Snowflake Inc. (SNOW) Earnings Call Transcript & Summary

Okay. So look, we're just delighted to have Mike Scarpelli, the CFO of Snowflake, here to join us. The last time we had this in person, you were a private company. And we had the same conversation and you -- you said that you loved being private because you can say whatever you want. You remember that? I can say whatever I want, I'm private. So...

Just to be clear, everything I said came to fruition, and more.

Right. Right. And more, and more. So I think what we'll do, if it's all right, is have you just give -- I mean, I assume everyone knows, but maybe there's some [indiscernible] we need to -- little grounding. So just a 2-minute version of what Snowflake does, right? And then we'll talk about how you got here and then we'll talk about a bunch of current topics.

Sure. Good morning, everyone. So Snowflake is a cloud data platform, really started as a cloud data warehouse doing a lot of first gen cloud migrations, and then we got into a lot of the Teradata, Netezza, Oracle, but we replaced a lot of Hadoop and SQL server and other things. And we quickly got more into data sharing. And we're spending a lot more time and are excited about the opportunity for people to actually write applications directly in Snowflake, and that's kind of what Snowpark is enabling along with Streamlit, the acquisition we announced last week, one of the fastest-growing enterprise software companies. The big difference in Snowflake over a lot of other companies is really the fact that we are a consumption model, 100% consumption model. We're multi-cloud. We have 30-something deployments around the world. Think of a deployment as like AWS East or GCP Switzerland. That's a deployment for us, and we enable cross-cloud. So we have customers that will be in various clouds and replicating back and forth from one that could be all within AWS or AWS to GCP, whatever. And that's really kind of at a high level. I don't know if you want me to go into a lot more detail.

No, no, that's perfect. That's perfect. That's perfect. And you joined in August of 2019, right?

I did.

So you had worked with Frank before, obviously, as I think everyone knows, at ServiceNow. But so when you joined...

And Data Domain before.

And Data Domain before that, the software dream team. So when you joined, what were -- so August 2019, what were the two or three primary things you had to get done? And then we'll talk about what those are today.

The #1 thing was really building an organization, really, a lot of the back office functions. I do a little bit more than just the typical CFO. I actually have the General Counsel reports in to me as well as our CIO and Chief Data Officer. But there really was no real back-office IT function in the company. We had to mature the legal organization from the contracting perspective with customers to be able to scale and really had to build out a lot of the FP&A that was taxes were completely ignored, a lot of those things. And then as well, there was no formal deals, desk function in the company. I own that as well, too. And really built a small team of sales finance people that really help our sales organization sell value to customers and structure the bigger, more complex deals.

Awesome. And then today, what would you say your top 2 or 3 priorities are?

Well, I'd also say to the other thing, there was a real gross margin issue in the company. And so literally, when I say there was no procurement function, no nothing, the first thing was really -- they had just completed a deal with AWS. And I pretty much started renegotiating a deal with AWS since they came in to take dramatic costs out. And then Azure came after that, currently working on GCP but plan on going back into AWS again, too, because we're well ahead of our commitments that we've made to renegotiate again too because we have real COGS in our company. We have the compute and storage from the cloud vendors. And for us to expand margin, we either need to make our system more efficient, which we're doing. We've really spent a lot of time with engineering to take out internal cloud costs. We had a lot of tooling with a lot of the monitoring and other things that we were doing where we were relying on third-party tools that the costs were just going up astronomical. So worked with engineering on a lot of investments we could do to reduce those costs by building a lot of things ourselves, which we've done. So really, the only way we're going to continue to get margin expansion is either getting at scale and a lot of the deployments around the world where we're not there yet. It's going to be taking costs out or renegotiating with our cloud vendors, and then continuing to have better discipline on our discounting. And that's something I'm super focused on because we talked about last week, we have new -- we're constantly enhancing our software. We don't increase our price for existing customers. But as we land new customers, they shouldn't be paying what those old customers got because they're getting more value for what they're doing. And you're never going to increase -- a lot of were saying, "Well, why are you giving this price concession to your..." It's not a price concession to them. They're just able to do more for less with what we have. And I'm not going to go back and increase -- how would you feel if I came to you and said you had a subscription for -- I don't know...

I can tell you exactly how I felt because I -- now the kids all yelled at me and I had to put it back on, but I canceled Netflix because they increased their price.

Yes, pissed me off. Yes. Right. So yes, you can't do that, right?

