The Pitfalls of Churn Rate. Setting the Scene

Setting the Scene

The shift from transaction-based and advertising-based businesses towards subscription is undeniable. The most valuable companies are attempting to build direct-to-consumer (“DTC”) subscription services. Apple Music has 36 million subscribers, and a separate video service coming next year. Over the past few years, Google has launched two DTC services: YouTube TV and YouTube Red. Amazon is attempting to bundle everything under the sun within Prime (60M — 90M subs). With unrelenting focus, Netflix is the gold standard with over 120 million paid subs.

This is the arena Spotify is entering as it prepares to become a public company. Behind Netflix, Spotify may be the most successful DTC subscription service. Its 70 million Premium subscribers is almost double Apple Music. With 60% of new subs coming from their free, ad-supported service, Spotify’s approach to freemium is exceptional. Each month, 160 million people listen to music on Spotify for an average of 25 hours per person!

For others stepping into this arena, it would be prudent to learn from the pioneers. Spotify’s recent F-1 filing is full of insights and details how they track performance. One of the most critical parts describes their improving churn rate and how that has boosted subscriber lifetime value (“LTV”).

LTV is the most important metric for subscription services. The concept of LTV, while complex, is critical to understand, and worthy of its own post. For now, let’s assume it’s important and that we should be thoughtful in its calculation. On that note, Spotify is making a common and dangerous mistake in their LTV calculation (emphasis mine):

Churn rate is the most common metric used to evaluate retention. But churn rate has inherent flaws that mislead companies towards unfounded reasoning and decision-making. Spotify is not unique in their dangerous mistake of using churn rate to determine LTV. There are other common misuses of churn rate beyond using it to determine LTV. Investors and entrepreneurs should shift their attention towards a superior metric: retention rates.

We dive into all these aspects in the post below, broken out into a few sections. Our first step is to define churn rate & retention rates, setting the stage for the rest of the post. The next section uses fictional stories to illustrate the following pitfalls with churn rate:

Talking Semantics: What is “churn rate”? What are “retention rates”?

Churn rate and retention rates are both important growth metrics. High churn and inferior retention are often driven by some combination of:

Services with unfavorable churn or retention have a hard time keeping people subscribed. With rampant cancels, these services burn through their addressable market. These services become more reliant on costly customer acquisition to fuel growth. The leaky bucket syndrome is a common catalyst for vicious cycles, which often end in demise.

While churn and retention rates are both useful, it’s important to use them in a thoughtful way. To do this, a fundamental understanding of the two metrics and how they are distinct is essential. Let’s begin by defining churn rate.

Churn Rate

Churn rate is an aggregate metric, measuring total cancels relative to total subscribers. Assuming a stable churn rate, we can predict next month’s cancels. However, churn rate is rarely stable beyond a few months, making it difficult to use for long-term predictions. More important, churn doesn’t provide insight into who is cancelling or their reasoning.

There are two popular ways to measure churn rate. In each method, the numerator is simply the total number of cancels for a period of time. In one method, the denominator is the average number of subscribers during same time period:

The other method uses a different denominator. It takes the sum of the beginning balance of subscribers, plus new subscribers added during the period:

The first method looks at cancels relative to average subs that could have cancelled. The second method looks at cancels relative to total subs that had the opportunity to cancel. Either methodology works, but it’s important to maintain the same methodology over time.

Retention Rates

Contrary to churn rate as an aggregate metric, retention rates are a per-usermetric. Retention rates track a group (i.e. cohort) of subscribers over time. They measure how much of the cohort remains after each successive month or billing cycle. The output is the average retention rates for that segment of subs, expressed on a per-sub basis.

As an example, we could isolate all new paying subscribers that joined in January 2017. We then track how many don’t cancel and make it through the second billing cycle, third, and so on. We can then combine all cohorts to determine our subscriber decay curve. Finally, we can use the decay curve to project longer-term retention and customer lifetime.

It’s important to establish a common rule for subscription businesses. Subscribers are most likely to cancel in the initial months of their subscription. Moreover, the likelihood someone cancels decreases rapidly the longer they’re subscribed. This produces an exponential decay curve. This is a crucial concept to understand as it causes most defects associated with churn rate.

Shortcomings & common misuses of churn rate

The simplicity of churn rate has made it the de facto metric to measure retention. However, this simplicity presents formulaic dependencies that can produce confounding and deceptive results. These dependencies also make it challenging to track churn over time, or compare churn between services.

Most important, churn rate is independent from LTV, so the two can move in opposite directions. By definition, retention rates align with per-user profitability over time, making it the backbone of LTV. The close linkage between retention rates and LTV is what makes it a critical metric. Contrarily, churn’s independence from LTV is what makes it dangerous.

