Exactly. All right. Good morning, everybody. Is my mic turned on? Okay, now it's turned on. Good morning, everybody. Thanks for making it to the second day of the Goldman Sachs Computing, Cloud, and Technology Conference. A real delight to be able to host, yet again, the amazing MongoDB management team, Dev Ittycheria and Michael Gordon. We are really happy to have you guys present. As you can see, the number of investors is roughly 2x of what it was last year, so definitely more interest and not less interest. The rate at which you guys are putting up numbers, maybe we'll have to upgrade you to the grand ballroom one of these days. No promises, no forward-looking statements, but if you keep up that Atlas growth or whatever it is, one day, ballroom.
So with that, the way—if I use terms like double click, segue, drill, then drinks are on me tonight at cocktails. So we'll try to not use clichés, but just use natural language search, because we're learning and we're gonna use natural language to conduct this meeting. With that, Dev and Michael, welcome back.
Thank you.
Thanks for having us.
So Dev, you've been on this journey, the shift away from relational to non-relational. You've been absolutely spot on. I keep asking you the same question, "So what does the company look like five years from now?" And, is the answer gonna be the same as you said last year? Or there are some nuances, some new developments, that really change your view or enhance your view as to what MongoDB is gonna look like and what you want it to look like in the next five years?
Yeah. So, I think last year we talked about our vision about a developer data platform, but let me just, for context, walk people through the journey we've been through so you understand where we came from to better understand where we're going. So, when I joined the company, this was nine years ago, actually this week, ironically.
Yes
We had just done about $30 million in revenue, and we're still considered an interesting toy, a cool technology, but we were still very unproven. And our job, and Michael joined me soon thereafter, our job was to really show people that we could be truly a viable technology for mission-critical use cases. We did that. Then, our job was to say that we could address a wide variety of use cases, that we could become a general purpose platform. We did that, and that's when our customer accounts started growing quite aggressively because of the variety of use cases people run. Then, the third step of the journey was to introduce a cloud service.
Now, we did this in 2016, before—this is before Snowflake was popular, before Elastic, before Confluent, and there was a lot of skepticism about could an independent company both partner and compete with the hyperscalers. And a lot of people thought we'd be roadkill for someone like AWS. We obviously proved that, and, and a year later, we, after we went public, AWS announced their MongoDB clone, and there was some panic, "Oh, my goodness, is MongoDB numbered?" And at that time, I said, "I think this is massive validation for us," and it turned out to be true. And Atlas was 2% or 3% of revenue when it went public, and now it's 63% of our business. So the next, the next stage of our journey is really executing on this developer data platform.
What that really means, simply translated, is we want to be able to enable customers to address a wide variety of use cases across a wide variety of deployment models. And so that's our Run Anywhere strategy. And so you're seeing, you know, customers, and I'm sure many of you have done your own channel checks, we are one of the most popular technologies in the software industry. You go to any corner of the world, someone using MongoDB for something. And they could be trading platforms on Wall Street, billing systems for telcos, streaming platforms for media companies, gaming, et cetera, and, you know, startups now building AI applications. So I think to answer your question succinctly, we believe that we'll be a modern platform to enable people to build the most modern and performant applications on MongoDB.
Got it. Michael, over to you. So you occupy a very important role at MongoDB. One of your responsibilities is to help scale the company from a systems standpoint, go to market, whatever it is. As you've crossed $1 billion in revenue, as you approach the next several billion, what are the key pillars of the strategy, growth strategy to help the company scale into a multiple of its current size?
Yeah, so a few different things. We're blessed with a very large market. And so if you think about the market today, per the IDC numbers, is a little over $80 billion in 2023, growing, I think, to $136 billion in 2027. And for those who are newer to this, that may seem like fairly rapid growth, right? You know, you just think of the database market, the developer data platform market as a relatively mature market, and shouldn't it grow at something closer to global GDP, not at, you know, 13%? And the reason is, as Dev sort of indicated, we're at this very critical role in enabling companies, our customers, businesses globally, to compete, right? Software is the basis for competition for companies today.
That's how you differentiate, you can differentiate with off-the-shelf package software. And so we're continuing to go after this very large market. We're, you know, now at $1.5 billion revenue scale, roughly, and we're closing in on 2% market share. So it sounds a little bit, boring and consistent, 'cause we've talked about this for a while, but the real focus is on execution. We have this very large opportunity ahead of us. We have this incredibly, you know, well-demonstrated product market fit. We have incredible, you know, developer mind share, sort of the hearts and minds in developers. But, you know, there's still, we still have to sell, we still have to go out there and tackle the opportunity. And that's sort of what we focus on day in, day out, is the execution.
