Tracking Growth Metrics for Subscription Services #1

(1) Subscriber Metrics

Hi there, I'm Leo, a Data Analyst at DelightRoom. When I first joined DelightRoom, one thing that felt pretty unfamiliar was how completely different the data structure for a subscription service is compared to the regular e-commerce I was used to. Because of that, I definitely went through some trial and error while designing our metric analysis framework. For instance, when analyzing revenue in standard e-commerce, just breaking down daily transaction amounts by various dimensions like buyers or products can yield pretty meaningful results. But for a subscription service, you have to factor in several new variables like subscriber retention, changes in subscription products, and the presence of trials. So, what kind of framework can we use to analyze metrics for a service with a subscription model?

Growth metrics for a subscription service can be looked at from two different angles:

• Centered around 구독자(Subscriber)

• Centered around 매출(Sales)

Some of you might be wondering what the difference is. You might think, "If subscribers increase, revenue increases anyway, and if revenue drops, it's no different from a drop in subscribers... so what's the difference?"

Here's why this distinction is important:

1. Some events that happen in a subscription service affect only one of these metrics. When a specific event occurs, these two metrics don't always move together.

2. In a subscription service, revenue metrics tend to lag slightly behind subscriber metrics.

3. There are plenty of cases where changes in subscribers and changes in revenue aren't connected at all.

1) Specific events might affect only either subscribers or revenue

Let's imagine a service that sells a subscription product billed at 1,000 KRW every month.

• New subscription event occurs → Subscribers increase by 1. Revenue also increases by 1,000 KRW.

In the case above, subscribers and revenue move together. But what about the following scenarios?

• Trial event occurs → Subscribers increase by 1. Revenue doesn't change.

• Trial period ends and a payment event occurs → No change in subscriber count. Revenue increases by 1,000 KRW.

• Subscription renewal event occurs → No change in subscriber count. Revenue increases by 1,000 KRW.

• Turn-off auto-renew event occurs → Subscribers decrease by 1. Revenue doesn't change. (It doesn't affect the revenue that has already been generated; it just means no payment event will occur in the future.)

Because changes in subscribers and revenue trigger at different times like this, it's not ideal to assess the health of a subscription service solely based on revenue. (The same goes even if you break down revenue into formats like MRR.) This is exactly why you shouldn't focus only on changes in transaction counts or revenue when measuring the impact of an ongoing experiment or service improvement.

2) The time gap between subscriber metrics and revenue metrics

Revenue metrics in a subscription service tend to lag slightly behind subscriber metrics. When a trial user comes in, they convert into an actual subscriber after a certain period, which generates revenue. Conversely, when a turn-off auto-renew event happens, it indirectly affects revenue after a certain period by "not making a payment." (Since it doesn't generate negative revenue, judging it as an impact can be a bit ambiguous in itself.) Even if both are ultimately affected, the moment they are triggered can be different.

3) Cases where subscriber count and revenue aren't connected

We can easily think of cases where revenue changes even if the subscriber count doesn't change at all. A prime example would be changing the product's price. If the price changes, revenue naturally changes even if the subscriber count stays the same. The same goes for when the product options that subscribers choose change. If an event to upgrade or downgrade a subscription product occurs, the subscriber count remains unchanged, but revenue can shift.

As a result, you need to look at subscription service metrics for subscribers and revenue independently, examine the leading metrics that affect each, and understand the missing link between the two. In this article, I'm going to focus specifically on subscriber metrics.

What's the most crucial metric related to subscribers? I think many people would say 구독자 수. While subscriber count is ultimately important, the number itself is a stock metric, meaning its resolution is too low to be used on its own. To get a more intuitive look at subscriber changes or the growth trajectory over a specific period, it's better to look at a flow metric like 구독자 수 증감(Diff).

One thing to keep in mind here is that you need to look at metrics for 'paid' subscribers. (This is different from the total subscriber count, which includes trials.) A trial user hasn't fully received the added value yet, so while they are a variable that impacts paid subscriber count, they shouldn't be lumped into the paid subscriber count itself. An increase or decrease in trial subscribers carries a different meaning than a change in paid subscribers. For example, an increase in trial subscribers is a positive signal, but a decrease is a neutral metric that's hard to label as strictly good or bad on its own. (If trial subscribers dropped because they converted to paid subscriptions, it's a positive metric, but if they dropped because their trial expired, it's a negative metric.) Therefore, if your goal is to see if the service is growing well, it's best to look at the change in paid subscribers, excluding trials. (Since just saying 'subscribers' might confuse some folks, I'll use the term 'paid subscriber' moving forward.)

So what are the factors that influence the net change in paid subscribers? If you think about it, there are really only two things that affect it. (Which is pretty obvious...)

• Increase in new paid subscribers

• Churn of existing paid subscribers

Even if you don't get a single new paid subscriber, as long as no existing paid subscribers churn, your paid subscriber count won't change. If you gain 10 new paid subscribers and 5 existing ones churn, your paid subscriber count ultimately increases by 5. As you can see, changes in paid subscriber count are determined by the movement of just two variables: 신규 유료 구독자의 증가 and 기존 유료 구독자의 이탈. Metrics like Install, SignUp, DAU, Retention, and ARPU might indirectly influence the paid subscriber count (when combined with other factors), but they aren't variables that directly impact it on their own.

→ As a result, if you just break down and closely monitor those two variables—the increase in new paid subscribers and the churn of existing paid subscribers—you can easily check and manage the net change in paid subscribers.

The diagram above illustrates everything I've just explained. Based on this framework, let's lay out the specific metrics we can monitor to check the health of our subscriber metrics.

Flow metrics: Metrics showing changes over a specific period

[User]

• A → Net change in paid subscribers (Change in paid subscriber count. Calculated as B minus C)

• B → New paid subscribers

• C → Canceled paid subscribers

• D → New trial subscribers

• E → New users (Criteria defined by sign-ups or installs, etc.)

[Conversion]

• 1 → Install to Paid Subscription Conversion (Conversion rate from install to paid subscription. At DelightRoom, we call this I2P)

• 2 → Install to Trial Conversion (Conversion rate from install to trial. At DelightRoom, we call this I2T)

• 3 → Trial to Paid Subscription Conversion (Conversion rate from trial to paid subscription. At DelightRoom, we call this T2P)

• 4 → Existing user's Paid Subscription Conversion (Usually aggregated as a conversion rate, but if the rate is too low to easily manage changes, aggregating it as an absolute number works fine too)

• 5 → Paid subscriber cancellation (You can look at the overall cancellation rate based on all paid subscribers, but it's better to break it down into a monthly subscriber retention rate)

Stock metrics: Metrics showing a snapshot at a specific point in time

• F → Unsubscribed users among existing (not new) users

• G → Subscribed users

In practice, the Alarmy subscription dashboard we use at DelightRoom is set up to effectively break down these Flow and Stock metrics by country and OS. It's designed so that we can check the daily net change in paid subscribers first, and then closely examine the leading metrics for both new subscriber growth and existing subscriber churn.

So far, we've gone over the metric analysis framework for subscribers. Conceptually, it seems straightforward, but actually building the necessary data marts based on raw data piling up from app stores and organizing the dashboards and metrics to fit this framework is no easy task. I plan to put together a separate post about that story if I get the chance later. (The subscription data available on Google Play and iTunes Connect have completely different data formats, labeling, and status values depending on the platform, and their aggregation standards and shapes are all over the place... The data preprocessing journey was quite a rugged one -_-;;;)

Read the next post

-> Tracking Growth Metrics for Subscription Services #2. Subscription Revenue Metrics