Why the shakeout effect matters in CLV modeling

Customer lifetime value (CLV) is often treated as a static metric. 

In practice, it is shaped by how different types of customers behave – and churn – over time.

One of the most important dynamics to understand is the “shakeout effect,” where early churn removes lower-value customers from a cohort, leaving a smaller, more stable group with higher engagement and more predictable purchase behavior.

This article takes a closer look at the shakeout effect in CLV analytics, why it happens, and how marketers should account for it when evaluating churn, retention, and long-term profitability.

What is the shakeout effect in the context of CLV analytics?

Imagine a cohort of new customers. 

As time goes on, all the “bad” customers drop, leaving only the “good ones,” with low propensity to drop, more engagement, better product-market fit, and more predictable purchase behavior. 

Therefore, the overall churn propensity decreases over time. This is called the shakeout effect and is a byproduct of having heterogeneity across customers.

As far as time goes, analysts typically use one-year windows or examine all-time purchase history, but it depends. 

For businesses with monthly subscriptions, the window after the first 30 days is essential to analyze, as no purchases after 30 days means new customers have churned.

If you’re looking at overall probability to churn over time, you’ll see something along those lines.

If you break out retention rates across various dimensions, such as UTM medium in the example below, you start to see this heterogeneity. 

In this case, email as a first touch is associated with a higher retention rate over time, roughly 27% after 500 days, while Google shows a lower retention rate, roughly 18% after 500 days.

Dig deeper: How to use CRM data to inform and grow your PPC campaigns

Why should the shakeout effect matter to marketers?

Not all customers are equal from a CLV standpoint. 

Businesses often lose money on a large percentage of newly acquired customers who churn before they register a CLV high enough to justify acquisition costs. 

Profitability is often highly concentrated in a smaller segment of highly loyal customers.

If marketers don’t account for shakeout and conduct an analysis of churn over a reasonable period of time, they may either overestimate long-term churn, assuming early churn continues, or overestimate CLV if they ignore the early loss entirely.

A strong high-level view implements the Lorenz curve and the Pareto principle, showing that 80% of CLV comes from 20% of customers. 

It’s critical for businesses to identify this core loyal segment, understand what these customers’ demographics and behaviors look like, and what they specifically like about the brand and products. 

There may be more customers like them out there – and the data can produce insights to help engage them with smart targeting and messaging.

How to identify heterogeneity in your CRM

One of the easiest and effective ways to explore your CRM data and get a sense of what is driving CLV up or down is ranked cross-correlation analysis (RCC). 

As an initial take, we want to know whether there are features in the data that clearly show a lot of variance in terms of CLV.

In the example above, customers with above-average CLV: 

• Show high purchase frequency.
• Are subscribed to the newsletter.
• Made a purchase recently.
• Initially subscribed to at least one product. 

While some of these features are redundant, such as purchase frequency being closely tied to product subscription, this view does a good job of suggesting what the main CLV needle movers are.

Another simple way to get a feel for CLV across dimensions is to visualize the distribution of the data.

• Is it normal, left-skewed, or right-skewed? 
• What is the median CLV by frequency? 

In the example below, using a ridgeline chart, we can see that CLV distribution is right-skewed, with Brazil having the highest CLV, at $2,014, and India the lowest, at $820.

Which dimensions you choose to analyze depends on what’s available in your CRM. 

At the very least, examine purchase frequency, purchase recency, channel, geo, and product purchased. 

For B2B specifically, I recommend using job title, vertical, and type of account, such as SMB, enterprise, and high-growth. 

When marketing offers more ways for customers to engage, I also find utility in including yes-or-no dimensions for newsletter and SMS subscriptions.

More advanced statistical methods, such as collinearity analysis, stepwise regression, and random forest, help account for collinearity challenges and estimate the importance of each feature in the data. I’ll keep that for another article.

Dig deeper: LTV:CAC explained: Why you shouldn’t rely on this KPI

CLV takeaways from the shakeout effect

In a nutshell, savvy marketers should:

• Account for the shakeout effect to accurately estimate CLV.
• Use both descriptive and predictive analytics to understand and predict what is influencing CLV.
• Identify and dig up insights into their core loyal segment to find similar customers in the future.