Obviously, the investments in the developer data platform help that, and you can see that with the incremental workload adoption that we're getting. But there isn't some sort of big bang, you know, that we need on the horizon to happen. It's really about just going after and, and capitalizing the opportunity. Our market is different than most-
Mm-hmm
... in that-
... It's not a sort of top-down, you know, sale and you sort of, you know, displace a competitor, and now every single application within Goldman moves to MongoDB, right? It's sort of workload by workload.
Mm-hmm.
And so that's different than most in software, and obviously, I recognize, you know, many of you are pride yourselves on your pattern recognition skills, so this doesn't quite fit the pattern. It's a little different, and we kinda-
Sometimes it screws you up, pattern recognition.
And so we have to go after a workload by workload. You know, and obviously, each subsequent workload is faster, and you get the benefit eventually within account of standardization, and you can see the sort of increased sales and marketing efficiency, but it's really a focus on execution.
Yeah. This is something that I not really planned, but as I listened to your talk, Michael, who are your role models for helping scale the business? I mean, who do you look up to in the tech industry or outside and say, "You know what? That's, that's a pretty damn good recipe for...
Yeah, there's not some one company that, like, we aspire to be them.
Mm-hmm.
Obviously, there's we have obviously a lot of respect for a lot of companies out there. I mean, you can start with AWS. Obviously, they built a great franchise. They created the whole cloud industry. Then there's other companies like ServiceNow, who've grown very fast and also delivered very healthy margins. Being able to use a core technology to expand just a bunch of different verticals-
Mm-hmm.
which is similar to what we're doing now in terms of expanding to a bunch of different use cases.
Mm-hmm.
I think those are the kind of analogs we think about, but ultimately, we kind of stay more focused on our-
Somebody's waking up.
We're more focused on trying to, you know, listen to customers, and I think that journey I described earlier was really functional. Also, listening to customers and how we realized how their buying behavior was potentially changing and what opportunities that gave us.
Got it. Got it, got it. Yeah, we had the good fortune of coming out to MongoDB Live in New York. I was just blown away by product after product after product announcement. I don't know where to begin, where to start, but we could go multiple places, but let's start with the vector search. I know that you've, you have fielded a lot of questions about the direction of this market. So maybe help us understand what exactly does MongoDB want to accomplish with your vector search product? What kind of opportunities this could open up for MongoDB? And then we can talk about the streaming capabilities, which were equally mind-bending.
Right. So, I think the reason vector has gotten a lot of hype, because obviously AI is top of mind for everyone, and the one kind of one of the key tangible ways people can look at AI is through these new things called vector search or vector databases, right? What I need to remind people here is that, vectors are really another form of an index. In fact, it's called a reverse index, and every database has an index. You know, we have an index, other databases have indexes, and my belief is that over time, every database or data platform will embed some sort of vector search functionality, much like they do today with regular indexes.
The differentiation of why someone will win is the developer experience they offer in terms of how well integrated into the platform, how well it enables the overall developer workflow to build applications. And that's where if you talk to people who are using these standalone vector database solutions, it's still quite painful. They have to marry the vector database with another database, in many cases, MongoDB, to store the metadata and have pointers to the actual data. They are now trying to build functionality that we already have, functionality like scalability, things like sharding and failover, things like, you know, distributed capabilities, and so on and so forth. And in some ways, we have seen this movie before. Three, four years ago, you've been asking me about time series databases or graph databases or NewSQL databases that we're trying to modernize SQL, right?
All those things kind of, you know, were the shiny new toys, and then they hit a plateau. The reason they hit a plateau is that ultimately, for customers to really embrace new technology, one, you have to have massive developer mindshare, and to get massive developer mindshare, you really have to add compelling value to their development workflow.
Mm-hmm.
And so we believe the best thing for developers is to embed vector search functionality into their existing workflow. MongoDB, leveraging the MongoDB query language, leveraging the document model, leveraging all the performance, scalability, and resiliency that we have built into the product, and it's available day one, and it just enables them to do more things. So our preview product has been massively oversubscribed. There's tons of customer interest, and customer interest from large enterprises.
Mm-hmm.
You know, even we have a lot of startups, but large enterprise interest is because they see the benefit of, of using MongoDB because we're already in there, inside the four walls of their enterprise, and so it's much easier to leverage MongoDB because they know MongoDB and know how popular it is inside the organization than using another bespoke solution. That's the last point. Customers have realized buying the next new technology for the next new use case has diminishing returns. The cost of learning, managing, supporting those different technologies becomes quite cost prohibitive.
Mm-hmm.
And so that's, again, back to our developer data platform. We want to enable more things on one platform that enables customers to consolidate onto a fewer set of technologies.
So when you talk about the several hundreds of customers that are building applications, generative AI applications, vector search is an important component of that?
Correct.
It's already scalable, it's proven-
I wanna be cautious. We have customers using our preview technology in production use cases-
Yeah.
but they're a minority. A lot of people are playing with it.
Yeah.
But people are already using the underlying MongoDB platform as the basis to run these AI applications already.
