Understanding Customer Churn

The secret to retaining customers in the face of what is often global competition is a key issue for many businesses. It is so easy for customers to 'play the field' these days, and they do! Both start-ups and established businesses experience this problem, but because understanding why it happens and what can be done about it is frequently seen as too difficult, addressing it often gets ignored and shoved in the too-hard basket.

So why do I have high customer churn rates?

Although not unique to start-ups, fledgling businesses often complain of high customer churn rates. This has been put down to various issues such as lack of loyalty to an unrecognized brand, the range of products on offer, inconsistent product quality, insufficient customer support and other issues related to possibly kicking off a business before enough of the key pieces are in place.

Established businesses, however, are also no stranger to high customer churn rates. Often, the causes are similar to those afflicting start-ups, but other factors are sometimes involved. Some of these are increased competition with other businesses, failure to innovate and adapt to changing market conditions, dissatisfaction with pricing changes and a failure to meet customer expectations.

Churn rates differ across industry

It appears that different industries suffer from different rates of churn. The following 2022 figures were derived from customergauge.com for US businesses.

Interestingly, it appears that despite the relative ease for their customers to find an alternative supplier, IT services and computer software industries have a relatively low rate of churn compared to the bricks-and-mortar industries.

This might be due to high-tech businesses being able to more easily identify and act on changes. After-all, writing some code to change the look and feel of a website is much easier than relocating a retail business closer to customers (for example).

Regular customer behavior monitoring is essential

Whether a business is a start-up or established, it is important to regularly monitor customer churn and try to determine the reasons for it. Acting on this intelligence will allow them to know where to make the most effective changes and improve customer retention rates.

So how does one go about doing this?

I prefer a data-driven approach, which allows for changes to be made in an experimental and systematic way. This usually involves first gathering lots of customer behavior-related data including purchase history, usage patterns, and internet site page-browsing history. Whilst you would think this would be a no-brainer for online businesses, it is unfortunately often the case that these companies have not in fact implemented processes to collect the correct data, making proper analysis difficult.

For offline – bricks and mortar – companies there are a number of ways to effectively collect analyzable data. These could include from sales and purchase data, assessments of in-store browsing behavior, loyalty program data and customer service call records. Whilst it might be slightly more difficult to collect this data than when a business is conducted online, there are certainly ways and means of doing so which are beyond the scope of this discussion.

Once the data has been collected, the next step involves conducting targeted behavioral analyses. This would involve running correlations between the identified behaviors and churn rates, and then using a combination of visualization and statistical analysis to identify trends, honing in on the factors that are likely to be the chief causes.

Unsupervised learning can give you actionable insights

I like to go one step further and implement an unsupervised machine learning model to identify natural behavioral segments within the customer population. Utilizing this approach, unthought of differentiators can often be found to be related to churn. By using the clusters as behavioral labels, predictions of whether and when a new customer will churn, can be made. By then running explainability analyses over the models, the extent to which each feature contributes to a particular behavioral classification can be articulated.

This approach can be quite rewarding, because you end up with actionable insights, allowing you to actually do something to try to fix the problem. An example might be finding out that a particular web page or message is usually the last one viewed before a customer churns, or that the average usage over the last 3 months usually declines precipitously just before a churn. In both these cases, churn can be anticipated, and attempts made to prevent it. i.e. either change the look and feel of the website or make some added attempts to target the users who are reducing their usage.

If you are having trouble working out why you are having difficulty retaining customers, or you have a machine learning challenge that needs a look, please feel free to reach out to continue the conversation.

#analytics #clusteranalysis #unsupervisedlearning #elad_data