Snowflake Inc. (SNOW) Earnings Call Transcript & Summary

Welcome back, everybody. Once again, I'm Brad Zelnick with the software research team here at Deutsche Bank. And along with my partner, Patrick Colville, we're delighted to be hosting Mike Scarpelli, CFO of Snowflake, for this session. Mike, welcome. Good to have you.

Thank you for having me today.

Fantastic. So the format of today's presentation will be a fireside chat. [Operator Instructions] But with that, maybe we'll jump right into it. Mike, congrats once again on all the great things happening at Snowflake, your strong Q2 results and many, many more.

But maybe if I can, I'd like to ask you about some of the takeaways from the quarter and then we can talk big picture, if that's okay.

Sure.

Awesome. So maybe first, the most obvious. I think investors were surprised to see cRPO growth step down from where it was. But I think maybe in context, it was a greater surprise that it had been north of 200% for the last several quarters for a business growing the top line of about 100%. So clearly implying that there are duration effects at play. But can you just remind us how you'd characterize the quarter from a new ACV perspective, since that's really the holy grail metric. And as well, how should we think about RPO and duration from here?

Yes. Well, RPO isn't the most important thing. The most important thing is revenue, how our customers are consuming because when they consume, they have to buy more. And when they come back to buy more, that's all about timing. And when we started disclosing RPO going public and we had these massive growth, I kept reminding people that don't get too excited about that RPO growth if you go back on all the calls, because last year was the first time we really ever sold multiyear contracts. And so yes, we were growing over 200% in total RPO, but it wasn't a true compare. And what made Q2 of last year a tough comp, if you recall, that's when we did that 100 -- our first $100 million 3-year deal with Capital One, which makes it tough. But we also signed up -- we also did a number of other multiyear deals in Q2 of last year. And so what's more important as looking at the current RPO, we actually increased the amount of current RPO quarter-over-quarter by 80-something -- almost $85 million. And so for us, the first half of the year, we are ahead of our internal plan. We continued to beat our internal revenue plan as well, too. So I think the last quarter was a great quarter, and you see that in the updated guidance for the full year that we gave.

That's all pretty clear. And maybe not to nitpick, and I've asked you these questions over time, Snowflake's model is different than a lot of subscription software businesses in that you you've got a consumption-based model. But with a lot of great businesses like Snowflake, growing triple digit at scale, some will look at the magnitude of outperformance as an indicator for great companies that are just crushing it. And percentage-wise, I think that ticked down a little bit in Q2. So maybe if I could, just a few quick questions there because I think this is something people are talking about. Does your forecast accuracy improve as the business gets larger?

So as we get more and more customers with history, that history makes it easier to forecast. But at the end of the day, it's easy to forecast 1 quarter out, it gets more challenging to forecast 2 to 3 years out for a customer. So clearly, as we move through -- when we start a year, as we go through the year, I have a lot more confidence in the full year forecast that we have. Your nitpick, if you recall, when we went public, and I think I said it on almost every single call we had with investors, this is not a business where you're going to blow your revenue out by 15% to 20%. If we do that, we did a very poor job of forecasting. And a 5% to 7% beat is a big beat, And we've been consistent coming into that range. Yes, we beat by a little bit more 1 quarter. But in my mind, then we did a bad job of forecasting. And there were some specific customers that their usage spiked faster than we anticipated that caused that. But we were well within our -- what we consider a good beat last quarter.

Well, I think it's a good problem to have, that customers are realizing even more value even faster even if it throws you a little bit of a curve ball in forecasting the business. Maybe just along those lines, are there some customers or workloads or maybe other ways of looking at the business that are more predictable than others?

Yes. So existing customers are easier to forecast based upon historical behaviors. Newer customers, you have to assume they're going to grow like similar customers in similar industries, and that becomes a little bit more challenging. But we've been -- we're pretty good with our predictive models that we have, and they're all built using Snowflake. So...

