Category Archives: churn

Important Subtleties in Calculating Quarterly, Annual, and ATR-based Churn Rates

This post won’t save your life, or your company.  But it might save you a few precious hours at 2:00 AM if you’re working on your company’s SaaS metrics and can’t foot your quarterly and annual churn rates while preparing a board or investor deck.

The generic issue is a lot of SaaS metrics gurus define metrics in a generic way using “periods” without paying attention to some subtleties that can arise in calculating these metrics for a quarter vs. a year.  The specific issue is, if you do what many people do, that your quarterly and annual churn rates won’t foot — i.e., the sum of your quarterly churn rates won’t equal your annual churn rate.

Here’s an example to show why.

saas churn subtle

If I asked you to calculate the annual churn rate in the above example, virtually everyone would get it correct.  You’d look at the rightmost column, see that 2018 started with 10,000 in ARR, see that there were 1,250 dollars of churn on the year, divide 1,250 by 10,000 and get 12.5%.  Simple, huh?

However, if I hid the last column, and then asked you to calculate quarterly churn rates, you might come up with churn rate 1, thinking churn rate = period churn / starting period ARR.  You might then multiply by 4 to annualize the quarterly rates and make them more meaningful.  Then, if I asked you to add an annual column, you’d sum the quarterly (non-annualized) rates for the annual churn and either average the annualized quarterly rates or simply gray-out the box as I did because it’s redundant [1].

You’d then pause, swear, and double-check the sheet for errors because the sum of your quarterly rates (10.2%) doesn’t equal your annual rate (12.5%).

What’s going on?  The trap is thinking churn rate = period churn / starting period ARR.

That works in a world of one-year contracts when you look at churn on an annual basis (every contract in the starting ARR base of 10,000 faces renewal at some point during the year), but it breaks on a quarterly basis.  Why?  Because starting ARR is increasing every quarter due to new sales that aren’t in the renewal base for the year.  This depresses your churn rates relative to churn rate 2, which defines quarterly churn as churn in the quarter divided by starting-year ARR.  When you use churn rate 2, the sum of the quarterly rates equals the annual rate, so you can mail out that board deck and go back to bed [2].

Available to Renew (ATR-based) Churn Rates

While we’re warmed up, let’s have some more fun.  If you’ve worked in enterprise software for more than a year, you’ll know that the 10,000 dollars of starting ARR is most certainly not distributed evenly across quarters:  enterprise software sales are almost always backloaded, ergo enterprise software renewals follow the same pattern.

So if we want more accurate [3] quarterly churn rates, shouldn’t we do the extra work, figure out how much ARR we have available to renew (ATR) in each quarter, and then measure churn rates on an ATR basis?  Why not!

Let’s first look at an example, that shows available to renew (ATR) split in a realistic, backloaded way across quarters [4].

ATR churn 1

In some sense, ATR churn rates are cleaner because you’re making fewer implicit assumptions:  here’s what was up for renewal and here’s what we got (or lost).  While ATR rates get complicated fast in a world of multi-year deals, for today, we’ll stay in a world of purely one-year contracts.

Even in that world, however, a potential footing issue emerges.  If I calculate annual ATR churn by looking at annual churn vs. starting ARR, I get the correct answer of 12.5%.  However, if I try to average my quarterly rates, I get a different answer of 13.7%, which I put in red because it’s incorrect.

Quiz:  what’s going on?

Hint:  let me show the ATR distributed in a crazy way to demonstrate the problem more clearly.

atr churn 2

The issue is you can’t get the annual rate by averaging the quarterly ATR rates because the ATR is not evenly distributed.  By using the crazy distribution above, you can see this more clearly because the (unweighted) average of the four quarterly rates is 53.6%, pulled way up by the two quarters with 100% churn rates.  The correct way to foot this is to instead use a weighted average, weighting on an ATR basis.  When you do that (supporting calculations in grey), the average then foots to the correct annual number.

# # #

Notes:

[1] The sum of the quarterly rates (A, B, C, D) will always equal the average of the annualized quarterly rates because (4A+4B+4C+4D)/4 = A+B+C+D.

[2] I won’t go so far as to say that churn rate 1 is “incorrect” while churn rate 2 is “correct.”  Churn rate 1 is simple and gives you what you asked for “period churn / starting period ARR.”  (You just need to realize that the your quarterly rates will only sum to your annual rate if you have zero new sales and ergo you should calculate the annual rate off the yearly churn and starting ARR.)  Churn rate 2 is somewhat more complicated.  If you live in a world of purely one-year contracts, I’d recommend churn rate 2.  But in a world of mixed one- and multi-year contracts, then lots of contracts are in starting period ARR aren’t in the renewal base for the year, so why would I exclude only some of them (i.e,. those signed in the year) as opposed to others.

[3] Dividing by the whole ARR base basically assumes that the base renews evenly across quarters.  Showing churn rates based on available-to-renew (ATR) is more accurate but becomes complicated quickly in a world of mixed, multi-year contracts of different duration (where you will need to annualize the rates on multi-year contracts and then blend the average to get a single, meaningful, annualized rate).  In this post, we’ll assume a world of exclusively one-year contracts, which sidesteps that issue.

