on customer satisfaction and loyalty in the private banking industry. International Journal of Bank Marketing, 31(4), 235–258.

Trebuna, P., Halcinova, J., & Fil’o, M. (2014). The importance of normalization and standardization in the process of clustering. IEEE, 12, 381.

Ziafat, H., & Shakeri, M. (2014). Using Data Mining Techniques in Customer Segmentation. Int. Journal of Engineering Research and Applications, 4(9), 70–79.