Snowflake Inc. (SNOW) Earnings Call Transcript & Summary

And to my right, of course, is Mr. Scarpelli, the CFO. And we'll start at the top, which is -- well, first of all, how are you?

I'm good. Can I say something before we do this? I have to tell you.

Yes. I can't wait.

I've been dealing with -- this is actually the fifth public company I've been dealing with, analysts with callbacks, for many, many years. And I have to tell you, Pat did something the other day that was the first [indiscernible].

I can't wait. I can't wait. What was it?

He was unbelievable. He actually had his son, who is a data scientist, who works at actually a customer of ours, come on the call and ask all the questions of Christian at a technical level, which was I've never ever seen an analyst have their son do that. That tells me 2 things: a, you're old.

Really old.

I've known Pat for many years. And b, the value you get out of really if you're a data scientist and you appreciate that, and hopefully you got something out of that, that your son learned, and you could appreciate the difference between Snowflake and what others are doing out there. So I really did appreciate you doing that. Wasn't expecting that.

Well, he...

And I also realized Pat's cheap because his house was so cold that his son has a jacket on and he has a jacket on, too. So [indiscernible].

I wrote it down. I'm not going to find it in time. I wrote down Zachary's reaction. I put it as a positive data point. So yes, so I have a -- we have 4 children. But 3 of them are -- hopefully, no one's in this -- they're around here somewhere. But 3 of them are normal, but 1 of them is just really, really off the charts, right? And he's a data scientist, and he talks slow. Did you notice that?

Yes, he's very methodical, [indiscernible].

Very slow, very slow. And so he would be pausing before asking the next question. I was like, "Oh, my God. Is there another question coming?" And then it was just like great.

He asked good questions.

It was great. It was great. And he -- so when I emailed you guys, I said, "I'm going to get my data science expert on." I didn't share who it was. Yes. So Snowpark is a big deal. That was what I took away from the conversation with Zach, and we'll get to that.

So let's start with how's business and then maybe we'll go straight to Snowpark after that.

Well, we just gave an update on the business last week, and nothing's changed since then.

Yes? Give me your characterization.

Q4 from a revenue standpoint, we pretty much hit our internal guidance from the beginning, and when I say internal, what we present to our Board. And so it's where we expected it to be. What we did notice was different was the -- our older customers seem to have contributed to more of the growth, and the newer cohort are growing much slower. And we think that is a function of newer customers today is still ramping on Snowflake. There's a lot more knowledge out there around people who have been working on Snowflake for a period of time, have learned the pitfalls of not deploying Snowflake properly, the pitfalls of not putting in place the proper governance. We have a lot more partners who are trained on how to implement Snowflake properly. We are doing a lot more training with customers on how to use Snowflake out of the gate. We have built a lot of capabilities within Snowflake to ensure that people are using it properly, like auto-suspend. It was always easy for people to -- you spin up a warehouse. There's 2 things you want to do. You want to be able to select the right size warehouse, and it was easy for people to select a really large warehouse. It was easy for people to disable the auto-suspend function. So a, you could be running a warehouse bigger than what you want. Now it's much harder to do that, and we do that for you. We have all kinds of alerting when someone disables an auto-suspend function, so that people know that -- and there's a question, "Do you really want to do that?" And b, people are just using Snowflake more efficiently. But I think, too, is the earlier customers were those digitally native customers that -- think of the Instacarts and others of the world that were just born in the cloud, that really it was growth at all costs. The cohort of customers we have now are the more established, mature companies. I don't want to say they're the laggards or the late adopters. But they're not those necessarily the fastest-moving companies. They've always had cost controls in place. And so we're seeing the newer cohorts just ramping more slowly. They still all have the same end state they want to get to. They're just going to go at their pace and in a very controlled fashion on controlling costs.

Okay. So you predict the future or you were predicting the future based on cohorts. And the new -- the cohort -- the new customers are actually not matching the same behavior as the older cohorts. Is that a fair assessment?

Correct, correct. And when we were going through and spending a lot of time in the second half of January and in February as we're rolling quota -- remember, at the beginning of the year is when we have to roll quotas out by rep for the full year, and we roll out consumption quota. There's a lot of discussion that happens between my finance team and the salespeople on accounts, and that's where it became evident a lot of those newer cohort of customers are just not growing at the same pace. And I don't mean customers we landed 3 quarters ago. I'm talking to customers we've landed over the last 1 to 3 years are just growing slower.