Makes perfect sense. And maybe just the last one before I beat this one to death. Is there a reason we should judge your variance from expectations on a dollar basis versus a percentage basis as the business gets much larger? Because I do remember you saying multiple times during the IPO process, a 5% to 7% beat, if I beat it by much more, something went wrong. But I imagine at a certain scale, does the -- does your approach may be changing in any way?

Not anytime soon. That may change in the future but not this year. And I don't anticipate it changing next year. But we're continuing to find investors, and our view of 5% to 7% beat is a significant beat and significant outperformance.

We totally agree with you. With that, I'm going hand it over to Patrick, who's going to...

There will be one day when you can actually give guidance, and if you land it within guidance people are happy. But I know we're not in that world.

It's been a long time since I've seen that, Mike. I remember the days.

I know. I know.

Patrick, why don't you jump in, just some thoughts and questions on vertical markets.

Yes. Brad, I guess the financial services vertical was your largest contributor to last quarter to net new ACV. Can you just help us understand the strength you're seeing there? And I guess the reason I ask is because banks are typically characterized to be more conservative and kind of measured in cloud adoption. So what's driving the strength in that financial services vertical?

Yes. What I would say is financial services aren't just the typical old banks, you have a lot of new modern fintech companies in there, and you can imagine who those are. As well, you have the likes of the insurance companies and others that fall in that financial services vertical. And it was our largest contributor to product revenue in Q2. Customers within this vertical include both cloud-native organizations. We have a lot of cloud-native financial services companies in there. They may be divisions of big banks or stand-alone companies. And I don't want to give the names, but you can guess who those cloud-native financial services companies are, and they're very meaningful to us. And then as well as enterprise with very large on-prem footprints that we're still just scratching the surface on the traditional banks is what I would say there. Yes, they're meaningful customers, but their spend relative to what they could, if you look at what a Capital One spends with us versus what some of these other banks, we have a lot of runway in front of them. And it just -- and unfortunately, it's a journey to get these guys to move workloads over to us. And many of them, we haven't even touched their Teradata environments yet. We've still just taken some of the Hadoop and some of the other on-prem stuff and moved, as well as new workloads that were created within the banks that they've just started out natively on Snowflake. And so we view the financial services industry as a very, very large opportunity for us going forward, which is why you see us hosting a financial services summit this month.

Yes. And I guess, conversely, the public sector segment is somewhat lagging, relative term, I guess. Are there procedural hurdles that are causing that lag? Or is it a timing thing? Just kind of help us understand, with that vertical, what might cause it to open up? And kind of critically as well, what's your best sense on timing?

Yes. So the #1 thing is we need to have FedRAMP high certification. We do have FedRAMP moderate and we have some customers. But it seems like more and more, specifically U.S. federal customers, want FedRAMP high certification now. We are working on that. We expect to have it in 2023. It's a process. I do expect that it's most likely going to be 2024 where the federal business is a more meaningful piece of our business. Now what I would say is that's upside because it's not very big today.

Yes. Okay. Interesting. And then just -- I mean, in this kind of last segment around vertical markets, I mean, the advertising industry is one that's on recent earnings calls you discussed and you talked about kind of data sharing in the advertising industry. Who's sharing data with who? And what part of the ad tech stack is that replacing? And then I guess, kind of interesting question to ask is, how does this kind of fit with compliance with GDPR and CCPA?

Sure. Good question. So yes, on our earnings call, we highlighted NBCUniversal Audience Insights Hub is what it was called, which NBC populates with third-party audience data. NBC invites advertising partners, and you can just think of who those are, to access the data and analyze it against their own marketing and sales information. But there's -- each party is able to share data insights without moving, this is the important thing, moving, copying or exposing any underlying PII, personally identifiable information, that's the big issue with this. And this is resonating very, very well with the big media companies that we're talking to.

Great. I'm going to hand it back to Brad now to chat through the product and the technology.

