Never Let Me Go: B2B Churn Analysis and Customer Retention
by Cyrus Safaie
“Why did they stop buying from us? Who are they buying from now? How can we get them back?”
It’s expensive to acquire new customers. Even if you can afford to obtain them, you must spend additional investment to bring them to the ideal level of customer life-cycle maturity. Don’t get me wrong: growth often demands new customer acquisition, but without high customer retention, all the effort given to acquire new customers merely offsets customer churn, often replacing longer-tenured, higher-profit customers with lower-profit ones.
When a customer stops purchasing from you, it’s not just one transaction that’s lost — it’s actually that customer’s total expected future sales!
Regardless of industry, customer retention and churn rate reduction is often many leaders’ main strategic objective given today’s highly competitive landscape. It’s not just about the sales organization and their quotas or the finance organization and EBITDA targets — changes in customer churn impact every aspect of the business, from supply chain (e.g. Perfect Order Index (POI), Fill Rate, Cost to Serve, etc.) to marketing (e.g. Campaign Management, MROI, Channel Management, etc.).
So what exactly is customer churn analysis?
Customer churn analysis is about finding patterns that suggest a customer may leave before they actually do, with the hope of identifying potential levers you can pull to keep them in the fold. Depending on the product/service or industry, “too late” can vary significantly — sometimes it’s a few days, sometimes multiple months.
As you may have guessed, there’s some science involved in mining these patterns and inferring drivers that can provide actual actionable insight. For years, modeling techniques like survival analysis and logistic regression, mostly selected for their easy interpretability, have been used to model customer churn. Breakthroughs in machine learning algorithms such as Deep Neural Nets and Tree-Based Ensembles, along with a significant increase in a typical business’s amount and variety of available data, have allowed many companies to greatly improve the accuracy of their churn models, as well as extend the time horizon over which they can be confident of their model’s results. These improvements empower businesses to take action before it’s too late, frequently saving millions in the process. However, some businesses have been reluctant to take the plunge and explore more advanced methods.
So why the low adoption rate of advanced machine learning techniques? One reason is a perceived lack of interpretability. And, perhaps, justifiably so: it doesn’t matter if we can predict a customer leaving unless we can complement it with a why. We need to be able to understand why they left in order to prevent it in the future. If your customer is not satisfied with the quality of your product, you won’t keep them happy by offering a discount; similarly, if customers are unhappy with how long your products take to be delivered to their doorstep, offering free delivery may not keep them around for long. That’s where advanced methods/metrics such as the Shapley Value and LIME can potentially help. In addition, text mining techniques can be used on customer emails, surveys, or customer service logs (plus many other sources of unstructured data) to understand why your customer is leaving and act preventatively.
Why are we excited about churn analysis in the B2B setting?
Let me start by clarifying that B2C (business-to-customer) churn analysis is probably more popular than in B2B (business-to-business) industries. It may be due to the higher level of analytics maturity in B2C companies, or simply that their survival depends on understanding their customers. Either way, that dynamic is changing as B2B companies discover the power of analytics and feel more pressure from competition than ever before.
Data availability has always been a challenge for most B2C companies, especially those with low e-commerce share. That’s one of the main reasons B2C companies invest in customer loyalty programs. However, such a strategy comes with many challenges of its own, such as customer targeting (should you focus on a customer or a household?) or fuzzy matching of credit card digits with loyalty numbers.
By comparison, B2B companies typically have very rich customer data. Vendors in this space often require that they be fully integrated into their customers’ internal systems before any purchases are processed. This installation ensures the vendor handles taxes, credits and payment terms, allowing them to collect vast amounts of data with ease. As a result, they know everything about the customers who purchased from them, including their tenure, life stage, sector, price sensitivity and more! In addition, the B2B customer purchase cycle and the assortment of interests are much more stable than in the B2C space, which results in an expanded ability to build churn models.
Customer churn analysis is both an art and a science, and a good understanding of its results can help your customer base and your bottom-line. Investment in this kind of analysis often has the support of all functional areas and, if executed properly, contributes positively to KPIs across an organization.