Over the last 1 to 3 years. How much slower are we talking about?

Well, enough that I lowered the guidance to 40% growth next year, which, by the way, is still good growth at the scale we're at. I'm not going to apologize for that type of growth.

What was it before?

In the end of November, I had -- when we were doing our preliminary planning, we were thinking 47% growth.

Yes. All right. So as an order of magnitude of what the cohorts you're seeing, is that a good indicator? Or is there something else in there, a [ few ] layer, an extra level of conservativeness or something on top of it?

That's the biggest thing there.

Okay.

And by the way, we're generally pretty good at forecasting consumption on an annual basis. And I'll tell you, we had set our 2022 plan -- in February of 2022, we went to the Board. We did miss that plan by 2%, I'm giving you full transparency, about $42 million or whatever it was. And that was before you had Ukraine. That was before you had interest rates. That was before you had the crypto implosion. That was before...

You're talking about 4%?

2%.

2%, yes, yes.

So I think we do -- my point is I think we do a pretty good job at forecasting.

Yes. Okay. So you piqued my interest when you started talking about the quotas. So yes, how do you -- how do the quotas work for your salespeople?

Salespeople get 2 quotas. They get a growth quota, and they get a consumption quota. And it could be anywhere from 50-50 to 50% of their pay. Their variable pay is based on growth, 50% on revenue. It could be as high as 90% on growth, 10% on revenue if you are truly a hunter going after new logos. Or it could be 10-90, 10% on growth, 90% on consumption if you're just managing 1 of our top 10 accounts.

Do they complain that it's not fair to expect them as a, hey, just Mr. Salesperson to be able to influence the consumption at some major bank, for example?

They -- some complain, some don't.

Some other are really happy, yes?

They don't complain directly to me because they know what happens when they complain to me. But at the end of the day, though, they can influence it, right?

[ Can they really influence them ]? [indiscernible] good sales guys do...

Here's a prime example. What they -- if they're out there and they're educating customers on actually how to use the product, how to use Snowpark, we really spent our last sales kickoff on educating salespeople on how to go in and ask the right questions to customers to identify whether there's a Snowpark opportunity within that account: a Spark replacement in EMR, Cloudera, whatever. And they don't actually have to sell anything. They just need to educate the customer and get a customer using, and that will lead to a customer having to buy more capacity sooner. That's what we need our sales reps to do. And by the way, AWS, Azure and Google pay all the reps on revenue, not on bookings.

I missed the significance of that.

What we're doing is what -- as an industry, you're saying reps can't influence a customer [ consumption ]...

You're saying they all do that?

That's how they're paying all their reps.

Yes, yes, yes. That's how they all do it. Okay. So let's shift to Snowpark, which honestly I don't get nearly as well as I should. So this will be the opportunity.

Well, if you don't get it, I'm probably not -- I'm not as technical as your son, that's for sure.

Well, yes, if only he were here. But so what was the opportunity with Snowpark that you felt like you were not capturing before?

Spark workloads on -- within data engineering. And what was happening is, is before we had Snowpark, customers would take their data out of Snowflake, move it into another system, run the Spark workloads and then move that data back into Snowflake. Well, that's very expensive because you're paying to move that data. By running it right within Snowflake, you know exactly: a, you're not having to pay for moving the data; but b, you have this security and governance of Snowflake. Once you move your data outside of Snowflake, unless it's in a system that's highly secured and governed, you don't know what may happen to that data and where it could go. That's the big benefit to customers.

Yes. And what are the -- at the very high level, what are the kind of workloads that you would take out of? What's the kind of analysis that you would do in Spark that you wouldn't just be able to do in Snowflake?

As I said, I'm not the technical person...

Yes, I don't know either. I don't know either.

I'm the accountant.

I'm going to look at [indiscernible]. What do you do in Spark that you wouldn't do at Snowflake?

Analytics.

Analytics?

No, you do analytics in Snowflake all the time. That's what we do. Maybe machine learning...

The machine learning stuff, yes, yes, yes.

It's machine learning, exactly.

Okay. And is that -- were they often doing it in Databricks?

Databricks, open source Spark, Cloudera, EMR.

Yes. Okay. So basically, you had sort of the...

It also enables you to write applications directly in Snowflake by having that, and they tend to be heavily analytics applications.