Thanks, Patrick. Yes, we've got several questions we wanted to get into. Maybe just for starters on Snowpark. What is it, Mike, that makes Snowpark unique? Is it a monetizable product? Or is it just about creating a good ecosystem for developers to stay within the platform to achieve all it is that they choose to do with data?

Yes. This is what I love about Snowflake. We really sell one product, and we have different versions of that product based upon the features. But Snowpark is not a separate product that's licensed. What Snowpark does, it enables developers to write applications directly in Snowflake, accessing the data in the data cloud, and that's going to drive consumption, and that's how we get our revenue. It's through additional consumption that Snowpark is driving because of the applications that are being written. The goal is really to enable developers, data engineers and data scientists to work in Snowflake with their preferred programming language. You the first release of that was in Java and Scala, and we have Python coming out in the fall.

Cool. Maybe just kind of related to that, when we think about that audience, when we think about developers and data scientists, et cetera, Snowflake seems to be -- maybe I'm misunderstanding this, but I had thought that it was generally more of a top-down purchase for most organizations. But a lot of the tools you've been rolling out have been aimed at developers and the like. Can you discuss that strategy and how do they kind of meet in the middle for the benefit of both the customer and Snowflake?

First of all, it's definitely not a top-down purchase. It's -- the top helps with getting the deals done, but you first need to start at the bottom. And I don't mean the very bottom, but you need to start with the people who are actually going to be living in the product. And so our new strategy has been really focused on extensibility of the platform, and this means that we're going to extend the use case of what we're doing. The capabilities, make it easy to use in the government environment. And the C-suite really loves the governance. That's what they like. And they like the idea of data isn't being siloed, it's across the organization, everyone can have access. But it's not a top-down sale. Yes, you have to go both ways, but you have to have that sponsorship from below because no executive in a well-functioning company is ever going to push down a technology where the people below aren't buying into it. Super important.

Well-functioning, I guess, is the important qualifier there. But I hear you. I hear you. I've seen it.

It's like when people call me trying to sell into me, I'm like, why are you selling into me, I'm not the buyer. I can influence the deal, but first sell on my people first. And when they tell me that we need to do this now, I'll get the deal done.

That's well functioning. Well, Mike, maybe just one more, and I'm going to hand it to Patrick. You guys launched Powered by Snowflake earlier in the summer, and you've already had, I think, about 80 partners, including some big names. What gets you most excited about this? And what kind of network effects can you drive with this program?

Yes. I think the Powered by Snowflake is super exciting. And one of the big networking effects is as you have partners that are developing their business on Snowflake, they're actually introducing us into their customers that we may not have had access to or we may have lost in a sales opportunity. I know a large insurance company, because it's something that's powered by Snowflake, that insurance company had said, "Yes, yes, yes, I'm going to stick with what I'm doing with one of the hyperscalers." But because of Powered by Snowflake, now the business side is all in on Snowflake, and we're back into that account and growing. So it's a huge network effect in helping open doors that we weren't otherwise able to get into. So I'm extremely excited about Powered by Snowflake. And we talked about the BlackRock example with Aladdin. And that's a prime example. And there's others as well, too. I think we now have over 80 partners that are doing things in the Powered by Snowflake.

Cool. Over to Patrick.

I guess, I mean before we pepper Mike with more questions, [Operator Instructions]. But anyway, Mike, back to you. What aspirations does Snowflake have kind of moving up the stack? Like you mentioned earlier that Snowflake's kind of one product for everyone. But is Snowflake going to move up the stack and kind of into data-driven applications?

We don't have aspiration or ambitions, I'll say, to build applications today. Does that mean we will never? I'm never going to say never. But I think it's really important, we want to be an open platform, and we want to let customers choose what applications they want to use, what third-party products. And so we don't really want to compete with our partners. And so I don't see us doing applications today. Our goal is really to provide the infrastructure and make it easy for our customers to use Snowflake. And at the end of the day, whether it's a third-party application or something that we're building, where we make our money is really on the compute. It's the value of our software, the consumption and that's what our real goal is to do. So I don't see us building applications anytime soon. I'd rather our partners do that.

