👋 Hey, it’s Sundar! Welcome to experiMENTAL where I share B2C marketing frameworks, how-to guides, and stories from 5 years of early Marketing at Uber and 15+ years at B2Cs.

While most companies focus on increasing retention, I prefer focusing on churn prevention. While you might think that’s the same, customers are looking for reasons to stay and are often disappointed by some part of the experience (onboarding, transaction, support, etc.).

But, before you can create a strategy around churn (or retention), you first have to define what churn is. That’s not always so straightforward. To help you define churn for your business, I’ve brought in guest author Drew Maniglia. 

Drew Maniglia has been focusing on customer churn for over 15 years. Prior to starting his startup, Drew built and led the data science team at Roku, implemented the churn models for Barnes & Noble’s Nook platform, and developed churn measurement for financial services companies.

He’s now the founder of Hakuin, a customer churn measurement and prediction company, helping B2C companies across media, e-commerce, fin-tech, and health-tech to improve customer lifetime value.

Disclaimer: Hakuin is not a paid sponsor and this is not an endorsement for Hakuin

What is Customer Churn? (✏️ to Drew)

Customer churn is one of the core dynamics playing out in every business. It’s simply a question of magnitude – just how many steps back for each step forward? At its most basic, customer churn is friction. It’s the headwind in the face of customer growth. It can also be compared to heat loss, or electrical resistance, or any other countervailing force that occurs in nature.

There’s so much human energy that goes into creating a product or service, bringing it to market, and recruiting customers. Churn is nothing other than the erosion of that energy.

Here’s a top-level definition that applies to all businesses: Customer churn occurs when customers turn off, depart from, or in any other respect cease to purchase our products or services for a prolonged period.

We’ll discuss below how this generic definition applies to different business models. Of course, the reason to define churn is to manage and reduce it. All metrics, churn and otherwise, should exist in service to action.

A useful churn definition helps companies:

1. Improve Customer Acquisition – Once we define churn, we can analyze and compare churn rates for different types of customers. The most relevant customer segment is the customer acquisition channel - how does acquisition relate to churn? We can update our acquisition strategies to focus on lower churn sources.

2. Improve Revenue Planning — We use churn analysis to project the number of customers and volume expected to churn out from our existing customer base in upcoming periods. 

3. Predict Customer Churn – Once we’ve defined churn, we can use machine learning to predict it at the user level. If we can predict it accurately, we can try to preempt it. 

But, before diving into churn definitions, we need to better understand customer cohorts which are the foundation of all churn analyses.

Customer Cohorts

A customer cohort is a group of customers who first engaged with your business within the same time period. By grouping customers into a cohort, we’re standardizing some of their user experience – the state of your product when they first engaged,  marketing campaigns, competitor activities, and overall market and economic sentiment. Think of the cohort of students who started kindergarten in September 2020. This cohort of students had a unique experience, and we’ll want to compare the academic performance of this cohort with other cohorts to understand what happens when your first introduction to school is remote.

Time is an key factor in any situation, and cohorts account for customer timing. When B2C subscriptions report on churn, they count the cancellations per cohort. This is what enables us to build churn curves and compare over time:

Let’s dive into how we should define churn based on your business model + product.

B2C Subscription 

When we hear “customer churn”, many immediately think of consumer subscription services like Netflix. For B2C subscriptions, customer churn comes with an explicit cancellation event. B2C subscription boils customer data down to just 2 events – subscription start and cancel:

When B2C subscriptions measure churn, they’re counting cancellations, by week, month or quarter. It’s fine to count these cancellations overall (e.g. 100 cancellations in December 2024) but more useful to count them within their customer cohort. 

B2C Subscriptions with Frequent Reactivation

Some B2C subscriptions benefit from an additional churn definition to capture the reactivations from churned subscribers. For many subscriptions, a cancellation is pretty near final for the majority of subscribers: a cancelled subscriber is probably not coming back any time soon.

For some subscriptions though, there’s a large resubscription rate, and the existence of intermittent subscribers gives rise to a secondary definition: Cancellation without reactivation in the following months. Think of streaming media companies like Disney+ or Peacock – it’s common for subscribers to opportunistically turn the sub on and off throughout the year.

To illustrate further, if a subscriber cancels, takes 3 months off, and then reactivates, is that really churn? If that’s common behavior, we may benefit from a broader view and define the B2C subscription starts and churns in much the same way that we would define a transaction-based churn. Each period of active subscription is a transaction (instead of shipping a product from an e-commerce site, we’re shipping a month of membership). In this definition, churn is the prolonged period of cancellation after which the subscriber is unlikely to organically re-subscribe. If the subscriber cancels for 1 month, there may be a high probability or return. However, as time goes by, we can eventually consider the subscriber gone – we no longer expect the return.

B2C Transaction

Similar to the above, when there is no explicit cancellation event, we need to define the period of customer absence after which an organic transaction has low probability. This is how we should define churn for transaction and usage-based businesses (e-commerce, ad-supported models, etc.) The customer should be considered churned once this period of absence has fully elapsed. The key variable here is the time period that defines churn, and it is a function of business dynamics and customer velocity. 

This time window is dependent upon the goods or services at hand. As the price of the goods increases, the transaction cadence usually declines: I’ve worked with online luxury retailers who expect customers to transact only 1x annually, and we’ve also served streaming media companies, where 14 consecutive days of absence means the customer won’t be back. 

When determining how long this transaction churn window should be, bear in mind that it needs to be long enough such that only a small share of customers who are lapsed for that period will return organically. A huge portion of Target’s customers probably go 1 day without a transaction, so it doesn’t make sense to define churn as 1 day of absence.

We’re trying to identify the volume of customers who will not be coming back spend with us again – lost value.  At the other end of the spectrum, five years of absence is too long to be useful, even if it covers 99.9% of customers (99.9% of customers who do not transact within 5 years are churned). We recommend defining the lapse period as that duration after which only ~3-5% of customers will return to make a purchase in the following period.

A quick note on low-transaction businesses: if an e-commerce platform only sees occasional transactions, it’s useful to infuse your churn definition with non-transaction engagement touchpoints. These most often come from the world of digital marketing. We recommend including events like email opens, site logins, and other such attributes to add more data. This way we don’t need to wait forever to declare a customer churned.  We can update our churn definition from something like “no purchase in 12 months” to “no engagement in 4 months,” of course first establishing that the marketing and site engagement can serve as a reliable proxy. This gives us a more handy tool to work with.

Data Structure

Unspoken here is the need to access and structure data. Everything mentioned depends upon analysis. We don’t need to go overboard with complexity, but we do need a way to manage and query data. The cohort framework mentioned above is a data minimalism at its best – you get a lot of mileage using something simple. The backbone of customer cohort analysis is a customer data-mart or feature store that puts useful information about your customer into a single place for ease of segmentation analysis: 

Every exercise like this is simple on its surface, but there’s always complexity under the hood.

Wrapping up (✏️ back to Sundar)

Defining churn is one of the most important exercises for a company because it immediately defines what good retention is. Defining churn should never be a hand waving exercise. Use data and knowledge of your customers to make a reasonable assumption of “this is the point of no return…usually…for like 95% of my customers”.

To learn more about churn visit https://hakuin.ai or email [email protected] to get in touch.

Missed my last newsletter?

Don’t be sad. Here it is: Why Last-Click attribution will never go away