Continuous customer segmentation with artificial intelligence

To target groups of customers with appropriate marketing strategies or offer especially valuable groups an even higher standard of support, many companies rely on customer segmentation methods. These may be based on the experience and intuition of marketing experts, for example, or on an analysis that finds patterns in large quantities of unstructured data. The latter approach, called “cluster analysis,” is very powerful and can identify segments even in cases where manual data analysis has reached its limits.

But even cluster analysis fails to provide a solution to one typical problem of customer segmentation: Segmentation always captures just a single moment in time, while the segments themselves are very volatile. Trends come and go, after all, and the preferences and needs of consumers change along with them. But these changes could render all the activities and results of a marketing strategy unusable.

To solve this problem, an interdisciplinary team of experts drawn from Arvato CRM Solutions and the ERCIS Omni-Channel Lab of the University of Münster has developed an innovative approach to segmentation based on “stream clustering,” the clustering of data from a continuous input stream. This makes it possible to identify and monitor segments over an extended period of time based on a stream of data, such as transactions. In this approach, the algorithm incrementally updates the segments. It identifies newly emerging segments and sorts out the ones that are obsolete. The results can be used at any time without complicated recalculations. This approach is fast and easy to implement, and it makes it possible to track changes in the customer segments.

The new method of segmentation has already proven itself in practice. It has been used for a retailer in the field of home furnishings, for example, where more than 1.7 million transactions were analyzed. First, the customer groups were divided along two dimensions: return rate and number of purchases.

In the case presented, the largest segment is made up of customers with very few purchases and low return rates. But there are also segments with markedly higher purchase frequencies and return rates. On the one hand, there are a few customers who make hundreds of purchases. These may be commercial resellers. On the other hand, there are a few customers who return almost every product that they buy. These customers appear to be unhappy with the service.

In the next step, the analysis was expanded to evaluate how often customers make purchases online, how many items they bought and how long ago their last purchase was. In this way, multiple segments were identified. To define these segments more precisely, the Arvato CRM team looked at their deviation from the overall average at a certain point in time and asked: what differentiates an identified customer segment from the average customer? This analysis makes it possible to label each segment with its characteristic attribute.

A total of five customer segments were identified for the client. The first and largest segment comprises customers with more than 100 purchases each — a very valuable segment. The recommendation given to the client was to do everything possible to keep these loyal customers happy.

The second identified segment consists of buyers who make 85 percent of their purchases in the online shop — a rate about 26 times higher than normal for the customers of the company. These buyers also seem to purchase more items per order. Digital marketing strategies may be the best way to address this segment and could be used to provide information about new products in the online shop.

The third segment also exhibits a high rate of online purchases. However, the return rate is also 13 times higher than the average, and products are often bought one at a time. This indicates that these customers are taking excessive advantage of the free shipping and returns policy. To optimize costs and earnings, Arvato CRM Solutions recommended that the client evaluate the profitability of customers in this segment. Unprofitable customers could lose the free shipping options.

The fourth segment represents a large customer base with very traditional behavior. In general, these customers purchase less frequently, do not use the online shop and return products only rarely. Interestingly, this traditional behavior does not appear to be related to age, since the average age here is similar to that of the other segments.

The smallest segment is the fifth. It represents customers who place very large orders but purchase less frequently.

Customer segmentation based on stream clustering provides an ongoing picture of the makeup of the customer base. It also indicates the value that different customer groups have for the company and shows where increased marketing activities may be worthwhile. It is not limited to the retail/ e-commerce field, of course. It can be applied in other sectors, too.

It allows companies to lay the foundation for targeted marketing campaigns aimed, for instance, at rewarding loyal customers, preventing defections or gaining new customers. The segmentation also helps a company select the right communication channels. If the intention is to target online shoppers, for example, campaigns using social media and email are preferable to expensive direct mailings. In other words, this new approach to customer segmentation allows companies to reach the desired customers with the right messages via the right channels, which also improves the customer experience. And these benefits are achieved continuously, because updating the clusters on an ongoing basis using the streams eliminates the key disadvantage of traditional customer segmentation.

Author: Editorial team Future. Customer.
Image: © Andrey Popov – AdobeStock