Okay. That's helpful. And I guess, I mean another thing you've been discussing on earnings call is stable edges. And you disclosed a metric around what proportion of customers have stable edges, the growth. Can you just help us understand, probably a good place to start with, just what is a stable edge?

Sure. So that's an important thing that most of our customers are doing some type of data sharing, but a stable edge for us is more meaningful because that's more looking at a persistent use of that data sharing where we look back over a certain period of time, are they consuming a minimum number of credits through that data-sharing relationship that creates that stable edge. And so in our mind, that's a more meaningful metric for us than just saying how many customers are -- we're engaged in the quarter in a type of data sharing because data sharing could be someone just looking up through our -- the data exchange, accessing data on another partner's data once a month, once a week. But a stable edge is they're constantly burning credits on a daily basis. and it's looking back over a certain period of time.

Can you share what the period of time is? Just to...

Yes, we look back over a 6-week period.

Okay. And it's typically using -- sharing data, consuming almost daily over that period?

Correct. It's a certain number of credits that they're consuming on a daily basis on average.

And I guess you said that 16% of customers now have stable edges in place. Clearly, 100% would be the optimum end goal. What do you think is -- do you think that's likely? Or do you think it's going to be kind of a more practical limit that will be below that potentially?

Yes. Once again, I think 100% of our customers, I think it's reasonable that we'll do some type of data sharing, but I don't think 100% will have stable edges because there are some use cases that customers just don't want to data share on a regular basis. And so yes, 16%, but that was up from the prior quarter. We did add 100 customers adopted stable edges in Q2, which we think is very bullish for the growth of when you look at that persistent usage. And you have to remember, a lot of our customers are still ramping on just getting their data into the cloud and starting to ramp on Snowflake. They're not even ready to do data sharing yet. That's the important piece, too, because it could take a customer -- depending on what that customer is, it could take them a year from when they sign Snowflake to when they're ready to data share. What I will tell you is the ability to data share is a compelling factor why customers choose us because they all have a vision to do it, but they know that's not going to be the first thing they do.

Yes. And I guess that's a really interesting segue into kind of my next question, which is the typical ramp for a customer. You've articulated in the past during the IPO process and kind of more recently that the time to value is typically like 6 to 12 months. I mean, is that still the case? Is that changing over time? And is there any way that you can kind of compress that time to value?

Yes. At the end of the day, the biggest hurdle with doing a -- and I'm separating it, the big on-prem migrations take time and they will always take time, and I'll explain why they will take time. These systems we're replacing are 10, 15, 20 years old. They are sucking in data, these on-prem systems, from many, many, many different sources. They have reports that they've been using for years that are formatted in a certain way that people are used to getting things. So to do those migrations, you first need to convert the code for that -- whether it's Teradata, Netezza or whatever into a format that then you can then transfer that data into Snowflake. And all the migration tools that are out there to convert the code, none of them are 100% accurate. If you're getting 90%-plus accuracy, that's great. Well, that part that's not accurate, that's a manual process. And we're talking millions and millions and millions of lines of code. Yes, we're well north of 95% accurate in doing these today. But then even once you've then transferred your data and got it into the cloud, you have to -- all those pipes have to get repointed into Snowflake from where it was sucking data before. And then there's all kinds of testing that has to happen by the customer on the output of all their reporting that they want to do. And many times, their legacy reporting was an on-prem BI tool. Now they're using a cloud BI tool and it just takes time. These are massive, massive systems. So yes, Deloitte talked about -- other financial analysts, they're focused on shortening that time and investing heavily. And we're doing everything we can, but it takes time. And I don't see that shortening much now. Customers who already have their data in the cloud, those can happen very quickly, the time to value. A new workload that's being generated within a company started out from scratch, those can happen very quickly, but they're not huge. Those new workloads tend not to be huge revenue items. It's those big on-prem migrations that drive the big revenue. And I'll give you an example. We signed a large retailer 2 years ago, did a $5 million deal with them. In that first year period, they barely consumed $1 million. This year, they're on track, that customer, to consume over $8 million. And we think that customer can [ go to ] $20 million over the next 12 to 18 months.

