Calculating churn is hard. There are so many things that you need to take into account!
Wouldn’t it be nice to have a practical walk-through on how to calculate SaaS churn? I couldn’t find one so I decided to create it myself.
Churn tells you how many customers you have lost. Churn is not a standardized metric – you can calculate it how you want.
The first step is to decide which customers you are going to include in your calculations. For that, you’ll need to define an “active” and “lost” customer.
Here are the most commonly used definitions for active and lost customers when calculating SaaS churn.
“CAN’T I JUST TAKE A COUNT OF NON-CANCELLED SUBSCRIPTIONS ONCE EVERY MONTH?”
For most of you, that will not work. You’ll want to normalize for growth. We’ll get back to that in the next post.
This is the most common definition. It’s the way i.e. Kissmetrics offers you by default.
It has several benefits:
Leaving out the trial users is a good thing:
You’ll want to measure your marketing funnel conversion rates separately. This way the measurements don’t overlap and seasonal ups and downs in marketing funnel conversion rate don’t shake your churn.
There is also one drawback in this definition. Your free plan subscribers don’t pay you. Based on this definition your free plan subscribers are not your active customers. That means you cannot i.e. calculate CLTV for all customers – only for the customers who paid. Yes, free plan users have CLTV too – it’s negative.
CLTV is not Customer Life-Time Revenue. CLTV is estimated Customer Life-Time Profit. Why the confusion? By definition, all “profit” measures are historical. Unable to use the word “profit”, some smart guy named this metric Customer Life-Time Value instead. Gee, what a great unambiguous choice! The respective historical “value” is called Customer Profitability. Many SaaS owners calculate just Life-Time Revenue, not full CLTV. That’s fine if your profit margins are super high.
Involuntary churners are people that you kicked out. You’ll want to kick out people who have consistent credit card problems, people who break the law (or usage agreement). In some cases you might want to kick out people who are not profitable.
Companies often want to track only voluntary churn. The problem is, if you just cancel people’s account as you kick them out, they are going to end up into the churn figures. To keep them separate, you’ll need to give them some other status than cancelled, i.e. “expelled”, “fired” or “sacked”.
You can calculate churn for a free plan by measuring real customer activity. But it comes with a price. You’ll have to build a system to measure, store and analyze customer activity data.
Many people who sign up for a trial never come around and actually try out the app. When you measure real activity, those people get dropped out of the calculations. Your churn rate will look better than if you would have included all people who signed up. As you follow up your marketing funnel separately you will not lose any information either.
But… pure activity based system makes it really hard to communicate your metrics with others. “Oh no, this is not the amount of signups or paying customers, we are only counting people who created a canvas and added at least 5 items to it and attached the tracking codes to their systems.”
For that reason many companies who use activity based system still leave the trial users out of the calculations.
This definition looks easy, but in the end it’s not. You’ll face most of the same challenges than when measuring real customer activity, because you’ll need use real activity to measure lost customers.
As the trial users are in the count, any seasonal variations in your trial conversions will be visible in churn. Then you’ll be wondering whether your old customers are leaving you, or if it’s just the season when you lose more trial users.
It’s easy to know when a customer is lost when we calculate churn for paying customers. Customers often make the decision to cancel long before they act, but we still get the cancellation pretty soon, unless the customer has a yearly subscription.
When you are calculating churn for a free plan, it is much harder to say for sure when the customer has left.
Trying to use plan cancellations to measure lost customers doesn’t usually work. People on free plan don’t cancel their subscriptions. They’ll just never come back.
You’ll have to come up your business-specific means to measure who are your active customers and who are just ghosts of the customers gone by.
Before you have enough activity data, you are forced to fall back using something like “hasn’t logged in within last 90 days”. Eventually, you’ll probably use something like “hasn’t performed key activity within last 45 days”. But it depends on your product and how your customers use it. In any case it often takes much more time to be certain that a customer has been lost.
You can use the following flow-chart to decide which definitions are the best for you.
Click the image to view it full size[pinit]
In general, counting paying customers is a good starting point. The paying customers are the heart of your business and you already collect the payment data.
Even if you would not use Kissmetrics, MixPanel or some other tool to track your marketing funnel conversions, it is easy to dig up trial-to-paid rate separately from your customer DB.
For lost customers, both cancelled+downgraded and non-paying customers are good options. It is a smart move to store the information on why the customer left, even if you would not be using it right away.
You’ll most probably want to keep calculating paying churn even if you end up tracking customer activity. Why? Because of the difficulties to recognize lost customer fast enough. In fact, unless your cost of service is high, you’ll probably don’t build an activity tracking system to calculate churn. You’ll build it to learn more from your customers and to predict when they are going to leave you. Then you can act before they churn.
