On the surface, it would seem that Snowflake’s sales has been doing just dandily, but a deeper look shows some trouble brewing.
On a side-note, I have been asked off-board what I do in times like these - meaning the current massive tech sell-off. Here it is. I put every company I own under a microscope to see if there was something I may have missed and I concentrate my portfolio in only those that I have no or very little doubt about. And I try to ignore the red ink in my portfolio.
But now, on to Snowflake. I think I may have found something that I have previously missed.
The red herrings
I recently read a report that highlighted two very impressive metrics as proof of accelerating sales, with more to come. The first was that NRR is exceptional and accelerated to 173%. And the second was that the number of enterprise customers spending more than $1m per year increased by 128% which is higher than sales growth. And because this number is higher than sales, it points to further growth ahead. They then go on to say that, in addition to this remarkable achievement, enterprise customers also account for a bigger part revenue - to 53%, up from 46% a year ago.
First, NRR.
Management has told us numerous times that it takes upwards of 6 months, and with big customers closer to 9 to 12 months, to ramp up spending. So to understand what the 173% is telling us, we really need to understand how they define NRR. This is their definition:
“We first specify a measurement period consisting of the trailing two years from our current period end. Next, we define as our measurement cohort the population of customers under capacity contracts that used our platform at any point in the first month of the first year of the measurement period. […] We then calculate our net revenue retention rate as the quotient obtained by dividing our product revenue from this cohort in the second year of the measurement period by our product revenue from this cohort in the first year of the measurement period.
This is important. They define NRR as the product revenue of customers at the beginning of two years ago, for that year vs product revenue in the next year. So product revenue of n-2 vs n-1. Our other SaaS companies define NRR differently, for example Monday defines NRR as the weighted average of the ARRs from each month of two years ago’s customer cohorts vs the ARR of the same cohorts 12 months later, up to the last quarter’s end. Datadog uses the same ARR methodology but with no weighting - so is even more recent: n-1 ARR vs n ARR.
I believe this is important because we are not looking at the same metric as for our other SaaS companies. ARR is a year forward looking metric based on the current point in time whereas product revenue is a year backward looking from the current point in time. In time these metrics will be similar, of course, but Snowflake’s definition is further backwards looking vs an ARR definition which uses more recent customer counts like Monday and Datadog. I believe that if Snowflake were to define their NRR as forward-looking ARR, their NRR would already be heading south - quite a bit lower than the 173% they published most recently as I outline below.
So what exactly does the 173% mean?
Snowflake is telling us that their customers who consumed product from them in the first month of Q4 2020 - i.e. just more than 1934 customers, which included just more than 106 fortune 500 customers, generated 73% more product revenue with them in Q4 2021-Q3 2022 than in Q4 2020-Q3 2021.
Here’s the thing though. If Snowflake adds lots of customers - especially large ones like F500 customers - in the quarter or so preceding that two years ago starting point, then NRR is bound to explode, because they are ramping up very large customers from almost zero.However, if their customer acquisition starts slowing, their NRR is bound to decrease quite rapidly because of this same dynamic.
And here’s the next, bigger thing. Their year-end is 31 January. Which means the 173% describes the behaviour of customers they had on the books in the first month of Q4 FY2020 - i.e. customers on the books in November 2019, and we are comparing their product revenue for November 2019 to October 2020 to their product revenue for November 2020 to October 2021.
So the first part of their product revenue for this calc - which is variable/usage based as they point out every time (“we are not a SAAS model”) - was generated in the peak COVID months. And we know from Datadog’s results for example that customers curtailed usage-based cloud spending at all the hyperscalers in that period (Datadog’s famous Q2 dip), so it’s not far-fetched to assume the same happened with Snowflake. Of course, thereafter there was a catch-up again.
So in my opinion, Snowflake’s most recently published Q3 2022 NRR tells us about a massive COVID bump, in addition to a big accelerator-effect on NRR from very fast customer acquisitions prior to November 2019
It does not tell us what is likely to happen next.
For that, we need to listen to management, and this is what the CFO had to say on the NRR point:
Citi conference 15 Sept: “Well, clearly, as we get bigger in our installed base of customers grows, I think it will be hard to maintain at the level we’re at, but we think net revenue retention will remain best in class for a very long time.”
What do we need to look at then?
CFO, Sept 14 Piper Sandler conference:
“I do think RPO - and current RPO more importantly - is important, along with revenue. And the reason revenue is the best indicator of the future is because that’s what customers are consuming. And once you consume, you don’t really decrease your consumption. Yes, you could run certain things. But so we give the metric that we expect that 60% or sorry 56% of our total RPO will be recognized over the next 12 months that was up from what was 54%, 55%, I think it was $85 million increase in current RPO quarter-over-quarter. That is a good metric to look at in conjunction with revenue, because you got to remember, if I end up beating my revenue in the current quarter, say by throwing numbers at $10 million that took out of my current, my current RPO that I’m reporting at the end of that quarter as well, so it’s important to look at both, and how those are both growing over time.”
And again, CFO, Q3 results:
“I would say the first thing is the guidance we give. The second thing is historical revenue growth patterns coupled with the current portion of RPO to build your models.”
Ok, so they guided for 84% yoy at the top for next Q. But for both Q2 and Q3 they guided for 92% and for Q1 they guided 96%. Last year Q4’s guide was 103%. So it went 103% → 96% → 92% → 92% → 84%. This is a pretty big drop in guidance from Q3 (-8%pts) and Q4 last year (-19%pts).
In terms of the other two things he says to watch, Revenue and current RPO, here are both, and how they are trending over time.