Wow.

And that's just the time it takes to do these migrations, and that was a massive Teradata. They just take time.

One more from me, Mike, and then I'll pass it back. Brad, forgive me, I feel like I'm hogging the microphone.

You're doing a great job.

So there's been a lot going on at Snowflake, a ton of new product launches, Snowpipe, Snowpark. Just help us understand, from your perspective, and I guess help investors on the line as well, which of those product lines that Snowflake have recently launched do you think is the most exciting, the biggest opportunity and the kind of the fastest time to value?

Well, I think, Snowpark, we're very excited about that because that enables people to really build applications in Snowflake. That was just -- we announced that, that went into private preview at the end of last year. We just released it in public preview in AWS in June, with the support for Java and Scala. All the feedback we've heard from customers has been solid. That will be GA at the end of this year. A lot of those customers to really start to ramp on it, they want to wait until it's GA, and we're super excited about that. But there's a number of things that we're doing. It's not just Snowpark. It's everything we're doing around data sharing. It's what we're doing around our support for unstructured data that will really kick in next year as well, too. And what you don't see, too, and we don't talk about that much is just the performance improvements that we continue to put in our software that really makes it cheaper for our customers to operate Snowflake. All those performance improvements that we do are 100% passed on to customers. That doesn't retain with us. Same thing like when we introduced a better storage compression ratio that we talked about at the end of Q1 when we were guiding for Q2, that had a massive impact for our customer spend on Snowflake. And what that -- when we make storage cheaper, they move more workloads over to us and they drive more compute. So we'll continue to do that.

Makes sense. Maybe I'll just jump right in. If you're okay with that, Patrick. We're all right?

Absolutely. Thank you.

Excellent. Yes, I had a couple. Just when we think about go-to-market and constraints to growth, and we think about ramped rep capacity, you're having to hire at a torrid pace. There's only so many reps out there in the market to hire that understand the industry, that understand the customer. How should we think about that and how can we -- what can you tell us just about where you are capacity-wise?

Adding salespeople isn't as difficult as what people think. I used to hear this comment at ServiceNow all the time, and we continue to find good people. And Snowflake is no different. We continue to find very good people and we're where we need to be, and we have an amazing pipeline of candidates that we'll continue to hire. I will say, as a percentage, we really are adding a lot. Asia Pacific is an area we're really focused on as a percentage of the new hires, as well as EMEA. But the U.S. is still getting the bulk of -- in absolute numbers, of our salespeople that we're hiring and we're finding those people. At the end of the day, salespeople are all about how much money am I going to make when I go there, and we have to offer a competitive compensation plan for these people, and we think we do.

And it sounds like, especially internationally, you've got enough territory to support the opportunity for them.

Absolutely. We hired a very senior person out of VMware, who was running Asia Pacific, who's based in Japan, actually a Canadian guy who's lived in Japan for 20 years, speaks Japanese fluently. He's running that region and he is doing a phenomenal job of building that team out throughout the region. So we're finding great people.

Cool. Awesome. And maybe just from an incentives perspective, how are -- I mean we understand that consumption is a big part of the comp plan. That's something I think was introduced around the time that you came on board, if I remember correctly. But how does that incentive evolve from here as it relates to consumption, deal duration or any other parameters I might not be thinking about, whether or not deals are coming in direct or indirect? How are you thinking about that?