Right. So -- but keep it really simple. So basically, we're talking about there's this machine learning opportunity, right, that you guys were not capturing as much of as you could. And in fact, it was worse, right, because your customers would take the data out of Snowflake, put it somewhere else and then there'd be all these inefficiencies in doing that. When did Snowpark come out?

GA for Python just came out last quarter, which really just pushed with the sales force at our sales kickoff in February.

Yes. I mean so we're right there, I guess...

We've had it for Java and Scala earlier, but Python is the most common programming language that the data engineers want to use today.

Yes. By the way, a little aside to that kid that you're talking about. Like, "Zach, what's everyone using ChatGPT for?" And he goes, "Well, Dad, I wrote a program in Scala, and I needed it to be," I think, "at Python. And so I asked ChatGPT to rewrite it for me in Python." And I'm like, "Did that work?" He goes, "Oh, yes." He submitted the code yesterday and got accepted. Crazy.

It is. That's actually what we see as one of the biggest use cases for AI is really helping develop code. I'm still trying to figure out how ChatGTP (sic) [ ChatGPT ] itself is going to make money.

Yes. Did you listen to this week's All-In podcast?

No.

No? I can't believe that I'd recommend -- honestly, it's so freaking good, right? And it's so good because that one guy is -- Dave Sacks is the head of venture firms, [ so there's probably, whatever ], 40, 50 portfolio companies, and he just kind of tells you what he's seeing, right, which is very rare to get a VC to do that, right? They generally hold things closer to the vest. But anyway, yes, this week's session is all about, they were not sure either, is the bottom line. They see where there's a lot of benefits coming to consumers, right? They're not so sure that that's not going to get absorbed by the big tech players. They're not so sure either.

It was interesting. There was a -- it was well publicized. There was a Princeton grad student over a weekend. He wrote an application in Streamlit using ChatGTP (sic) [ ChatGPT ] to tell whether a paper was machine-generated or written by humans.

Yes. We...

And that had over 8 million views.

Yes. So we had this conversation, and our 14-year-old is like, "Yes, dad, why are all the teachers so freaked out at Redwood High School about this ChatGPT thing? What is it anyways?" And so we have the data scientist and the 14-year-old at the table, right? And so Zachary goes, "[ Gigi ] ask ChatGPT" -- so she's in ninth grade. He says -- he goes, "What are you studying in school right now?" And she goes, "Global warming." And he goes, "Ask ChatGPT to write a 5 paragraph -- ChatGPT, write a 5 paragraph essay about global warming that's appropriate for a 10th grader," right? That's the beauty, right? That's appropriate for a 10th grader. And then the next example is even crazier. So my wife is an author, and she's got a bunch of articles on the web. And she was about to interview someone for a class that she's doing, who is the CEO and female founder of a video game company. And it was write the interview questions that she should ask the CEO/founder of the video game company in the voice of Samantha Parent, my wife, and it goes -- and she gets a little bossy when she's on this topic. There's a lot of musts. And so literally, the questions come out in her voice. But how is that all going to benefit Snowflake? I don't know.

So what we see happening, and I'm just going to -- I'll call them large language models, we're never going to be the one developing these large language models. Why? To develop something like that is like -- it's -- it could be $100 million in compute to develop these models. But these models need to be fine-tuned on real data. We have the data. So the big thing that we think is enabling those models to run directly on the data in Snowflake is how we think we will benefit. So companies will take those large language models. They license and run it against their Snowflake data for their business to fine-tune the models for them. That's how we believe. What's interesting to us, though, is people to be able to write their own queries and Snowflake using ChatGTP (sic) [ ChatGPT ] to develop the queries, that's -- we're already seeing people do that. There's all kinds of YouTube videos around that out there.

Give us an example.

You want to query a certain data and you want to write the code in Python, similar to what you're saying, but not convert it, actually write the queries themselves. You can do.

Yes. Wow. Wow.

By the way, they're not always 100% accurate [indiscernible].

No, no. Okay. Cool. So let's go back to what you're seeing in the macro. So lengthening sales cycles, more approvals, all that stuff that everyone else is talking about.