Rev Q1 Q2 Q3 Q4
2020 43.7 60.3 73.0 87.7
2021 108.8 133.1 159.6 190.5
2022 228.9 272.2 334 350
cRPO Q1 Q2 Q3 Q4
2021 733.2
2022 773.3 841.0 992.2
From this we can see that cRPO vs revenue growth qoq growth for the last 3 quarters was as follows:
Revenue QoQ: 20% → 19% → 23%
cRPO QoQ: 5% → 9% → 18%
So Revenue growth is outpacing cRPO growth. Not great as an indicator of future growth.
And to get to a trend, let’s divide current RPO by the quarter’s run-rate revenue (most recent quarter last). It gives:
Current RPO as % of run-rate revenue: 96% → 84% → 77% → 74%
Which means that current RPO is becoming smaller and smaller relative to their most recent revenue. The revenue in the bag so to speak, is becoming less and less.
So the key things the CFO tells us to look at - guidance, revenue and current RPO - are not that inspiring.
But what about that 128% growth in >$1m customers?
On to the second of the red herrings above: >$1m customer growth (And here, for full disclosure, I would like to point out that in a previous post I got tricked by this herring myself: https://discussion.fool.com/thinking-about-snow-34914454.aspx) Contrary to how our SaaS customers define their >x spending customers, namely by using ARR, Snowflake defines these customers as customers who spent more than $1m in product revenue with them in the preceding 12 months. So let’s unpack this again. These customers went from 65 in Q3 2021 to 148 in Q3 2022. So they had 65 of these in the spending period of 12 months prior to October 2020. And they have 148 of these in the 12 months spending period prior to October 2021. So again - COVID constrained year 1 vs non-COVID constrained year 2 = big growth %.
Another way to look at this. In the same two periods, they had 171 and 223 F500 customers. So the number of >$1m spending customers as a % of the 171 F500 customers on their books 31 October 2020 was 38%. And of the 223 F500 customers on the books on 31 October 2021 that proportion was 66%. Now of course there are non-F500 customers spending >$1m too, but it is safe to assume many, if not most of the F500 should also reach that milestone. So of the large F500 customers they had previously signed up, many more than right after COVID have now reached a significant ramp-up of revenue. Meaning there is less ramp left from existing customers vs in prior periods.
So both the growth in ttm >$1m customers and the sky-high NRR are two ways of telling the same story: COVID plus rapid customer additions two years ago.
Now that we know that their most recent NRR and >$1m customer metrics most likely reflect a big COVID bump, was boosted by very rapid customer growth pre-Nov 2019, and shows old information, let’s turn to the important driver of future growth: recent customer additions.
The slowing sales machine
Let’s look at absolute number of customers added:
Total customer additions
# Q1 Q2 Q3 Q4
2020 458
2021 328 397 437 585
2022 393 458 426
YoY Q1 Q2 Q3 Q4
2021 28%
2022 **20% 15% -3%**
Looking at the pace of customer additions, this has been slowing in absolute terms over the last several quarters. They signed up fewer customers in Q3 of this year than they did a year ago (426 vs 437; the first absolute yoy decline in my data set), and only marginally more in the first 9 months of this financial year vs the previous one (1277 vs 1162). Not great
Let’s now turn to the addition of customers who have the potential to become very big spenders.
To see what is on the other side of the horizon, we need to look at the number of F500 customers added:
Fortune 500 additions
# Q1 Q2 Q3 Q4
2020 20
2021 17 15 13 21
2022 4 19 8
The table above does not need a qoq or yoy analysis to make the point that the number of F500 customers added has decelerated a lot over the last couple of quarters. They added 31 in the last 3 quarters vs 45 in the same three quarters a year ago.
But wait, that’s not all. In the last several quarters, they have been significantly ramping up the sales machine. Below is the number of S&M employees per quarter, and then two tables showing the effectiveness of this machine in landing new customers, in the form of customers added per 100 S&M employees, and F500 customers added per 500 employees.
Number of S&M Employees
# S&M Q1 Q2 Q3 Q4
2020 907 989
2021 1082 1141 1177 1257
2022 1450 1570 1672
Customer adds/100 S&M Employees
# Q1 Q2 Q3 Q4
2020 46.3
2021 30.3 34.8 37.1 46.5
2022 27.1 29.2 25.5
F500 adds/500 S&M Employees
# Q1 Q2 Q3 Q4
2020 10.1
2021 7.9 6.6 5.5 8.4
2022 1.4 6.1 2.4
In FY2021 each 100 S&M employees brought in more than 35 customers in the quarter, on average. This year that number was below 30 for all quarters and reached the lowest yet - 25 - in Q3. And on F500 it’s much worse of course. For every 500 S&M employees, they added a bit more than 2 F500 customers in Q3 of this year vs 10 in Q4 of FY2020. So they have more - many more - sales and marketing professionals bringing in fewer customers overall, and much fewer fortune 500 customers.
So? What did I do?
SNOW will have many more quarters of very strong growth, driven by significant new customer additions brought on a year or more ago, but that pace is going to decelerate pretty fast in my view. NRR is probably already at its peak. However, for me the clincher was the massive decrease in sales efficiency as measured by customers added per S&M employee. It’s still a great company and there are many other fantastic things about it that I did not analyse in this post - it’s tech, leadership, momentum, etc. etc. However in this environment I’m looking for companies with damn nigh no flaw, and Snowflake did not make that cut on the above analysis.
I therefore sold my position in Snowflake yesterday and used the proceeds to buy ZI, ZS, and S (MNDY is already my #1 position otherwise I would have added there).
Happy to hear dissenting views!
-WSM.
