Images: Urs Wehrli

Finding customer needs using Cluster Analysis

Published: Author: Oliver Staubli, CEO & Data ScientistTags: ArticleE-CommerceData ScienceExamplesExploratory Data AnalysisData Visualization

Whether your company sells clothes, cars or shampoo, with every product sold you should learn more about the past needs of your customers. The longer the customer relationships and the more customers you have, the greater the chance that exciting patterns are hidden in your transactional data. The knowledge about your customers' needs is ideal for cross- and up-selling. Through improved targeting (matching the right offer to the right customer), your customers are more likely to convert. In addition, customer loyalty increases because the customer feels understood and listened to.

With the help of the cluster analysis it is possible to distill from thousands of customer profiles, with hundreds of dimension (distribution of the product purchases on the product categories) automatically the "typical" need profiles. The following visualizations are based on synthetic data of a fictitious company to explain the characteristics of the cluster analysis. In the following radar chart, the "typical" profiles found with the cluster analysis can be easily read.

With the help of cluster analysis it is possible to distill automatically from thousands of customer profiles, with hundreds of dimension the "typical" need profiles. The dimensions can be the percentage of product purchases in each product categories offered by your shop. The following visualizations are based on synthetic data of a fictitious company in order to explain the characteristics of the cluster analysis. In the following radar chart, the "typical" profiles found with the cluster analysis can be easily recognized.

In the synthetic data 10 clearly distinguishable clusters (customer groups) were found. For example, when selecting the CL-4 cluster, one can see that the customers assigned to this cluster typically bought about 52% of their products in the sector "toys" and 25% in "children's clothes" sector, but bought practically no "jewellery" or "luggage".

The result of a cluster analysis

Cluster analysis automatically finds the ideal customer grouping in terms of homogeneity (similarity within the cluster) and heterogeneity (distinguishability of the clusters). In the end, the following facts are therefore available:

  1. How many "typical" customer groups are hidden in the data.
  2. Which are the most important dimensions (product categories) to differentiate the needs of the customers
  3. How the individual cluster profiles differ in the dimensions found.
  4. Which customer is assigned to which cluster and thus the size of the individual clusters.

The following interactive bubble chart helps in understanding the differences in the resulting cluster profiles. Clusters horizontally as well as dimensions vertically can be sorted by hand to group similarities.

Certain dimensions and clusters stand out with large "bubbles", i.e. a large proportion of product purchases in this category. For example, clusters CL-7 and CL-9 appear to be similar in three dimensions, but are clearly distinguishable in terms of the number of customers and in the children's clothing segment. In this way, all clusters should be compared with each other in order to identify the core characteristics. It is also helpful in this process to assigen descriptive names to the clusters: Cluster "CL-4", for example, could be described as "the Multimedia Cluster", i.e. a typical customer from this cluster buys only multimedia products.

The added value of a cluster analysis

In addition to better understanding the needs of your customers, the results of a cluster analysis can be directly applied to marketing activities, resulting in the following benefits:

  1. Better conversion: A personal approach based on needs increases the chance of conversion.
  2. Better customer loyalty: Customer feels understood and customer loyalty is strengthened.
  3. Better targeting: The customer groups are sharply separated and thus minimize scattering loss in targeted campaigns.
  4. Higher utilization: Marketing budget can be used optimally without wastage.

Of course, there are many other possible applications for the results of a cluster analysis. For example, several online shop operators already use our cluster analyses to address their customers directly in the e-shop in real time based on the corresponding cluster profile: Thus, search results, navigation and campaign images change depending on the cluster to which the current customer is assigned to based on his previous purchases.

Even if the current click behaviour of a customer deviates from his usual shopping profile, or for new customers without a purchase history, the "next best" cluster affiliation can be calculated in real time, thus allowing the currently best marketing material and matching products to be displayed. We would be happy to give you more information about our implemented projects in person.

~

Did we arouse your interest to get a cluster analysis applied to your own customer data? We would be happy to help you with the preparation of the data and with the cluster analysis.

Get in touch