The examples below illuminate a few common shortfalls associated with churn rate.

1) Churn rate is completely independent of improvement in LTV

This is the most important shortfall: an improvement in churn is not dependent on an improvement in average customer lifetime. Churn can (and often will) decrease without any improvement in average customer lifetime. As an extension, all else equal, LTV can deteriorate alongside improving churn. The disconnect between LTV and churn rate limits the usefulness of churn rate.

Let’s use an example to clarify churn’s independence from per-user retention and LTV. The image below shows annual net subscription growth (left), cancels (middle), and churn (right). There are two important assumptions beneath the performance of this service:

With no change in the subscriber decay curve, the same percent of new subs cancel in their 1st year, 2nd year, and so on. Combining a static decay curve with no change in new subs produces the layer cake of annual cancels in the left chart. The bright red segment represents subs cancelling in the first year of their subscription. This segment remains the same for all five years (7.5k cancels for Year 1, Year 2, etc.). In other words, the same relative share of the 12,000 new subs added each year are cancelling in their first year.

Total cancels are growing over time due to the aging of older cohorts. During Year 2, we have subs cancelling in the first year and second year of their subscription. The latter is what drives the increase in cancels during Year 2. This dynamic continues each year, with an incremental layer added on.

Alongside no growth in new subscribers, this places downward pressure on subscriber growth. In the middle chart, there’s a gradual descent in net subscription growth during the 5 years. More important, with no improvement in the decay curve, LTV is remaining constant. Absolute subscriber growth is decelerating and per-user value isn’t improving. This is not the growth story you want to tell investors.

That said, churn rate is experiencing a precipitous decline! So at least the team has that going for them…

2) Churn spikes during periods of exceptional subscriber growth

There’s an important byproduct of an exponential decay curve that erodes churn’s effectiveness. Periods of elevated growth in new subscribers will produce periods of elevated cancels. This is because there will be an influx of subs in their first billing cycle, when they are most likely to cancel. These periods of strong subscriber growth are often accompanied by a spike in churn rate.

Let’s walk through a hypothetical example using the images below. The left image shows monthly new paying subscribers, cancels, and net growth. The image to the right shows monthly churn rate. Notice the exceptional subscriber growth in March, which is well above other months.

However, given the growth in first-month subscribers, there’s a complementary increase in cancels. The sharp increase in cancels causes monthly churn to spike. If we over-fixate on churn, we will overlook the unprecedented subscriber growth. More important, we have no idea how long the new subscribers during March are sticking around.

3) Growth stage & other factors impact the application of churn rate

Churn rate is often used to compare retention across different services. In doing so, it’s crucial to consider growth stage of the different services. A recently launched service will have a high share of subs in the initial months of subscription. An established incumbent will have a higher share of tenured subs that are unlikely to cancel. Shifting perspective to a single service over time, we can also conclude that churn should go down over time.

One more hypothetical example. The table below shows churn for 3 services in different periods of the product life cycle. All experience the same number of cancels for this particular period of time. The launch company’s churn is almost double the incumbent, but has more attractive growth. More important, the launch service added 6x as many new subs with the same cancels as the incumbent. This indicates the launch service has superior retention & customer lifetime.

In addition, the type of relationship the service has with its subs has a significant impact on churn. A service with multi-year agreements and early cancel fees (i.e. cable & telecom) will have low churn (1%-2%). Monthly services that promote “cancel anytime” (Netflix, Spotify) will have higher churn. The structure of the business models are different, making comparisons challenging.

4) Do NOT use churn rate to determine LTV

To bring it all together, let’s talk about the most dangerous use of churn rate: using it to calculate LTV. A common method for determining average customer lifetime is to divide 1 by the churn rate. For example, Spotify’s monthly churn rate of 5% would imply a 20 month average lifetime (1 / 5% = 20 months). We assume the probability a sub cancels is the same each month (5%), and measuring how long it will take to reach 100% (or “1”).

This is a dangerous method to calculate average lifetime for a few reasons:

Determining customer lifetime is the backbone of the LTV calculation. Using churn rate to determine customer lifetime is an oversimplified method. The simplicity amplifies the margin of error, increasing the likelihood of being duped. Don’t be duped, use retention rates and the decay curve to calculate customer lifetime.

Summary

Improving retention and extending customer lifetime creates tremendous value for companies. There is no silver bullet when it comes to improving user retention. It boils down to improving the perceived value of your service relative to its cost — an obvious and far-reaching challenge. Improving customer lifetime is an ongoing, cross-company effort. It must be an integral part of the team’s culture.

Nothing is more dangerous to this ambition than using a flawed metric to track retention. For churn rate to be an effective metric, it requires an understanding of its limitations. Further, churn should be one component of a holistic toolkit, and retention rates should be your primary tool.