Sales cycles themselves, well, I've been saying since day one, these large Global 2000 accounts are 1- to 2- to 3-year sales cycles, it really depends, but they all start small. We -- generally, when we land a Global 2000, the average size is like 100,000. It's not like they're big. It's their follow-on deals typically are bigger. Don't see anything changing there. What I do see is customers wanting -- and it's more with our existing customers where they've consumed faster than their contract rate, they want additional discounts. They want an economic benefit to do a new deal. If not, they're just going to continue to buy under their existing contract. And we're just saying, "Fine. Buy under your existing contract." It's just buying capacity, and that's what we saw. Some of our largest customers just bought enough capacity to bridge them through to March rather than do an annual contract. Why? Because they can under their contract, and they can continue to do that until July. But in July, they have to do something of equal to or greater than their old deal or they lose the discount they have. And so I'm hearing more and more customers wanting different payment terms. If you went back over a year ago, interest rates were virtually 0 and holding cash wasn't a big deal. Today, you can earn 4.9% in overnight money.

Okay. Good. We have 5 minutes. So let's open it -- I have a couple more I want to ask you at the end, but let's open it up to any questions from our audience.

Who's your most, I guess, aggressively growing competitor? [ And how is your strategy addressing that ]?

There is no change in the competitive landscape. It continues to be Google BigQuery #1. It has been since the time we went public. Databricks is probably -- I don't even want to say #2 because Microsoft, I would say, would be #2. Databricks is the other one. And our whole Snowpark really goes directly at them. But the reality is they coexist in many of our accounts, why we brought them into many of those accounts early on when we partnered with them. But it's definitely Google, I would say, is the most competitive. And Google and Microsoft have the best -- they can do a lot of bundling, which they do. Databricks doesn't have that pricing pressure that a Google or a Microsoft can put on us. And by the way, we've been dealing with that for years with those guys offering stuff for free, and free isn't free. You got to look at the total cost of ownership.

How do you think about the public sector opportunity? And would you expect to [indiscernible]?

So...

We have to repeat the question.

Yes, I'll do. Question was is what do I think about the public sector opportunity and when do we expect to have FedRAMP high certification. I expect to have FedRAMP high certification very soon. We've submitted everything, and it's just waiting on the -- getting awarded that. It could be -- as far as I know, it could be a week, or it could be 2 or 3 months waiting on them. The public sector opportunity is only upside because they're pretty small. It's less than 1% of our business today. I do think public sector, in general, and I'm not just talking the U.S., is a big opportunity for us. I just came back from Korea, and I was actually meeting with a consortium of companies that are advising the Korean government on their digital strategy and what they want to do. And this is going on around the world. There are data sovereignty issues that we're working through but -- U.K. is another good one there. But overall, public sector could be 10% of our business, but it's going to take some time to get there. We're doing very well within state and local section of government that doesn't require that FedRAMP high. But it almost seems like the goalpost had moved where we thought FedRAMP moderate was going to be enough and now they're saying, no, they need FedRAMP high. So hopefully, it's very soon. And then IL5 will be next.

What's IL5?

It's another higher certification.

Is that another level? Is it really?

Yes. IL6 is -- I don't think we'll ever get there. It's probably the most rigorous.

I mean I got to learn more. Okay. I am curious though. So when you say we're waiting back to hear from them, who's them? Is it like a particular department? Or is it -- how does that go?

So what happens is, is you have to do all this stuff and demonstrate that you have all the controls and everything and procedures in place to operate in FedRAMP high. And then you literally hand all your files over to a third party to do an audit. That third party -- and you need a sponsor within the government. And we have a sponsor within the government that has to pay for that third party to do the audit of everything for them to sign off...

Is that how it works?

Yes.

They have to pay for it.

It's a lot more than that.

Yes, but they have to pay for it...

[ But in fact ], the sponsor has to pay for it. Yes.

Interesting. Okay. 1.5 minutes. What do you think investors don't get as well as they should about this story?

I get asked this question all the time.

Well, I thought it was -- I thought I had the most original question. Is it not that original?

I think in general, investors get it. I'm not going to say they don't get anything. I think a lot of people don't appreciate that how long some of these migrations are going to take with customers and that we have some customers that it's going to take them 10 years before they have all their data move to Snowflake because their on-prem data [ state ] is so big and they move so slow. I think if you look at -- and a lot of people think that we've been just an on-prem data warehouse migration. Less than 20% of our business has actually been these big on-prem data warehouses. We have a lot of new digitally native companies that were never an on-prem migration. And if you look at, we've signed up over -- well over -- I think it's over 1,500 on-prem data warehouse migrations of principally Teradata, and less than 100 of those have -- customers have actually shut Teradata down. It's a long tail.

Yes. That's a durable opportunity. All right. Mike, it's always great to have you here. We appreciate it.

Thank you.

Thanks for coming.