Yes. So first of all, the vast majority of our deals are direct. Very, very, very little goes through partners or the channel. Partners tend to influence deals more. But at the end of the day, we're hosting customers' data, they want a direct relationship with us. Yes, some of our business goes through the marketplaces like AWS or Azure marketplace, and that's more customers want to do that because it allows them to burn down the commits they made with AWS and/or Azure by doing this. It's actually not very expensive for us to put them through there. I view it more as a tax. But it's single basis percentage or single percentage points that we're paying there. I don't envision any changes to our comp plan for next year. And you're always tweaking things with different [ specs ]. But at the end of the day, it's a misconception that we just pay on consumption. We still heavily incentivize our people to close deals, bookings. They get paid on both bookings and consumption, but it -- and there's another multiyear bonus if you do a multiyear deal. But there are reps where they may be on a 90-10, where 90% is paid on consumption, where someone like one of our biggest accounts that's doing $30 million in revenue a year to us already, you know growth isn't going to be that much with that customer and consumption is more meaningful. But then there's other reps that get paid 90% on growth and 10% on consumption. So we have different packages based upon the opportunities that reps have in the accounts they're managing, and it appears to be to be very competitive.

Cool. I think we're getting short on time, so we got to squeeze in a handful of important questions. One that I actually see Patrick came in on the chat, I will leave for you to prioritize that next. But maybe just real quickly, Mike, if we look back, the history of the software industry, a lot of the magic seems to happen where you get sort of the off-balance sheet resources out there really -- even if there's a direct relationship with most of your customers given the nature of what you do, but helping to be demand generators not just fulfillment mechanisms, and they themselves investing in resource and skill set and everything else. Where do you think maybe the unexpected source of leverage from a partner perspective might come from in the future for Snowflake?

Well, we're seeing -- the partners are critical to us, the big GSIs, the likes of Deloitte and others. And I guess what I would say is we're counting on that. I'm not expecting a -- I'm not expecting it to be beyond what I'm forecasting already. And if anything, that would be upside and it's hard for me to give a firm what that could be beyond what we're already forecasting. I'm probably not answering your question the way you want it answered, but that's the best I can do.

No, fair enough. Patrick, I think we -- if I have that right, I think we've got time for maybe just a couple more. Why don't you have at it, please?

Yes. I mean can we ask about competition, and an investor wants to know the answer to this. And actually, to be honest, it's around Databricks. Are they a competitor? Are they a partner? How -- I guess, how has that evolved over time? And then in terms of kind of products, what's similar and what's different?

Yes, well, what I would say is that Databricks coexisted many of our accounts today. I think they're probably in 700 or 800, don't quote me on that number, of our accounts and they tend to be more on the data engineering, data science side. We don't compete with -- when we were looking at a large on-prem migration, our competitors are the 3 hyperscalers, it's Google, Azure and AWS. And I would say Google is probably the most competitive with us. We tend to partner much better with AWS and Azure, and especially AWS and Azure really partner with us when they know Google is going to want to try and get into one of their accounts.

Yes. That makes...

Databricks does a lot of marketing around things. But what we see out there, we don't see them doing a lot of those things they talk about from a marketing perspective. What I will say is we talk about what we actually do.

Okay. Talking about marketing, can I kind of squeeze one more in? I mean GCP Omni, Azure Arc, the pitch...

Sorry, what was that?

The pitch 18 months ago when you guys IPO-ed was around being multi-cloud and independent. I mean GCP Omni had only just launched, I think, at the time, and Azure Arc has launched more recently. Has those products launching change dynamics at all in the way that you approach customers and this kind of messaging about -- around being multi-cloud and independent?

Zero, because we don't see those things really in practice and use. And once again, it's more marketing. Do you know a single customer who's using the Google omnichannel in any meaningful way?

We struggled to serve as many...

Yes. We -- that's why we figured, if anybody was going to come across competitively, it would be you. But I can't say we found anything that makes...

I've never heard of people ever mention them in a competitive situation. Once again, it's more marketing.

Fair enough. Well, with that, Mike, it's always tremendous to see you and even better to see you here today at the Deutsche Bank Tech Conference. Thank you so much for the time. On behalf of Patrick, myself and everybody here, thanks to everybody else who tuned in for this session, and we hope to see you all soon.

Well, thank you for having me. Have a good day.

You, too.

Bye now.