[4] ATR is normally backloaded because enterprise sales are normally backloaded.  Here the linearity is 15%, 17.5%, 25%, 42.5% or a 32.5/67.5 split across the first vs. second half of the year (which is pretty backloaded even for enterprise software).

[5] The spreadsheet I used is available here if you want to play with it.

The Leaky Bucket, Net New ARR, and the SaaS Growth Efficiency Index

My ears always perk up when I hear someone say “net new ARR” — because I’m trying to figure out which, of typically two, ways they are using the term:

  • To mean ARR from net new customers, in which case, I don’t know why they need the word “net” in there.  I call this new business ARR (sometimes abbreviated to newbiz ARR), and we’ll discuss this more down below.
  • To mean net change in ARR during a period, meaning for example, if you sold $2,000K of new ARR and churned $400K during a given quarter, that net new ARR would be $1,600K.  This is the correct way to use this term.

Let’s do a quick review of what I call leaky bucket analysis.  Think of a SaaS company as a leaky bucket full of ARR.

  • Every quarter, sales dumps new ARR into the bucket.
  • Every quarter, customer success does its best to keep water from leaking out.

Net new ARR is the change in the water level of the bucket.  Is it a useful metric?  Yes and no.  On the yes side:

  • Sometimes it’s all you get.  For public companies that either release (or where analysts impute) ARR, it’s all you get.  You can’t see the full leaky bucket analysis.
  • It’s useful for measuring overall growth efficiency with metrics like cash burn per dollar of net new ARR or S&M expense per dollar of net new ARR.  Recall that customer acquisition cost (CAC) focuses only on sales efficiency and won’t detect the situation where it’s cheap to add new ARR only to have it immediately leak out.

If I were to define an overall SaaS growth efficiency index (GEI), I wouldn’t do it the way Zuora does (which is effectively an extra-loaded CAC), I would define it as:

Growth efficiency index = -1 * (cashflow from operations) / (net new ARR)

In English, how much cash are you burning to generate a dollar of net new ARR.  I like this because it’s very macro.  I don’t care if you’re burning cash as a result of inefficient sales, high churn, big professional services losses, or high R&D investment.  I just want to know how much cash you’re burning to make the water level move up by one dollar.

So we can see already that net new ARR is already a useful metric, if a sometimes confused term.  However, on the no side, here’s what I don’t like about it.

  • Like any compound metric, as they say at French railroad crossings, un train peut en cacher un autre (one train can hide another).  This means that while net new ARR can highlight a problem you won’t immediately know where to go fix it — is weak net new ARR driven by a sales problem (poor new ARR), a product-driven churn problem, a customer-success-driven churn problem, or all three?

Finally, let’s end this post by taking a look and then a deeper look at the SaaS leaky bucket and how I think it’s best presented.

leaky1

For example, above, you can quickly see that a massive 167% year-over-year increase in churn ARR was the cause for weak 1Q17 net new ARR.  While this format is clear and simple, one disadvantage of this simpler format is that it hides the difference between new ARR from new customers (newbiz ARR) and new ARR from existing customers (upsell ARR).  Since that can be an important distinction (as struggling sales teams often over-rely on sales to existing customers), this slightly more complex form breaks that out as well.

leaky2

In addition to breaking out new ARR into its two sub-types, this format adds three rows of percentages, the most important of which is upsell % of new ARR, which shows to what extent your new ARR is coming from existing versus new customers.  While the “correct” value will vary as a function of your market, your business model, and your evolutionary phase, I generally believe that figures below 20% indicate that you may be failing to adequately monetize your installed base and figures above 40% indicate that you are not getting enough new business and the sales force may be too huddled around existing customers.

Survivor Bias in Churn Calculations: Say It’s Not So!

I was chatting with a fellow SaaS executive the other day and the conversation turned to churn and renewal rates.  I asked how he calculated them and he said:

Well, we take every customer who was also a customer 12 months ago and then add up their ARR 12 months ago and add up their ARR today, and then divide today’s ARR by year-ago ARR to get an overall retention or expansion rate.

Well, that sounds dandy until you think for a minute about survivor bias, the often inadvertent logical error in analyzing data from only the survivors of a given experiment or situation.  Survivor bias is subtle, but here are some common examples:

  • I first encountered survivor bias in mutual funds when I realized that look-back studies of prior 5- or 10-year performance include only the funds still in existence today.  If you eliminate my bogeys I’m actually an below-par golfer.
  • My favorite example is during World War II, analysts examined the pattern of anti-aircraft fire on returning bombers and argued to strengthen them  in the places that were most often hit.  This was exactly wrong — the places where returning bombers were hit were already strong enough.  You needed to reinforce them in the places that the downed bombers were hit.

So let’s turn back to churn rates.  If you’re going to calculate an overall expansion or retention rate, which way should you approach it?

  1. Start with a list of customers today, look at their total ARR, and then go compare that to their ARR one year ago, or
  2. Start with a list of customers from one year ago and look at their ARR today.