Whatever definition you’ll use, use it consistently in all customer-related business calculations. You can’t, for example, decide to calculate churn using paying customers and then try to use it to calculate CLTV for every customer who signed up. (Except, being the master of your online empire you can totally do whatever you want – you’ll just get fucked up results)
This was actually one of the hardest decisions in calculating churn. Next time, we are going to have fun with segments, cohorts and different churn types.
Have you ever wondered why there are so many different Churn Rates? What’s the point?
Let’s find out! Even though this article does not answer a direct business question, it’s important to understand how the Churn Rates work.
When you understand the differences, you’ll be able to answer questions like:
And more importantly – you can be confident that you answers aren’t wrong.
The table below lists the Churn Rates and their features. ￼
Subscription, User and Customer Churn should be calculated separately if necessary. They are used for the same purposes and they are often the same number, so only Subscription Churn is listed in the table.
Normalization means that the effect of time is taken into consideration.
When talking about Churn, normalization requires 2 actions.
Normalized Churn Rate:
Normalized rates tell you about the behavior of a single average customer.
The non-normalized rates tell you about your business as whole.
That means we’ll use Net MRR Churn Rate for financial projections and estimating how the whole customer base behaves.
And we’ll use Subscription and MRR Churn Rates for estimating average customer behavior.
FirstOfficer.io shows Net MRR Churn Rate, MRR Churn Rate and Subscription Churn Rate.
You can – as long as you offer just one plan and no annual subscriptions.
The one Churn Rate you’d start with is the Net Subscription Churn Rate – lost subscriptions compared to the subscriptions you had at the beginning of the month.
Net Subscription Churn Rate is easy to calculate and can be used for almost anything – but its meaning changes when your business develops.
It gradually stops working for the purposes you initially used it for.
When you add more plans, you should start using MRR Churn Rates so that you can see if people in different plans behave differently.
When you add annual plans, you should start using normalized Churn Rates when you want to calculate unit metrics like Customer Life-Time Value (CLTV).
You might get worried that your business performance is crashing – when you just had lots of annual subscriptions up for renewal. Or you may believe that your CLTV values are 3-5 times higher than they actually are.
The effect is not theoretical – I’ve seen cases like above.
Here is an example:
In this example your overall customer base shrinks at 10% monthly rate. Average customers however, when having an opportunity to churn, leave at 20% monthly rate.
Now we can look at the questions in the beginning of the article again.
What if… you stopped getting new customers? How long would it take to lost all the monthly subscription revenue?
This question concerns your whole customer base and you are interested in money (vs. subscriptions). That means we use Net MRR Churn.
Just invert it… (1/churn rate) *100% … and you have the answer in months.
Here are a couple of pre-calculated values:
It’s not the most accurate formula, but works fine with rates smaller than 10% and it’s a rough estimate anyway. In practice, you’d use this churn rate to create a projection which includes also the expenses and your cashflow.
Net Churn Rates are sometimes volatile, so it’s best to take an average churn rate from a longer period and check that the annual customers are evenly distributed throughout the year.
This question is about the average lost customer – so we need the normalized Churn Rates.
One of our Churn Rates measures money, the other people. If we lose more people than money, we lose low-value customers. And if we lose more money than people, we lose high-value customers.
When we compare the MRR Churn Rate to Subscription Churn Rate and the MRR Churn Rate is bigger, you lost high-value customers. If the MRR Churn Rate is the better one, you lost low-value customers.
The simple answer is ‘No’.
The non-profit-based CLTV should always be adjusted with gross profit margin – and even after that it is often not reliable even when calculated right. As you just learned, when your churn rates are low enough, CLTV tries to estimate behaviour years onwards.
Jason Cohen has an excellent article on this topic: The mistake of 1/c in LTV calculations.
That said, if you choose to use CLTV, here’s the key to assess it:
The calculations are for your average customer.
If you mostly lose high-value customers, the CLTV will be overestimated.
If you lose low-value customers, the CLTV will be underestimated.
So you can actually use the answer from the previous question to answer this one.
If your MRR Churn Rate and Subscription Churn Rate are similar, the CLTV’s are trustworthy. But also check that the ARPU of new customers isn’t very different from the ARPU of all customers.
Often a single CLTV value for whole customer base is useless unless you offer a single monthly plan. Most SaaS businesses should use plan-specific CLTV values.
It’s also worth to follow up the realized Total Contract Values (TCV) – sometimes even per cohort.