I guess they won in the end. But for 2 weeks. All right. So how's business, what would you say?

We just had a record Q4. We booked over $1.2 billion in contract value in Q4 alone. I think we exceeded our internal...

Think about that, you booked over $1 billion in contract value in the quarter.

$1.22 billion, to be precise. $300 million more than what we were planning on doing. So business is extremely strong in the company.

And yet the stock was down. I forget where it ended up the -- it came back, but in the aftermarket, it's down like 22% or something like that. So tell -- what was going on there? What were people so bent out of shape about? And what's the right answer?

Well, first of all, this is the first time we ever guided to 2023 beyond the IPO model that we gave in 2021 to the investment bankers and the analysts, which, by the way, at the time, we said we're going to do $1.680 billion in 2023, and we just guided to $1.9 billion, the high end. And what a lot of people don't understand still in the consumption model where you're selling software, and as the software gets more efficient and when the hardware gets more efficient, there's savings because you're able to do more within that credit you're purchasing. And I don't think -- when I reminded them that there's going to be about $100 million headwind, which back in May of 2021 on our conference call, if you listen to, I actually told people, there was going to be this headwind in revenue in 2023 with the new AWS chip platform that they rolled out. Graviton2, they call it. So I think that's what they didn't like. And I don't think people focused on that. We had extremely strong free cash flow. I think it was actually 27% in the quarter. For the full year, we were -- remember, when we started the year, we actually said we were going to be cash flow breakeven. We did $140 million in free cash flow. We're guiding to 15% free cash flow next year, we'll still growing [indiscernible]

I mean that's something with these rates.

But there's not too many companies doing that.

So let's dig into this consumption issue a little more. So people -- I understand the consumption model. The comps -- the consumption model is you buy credits, right? And the credits let you do what?

So a customer buys a credit. And those credits are -- and a credit is a unit of measurement that we have and it's well defined in our contract. I'm not going to go through all of that. But you're literally paying for compute by the second of what you're using. And it depends upon the warehouse size that you -- the bigger the warehouse size, it's a bigger amount per second because you're using more servers at that time. And then you pay -- for storage, we take, I think, 6 snapshots of your storage a day and you pay based upon your monthly average of storage. You pay the -- it's pretty much the list price of what AWS or Azure or GCP have. And so if you're able to do more with your credit, you're not going to consume as much. You may say, "Well, why would you do that for your customers?" Well...

Again, just the length. They're -- people were able to do more now with the credit because?

Because they can process more queries in that same amount of time that they were burning 1 credit before, so...

You guys speeded the system up?

We -- well, we fed the efficiency of the software is one, that's the warehouse scheduling service. And not all customers are going to see the same benefit. If you just have one big warehouse that really just runs one query all the time, you're going to see no benefit out of warehouse scheduling service. But if you have a warehouse fill with lots of data and you're running thousands and thousands of small queries, you're going to see a massive performance improvement in the system. When you have a new chip, CPU, you can process queries faster, but not all CPU enhancements -- but just because you may see a 30% CPU improvement doesn't necessarily mean that the software is going to improve by 30% because you still have the -- you're still I/O bound. And you still have to take the data into the -- into that -- the new compute platform and then you need to put it back out. And so on average, we're seeing a 10% to 20% performance improvement by customers. And as we become cheaper, people put more workloads into us. And as we become faster, there's more workloads because of latency before that they wouldn't run on Snowflake. And we have many customers that have told us, "If you can get your performance to this, we'll move more of these workloads." And so we know because we've been doing this for quite some time, there will be more workloads by customers moving into us. And as an example, many times when we do a big on-prem migration, you're not shutting down the legacy system 100%. I think we have hundreds and hundreds of on-prem migrations that were in various stages. We've only -- I think there's only about 50 customers that had completely shut down their legacy system, whether that's Teradata, Netezza or others. Why? Because many customers just moved their most key workloads into Snowflake. And as we become cheaper, they're willing to move other stuff to Snowflake. And it is a multiyear journey for most companies to do an on-prem migration. Why? Because they're hard.

Yes.

There's not that many companies that have been successful in doing large Teradata migrations. As far as we know, we are the only one that has done large -- yes, there's some small ones. But I know GCP has struggled with a very large retailer for 2 years to do a Teradata migration and they're failing miserably. And why I know this, too, many times, we only get involved after one of the hyperscalers failed. We have a big one that -- Microsoft failed. They tried for a year and couldn't do it, and we were successful.