Number 2 is the obvious answer.  You should include the ARR from customers who choose to stop being customers in calculating an overall churn or expansion rate.  Calculating it the first way can be misleading because you are looking at the ARR expansion only from customers who chose to continue being customers.

Let’s make this real via an example.

survivor bias

The ARR today is contained in the boxed area.  The survivor bias question comes down to whether you include or exclude the orange rows from year-ago ARR.  The difference can be profound.  In this simple example, the survivor-biased expansion rate is a nice 111%.  However, the non-biased rate is only 71% which will get you a quick “don’t let the door hit your ass on the way out” at most VCs.  And while the example is contrived, the difference is simply one of calculation off identical data.

Do companies use survivor-biased calculations in real life?  Let’s look at my post on the Hortonworks S-1 where I quote how they calculate their net expansion rate:

We calculate dollar-based net expansion rate as of a given date as the aggregate annualized subscription contract value as of that date from those customers that were also customers as of the date 12 months prior, divided by the aggregate annualized subscription contract value from all customers as of the date 12 months prior.

When I did my original post on this, I didn’t even catch it.  But therein lies the subtle head of survivor bias.

# # #

Disclaimers:

  • I have not tracked the Hortonworks in the meantime so I don’t know if they still report this metric, at what frequency, how they currently calculate it, etc.
  • To the extent that “everyone calculates it this way” is true, then companies might report it this way for comparability, but people should be aware of the bias.  One approach is to create a present back-looking and a past forward-looking metric and show both.
  • See my FAQ for additional disclaimers, including that I am not a financial analyst and do not make recommendations on stocks.

Churn:  Net-First or Sum-First?

****

Please note that this post has been superseded by A Fresh Look at How to Measure SaaS Churn Rates.  I’m leaving it posted to protect in-bound links only and to provide a referral to my latest material on this subject. 

****

 

While I’ve already done a comprehensive post on the subject of churn in SaaS companies and some perils in how companies analyze it, in talking with fellow SaaS metrics lovers of late, I’ve discovered a new problem that isn’t addressed by my posts.

The question?   When calculating churn, should you sum first (adding up all the shrinkage ARR) or net first (net shrinkage vs. expansion ARR and then sum that).  It seems like a simple question, but like so many subtitles in SaaS metrics, whether you net-first or sum-first, and how you report in so doing, can make a big difference in how you see the business through the numbers.

Let’s see an example.

net1

So what’s our churn rate:  a healthy -1% or a scary 15%?  The answer is both.  In my other post, I define about 5 churn rates, and when you sum first you get my “net ARR churn” rate [1], which comes in at a rather disturbing 15%.  When, however, you net first you end up a healthy -1% (“gross ARR churn”) rate because expansion ARR has more than offset shrinkage.  At my company we track both rates because each tells you a different story.

Thanks to the wonders of math, both the net-first and sum-first calculations take you to the same ending ARR number.  That’s not the problem.

The problem is that many companies report churn in a format not like my table above, but in something simpler like that looks like this below [2].

net2

As you can see, this net-first format doesn’t show expansion and shrinkage by customer.  I think this is dangerous because it can obscure real problems when shrinkage ARR is offset, or more than offset, by expansion ARR.

For example, customer 2 looks great in the second chart (“wow, $20K in negative churn!”).  In the first chart, however, you can see customer dropped 4 seats of product A and more than offset that by buying 8 seats of product B.  In fact, in the first chart, you can see that everyone is dropping product A and buying product B which is hidden in the second chart that neither breaks out shrinkage from expansion nor provides a comment as to what’s going on.  My advice is simple:  do sum-first churn and report both the “net ARR” and “gross ARR” renewal rates and you’ll get the whole picture.

Aside 1:  The Reclaimed ARR Issue
This debate prompted a second one with my Customers For Life (CFL) team who wanted to introduce a new metric called “reclaimed ARR,” the ARR that would have been lost on renewal but was saved by CFL through cross-sells, up-sells, and price increases.  Thus far, I’m not in love with the concept as it adds complexity, but I understand why they like it and you can see how I’d calculate it below.

net3

Aside 2:  Saved ARR
The first aside was prompted by the fact that CFL/renewals teams primarily play defense, not offense.  Like goalies on a hockey team, they get measured by a negative metric (i.e., the churn ARR that got away).   Even when they deliver offsetting expansion ARR, there is still some ARR that gets away, and a lot of their work (in the customer support and customer success parts of CFL) is not about offsetting-upsell, it’s about protecting the core of the renewal.  For that reason, so as to reflect that important work in our metrics, we’ve taken a lesson from baseball and the notion of a “save.”  Once the renewals come in, we add up all the ARR that came from customers who were, at any point in time since their last renewal, in our escalated accounts program and call that Saved ARR.    It’s best metric we’ve found thus far to reflect that important work.

# # #

[1] I have backed into the rather unfortunate position of using the word “net” in two different ways.  When I say “net ARR churn” I mean churn ARR net of (i.e., exclusive of) expansion ARR.  When I say net-first churn, I meant to net-out shrinkage vs. expansion first, before summing the customers to get total churn.

[2] Note that I properly inverted the sign because negative churn is good and positive churn is bad.