Why are they so hard? And why are you guys -- you think AWS and Microsoft and Google, you'd think they would have pretty good technical skills. So what is it that -- first of all, why is it so hard? And then why are you guys better at it?

So we, with a partner, have written some pretty good tooling to be able to translate the proprietary language that the code is stored in. Teradata is one of the more challenging ones. But on top of that, it's also getting the new cloud system to work. And technically, we never lose to Google, AWS or -- why we lose is because of relationships where they're a big Microsoft shop or they're a big Google shop. And I would say we actually partner extremely well with AWS. That relationship has completely blossomed. When I first joined the company, we did not have a great relationship. Now last year, we co-sold over 1.2 billion with the hyperscalers, and the vast, vast majority of that was with AWS. We actually have deals where AWS will throw dollars to us to help get the deal so that it doesn't go to GCP. GCP is the one we co-sold 0, and they're the most competitive with us.

What is going on there? So first, how did you get the AWS relationship to blossom?

Partnering with them and showing them that having Snowflake in AWS will lead to you selling more of your software around us, such as SageMaker. It's funny, I had a call 3 weeks ago with the people at Google, and I pointed out to them there were 300 instances where you guys compete it to the very end with BigQuery and we won. And all of those customers ended up in AWS or Azure when they all could have been in Google if you would had just partnered with us and you would have been able to sell some of your AI or ML technologies around it.

So you think that would be a [indiscernible] argument?

They're not an open platform, Google.

What's that?

Google is not an open platform. They say they are, but the reality is they're not a great -- they don't like to partner very well, especially if they feel threatened, where I would say AWS and Microsoft partner very well.

Interesting. Interesting. Okay. Let's talk about data sharing, what that means and then I'll tell you what our CIO just said in the keynote they did about Snowflake is super interesting. But you go first. What is data sharing?

Sure. What data sharing is you -- let me step back and say one of the key benefits of Snowflake is we have extremely good security and governance. We know exactly who is accessing what, and you can control which data gets accessed by who. And through data sharing, the data is never transferred. You're just giving someone and it could be someone within your company or a different division or a completely separate company, the ability to access that data without ever transferring that data. And you know exactly who is accessing that data. You can control the amount of time you want to have them accessing that data. And so that person can query that data directly in their system. And obviously, it's much easier when the data is in the same deployment you're in, but you still can technically do it if you're on a different deployment of Snowflake. There's just -- it could be egress charges if you're -- because you're still accessing that data to do things with it.

Let's see if I wrote it down so I don't misquote them. But you were saying that -- Snowflake, okay. We just started...

You're still in your migration process.

High-performance cloud native sits on top of AWS [indiscernible], right? So talk about the personalization experience. So they can get data like Experian, Dun & Bradstreet, Merkle -- whatever Merkle is. Do you know what Merkle is? I don't know. Some kind of marketing data. We...

Yes, it's marketing data, that's what you're using it for. I think you're using it on the invest -- the personal banking side, doing 360 reviews.

So it used to be, we'd send them a file, they would have their own database, they would have to do a merger, they would send it back, we would have to do ETL processes. And that whole thing would take at least a day or 2. They said Snowflake has the Merkle data in a clean zone where we can't see it, but we can join the two. We avoid ETL, we avoid extraction, we avoid the merging. And it happens immediately instead of having to wait a day.

Correct.

All right. So what we -- explain what's he talking about.

Yes, so what he is talking about in the ad tech space is probably the best thing to talk about there. So retailers -- and the retailers don't like to share their customer information with a lot of people. It's very proprietary to them. Think of it -- and I'm just using -- this is not a real -- actually, I can give you a real one, I just don't know if I'm supposed to say their names, but there's a sporting goods store. I won't give you their name, okay? They have all their loyalty programs with all of their customers. And so they know everyone's e-mail address, for instance. You have a -- well, this is public, you have like Hulu. Hulu wants -- that retailer wants to go to Hulu to do direct advertising through a clean room technology that sports retailer can put all of their customer loyalty information in there. Hulu knows all of their subscribers. They can -- through Snowflake, they can quickly see the overlap and who are the ones. And then that sports retailer can have a targeted campaign specifically to all of the Hulu customers who are in the loyalty without anyone knowing any of the information of one another. That's what a clean room technology does. And that's all through data sharing. And that is a prime use case as with all of the retailers right now and the ad tech industry.

That's super cool. Yes. And there's no cookies in there? No...

There's no cookies or anything. You get around all the GDPR and everything.

Wow. Okay. Okay. That's super cool.

Obviously, there's more of it from a technical standpoint, but that's the way it works.

Okay. And then -- so that's data sharing and then the next thing on the agenda is Snowpark and the apps. Explain to us what's the idea there.

So to write applications, you need to have a -- you need to make it easy to write things. And Snowpark makes it easy for developers to write in their programming language of choice -- excuse me, applications in Snowflake. We first rolled out doing it in Java. That's in GA right now. But the reality is, is most data scientists and others like to work in Python today, that's in private preview today, and will be in GA at the end of this year. Couple that with the acquisition of Streamlit that we announced last week that has over 40,000 developers. It's open source technology. One of the challenges with Streamlit and why they wanted to be part of us is they were going to build out their own Streamlit cloud actually in Google, but they were really struggling with the security and governance piece. And we already have thousands of customers who have used Streamlit to write applications who were pulling data from Snowflake. So our vision is we're still going to support the open-source community. But if you want to be able to have your applications in a secure, governed way, you'll do it in Snowflake directly.

And so they were going to do that in Google?

Eventually, we'll get to it. If you if you're running in GCP, you can do it in Snowflake, in GCP or AWS or Azure. But obviously, 83% or 82% of our revenue is associated with AWS. So we're going to get that built out to do it on AWS.

Yes. Okay, cool. I meant to leave more time for questions from the audience, but -- questions from the audience.

They're used to doing it on Zoom where they can do it through the chat function.

I have plenty of questions. You'll definitely get one.

Talking to data sharing business, can you speak to -- how that is -- partnership with LiveRamp in that sort of arena and what essentially equates to? What do you bring to the equation?

To be honest, I'm not that familiar with it. I know we're doing stuff with LiveRamp, but I don't know specifically what we're doing with LiveRamp. Jimmy, do you know what we're doing with LiveRamp?

Yes. [indiscernible]

I know more the Habu and VideoAmp on the data clean room stuff, which there seems to be more going on with them.

Can you talk about the main areas of investment for next year or 2 and [indiscernible]

We have to repeat the question, it's been so long.

Yes. Okay. So the question was what are your main areas of investments over the next few years? Is that correct, Bob?

On that M&A [indiscernible]

Yes. So I'll start with M&A. M&A is not a big piece of our strategy. Yes, we will continue to acquire most of our M&A to date other than Streamlit have really been around acquihire. There's been maybe a little bit of technology, but it's really more about finding people with great domain expertise, good engineering teams. And most of our bigger M&A from a people perspective have actually been in Warsaw. We've now done 2 acquisitions in Warsaw. And they're not that big but you can get very good deals, like 70 people here. We're actually looking at a couple more right now in Poland as well. We tend to do a lot of investing through Snowflake ventures in AI, ML technology companies. And why we do that? Those investments tend to drive deeper technical integrations to get the data from more into Snowflake. And we only invest in companies that drive consumption in Snowflake. And Streamlit was our biggest from a dollar standpoint, it's gotten cheaper now, that they actually wanted stock instead of cash. We wanted to give them cash. So I'm glad we did that in stock now. But that really is not a big company. It just has good technology. It's open source and what it's doing to drive to us. I think they only have 80 people in the company. So there's nothing in our road map that will require us to do big M&A. We'll -- it doesn't mean we will never, but I find it hard to believe we would do anything that big in terms of number of people. And we're not looking to acquire revenue, it would be more technology and people. In terms of where our investment focus is, it's really continuing to add our engineering talent and expand our sales and marketing organization. We added 1,500 people last year. We'll add about 1,500 net plus people this year. There is about $70 million in CapEx in our plan for next year. About $20 million of that is software that we have to capitalize because the FASB says we have to. I'd rather write it off as we go. And then there's about $50 million associated with just all the heads. We've hired so many heads during the pandemic, and we continue to. People are going back into the office. They need to build out offices, leaseholds and stuff around the world. So that's pretty much what our investments look like.

We had our first week back to the office. And the funniest thing was the people who were in who were complaining that IT couldn't keep up with getting all their workstations set up, and those were the people you knew had not come in for 2 years. Mike, thanks so much for coming. It was great to have you here. We really appreciate your support. Yes.

Thank you, any time.