Yeah, but-
So back-
[crosstalk] Not all of them are vector.
Yes.
Yeah.
Correct.
Okay. Got it, got it, got it. So on that particular topic, what is the difference between vector search capability and a full-blown vector database? I mean, does that-
We think, we think they're one and the same. It's just that-
Mm-hmm
The way it's encapsulated, you have to install a separate point solution or you embed it into a broader platform.
Yeah.
It still serves the same needs.
Yeah.... I recall it's about 12, 13 years back or so, we had this explosion, Cambrian explosion of non-relational databases. And, you're right, they were-- I remember writing a report, we talked about 25, 30 companies. Your MongoDB was one of them, and I think it was called a different name, 10gen, 12 years back.
Yeah, the company name originally was 10gen-
10gen, yeah.
But the product was always called MongoDB.
Yeah.
And it was a decision made prior to me joining-
Yeah
... but the company realized we should just leverage the product name-
Yeah
as the company name.
Exactly. Yeah, it was a good decision on your part.
Dev, you talked about, you know, the direct benefits, MongoDB could see from generative AI with vector search, but can you talk about maybe some of the indirect benefits? 'Cause I think that's one of the more compelling aspects of the narrative around, you know, app modernization or developer productivity yielding new workloads. Can you talk about that opportunity?
Yeah.
And exposure.
Right. So we believe we're gonna be beneficiaries on two key dimensions of this AI wave. One, all these code generation and code assist tools will make developers more productive. Now, I think it's too early to give you a quantitative number. Some people say 20%, some people say 40%. You know, there's a lot of numbers bandied about, and no one's really done a true A/B test, and the tools are still quite early. But there's no question that developers will be more productive. And some people said, "Well, if, if it's 30% more productive, then do you need 30% less developers?" Actually, I disagree with that vehemently, because every development team I know, including ours, has a big backlog of things they would like to do, but they just don't have the development capacity to do them.
Now, suddenly, if you increase development capacity by 20%, 30%, 40%, you're gonna be able to do more things. So by definition, you're gonna produce more software, produce more applications, which means needs more databases. So we're a net beneficiary of that. The second thing is modern applications need modern platforms for all the performance and scalability requirements. So I think we're, you know, have an inherent advantage over some of the legacy platforms. The other thing that's happening is what generative AI is doing is basically moving AI from the world of data scientists, where it's still somewhat theoretical. People build these machine learning models, but you have to figure out how to deploy the model, how to actually use it.
Some of these models never get used, to being moved to the developer, where now that's where the, you know, every business is, is encapsulated in software. Now you're making things real. Now you can make things like I can, you know, be more intelligent around my supply chain. I can be more intelligent around, like, how I'm driving more efficiency of my support staff. I can be more intelligent around, like, you know, what new products to build. And so I think the world of moving to, to developers will make AI more real for companies, and then you'll see this ex- you know, explosion of new things that they can do.
And you talked about the importance of app modernization, and you recently released Relational Migrator for general availability.
Yeah.
So, can you talk about the importance of that product and maybe the reception with this new generative AI paradigm shift?
Yes. So as Michael mentioned, we have less than 2% share, and the bulk of the, you know, workloads and the wallet share in our space is essentially SQL apps. And we have lots of customers. When we went public, we mentioned that 30% of our new business was SQL migrations to MongoDB. By the way, that's also a lot of people never thought that we could even win any bit of that business. The biggest challenge we have in getting people to migrate an app is the switching costs, right? Because there's three components to migrating an app. One is the schema. How do I map a SQL schema to a MongoDB schema? Second is the data. How do I move the data from, you know, a relational database to MongoDB? Those two things, frankly, we've already automated with Relational Migrator.
The third thing is, how do I rewrite either all or parts of the code to run on MongoDB? That is the most manually intensive part of the job, and that's what gives a lot of customers pause. And so we believe that as code assist and code generation tools become better, there's an opportunity to leverage AI to reduce that switching cost, where all of a sudden I can start, you know, auto-generating code that will allow my app to now run on MongoDB versus being stuck on relational. And so that is... we view as a big opportunity for us.
So Dev, on that point, the code assist, what products are you guys using?
So we-
As a developer.
We announced a partnership with Google where they're already using Codey to train-
Yep
... Codey on MongoDB. So we have, obviously, a huge corpus of data-
Mm-hmm
... best practices, et cetera, that are publicly available that we're giving. I don't wanna say too much, but you can expect that there'll be some other announcements with, some other, you know, partners as well on this front. Because we want all the code generation tools to be trained on how to, you know, program and code in MongoDB. Because MongoDB is very popular, they're also very motivated to do that.
Yep. I'd love your view. I mean, you've, you've been in the tech industry for a long time, and remember working with you on the-
That's your way of saying I'm old.
What's that?
That's your way of saying I'm old.
No, so am I. I mean, I remember the IPO that I worked with you on. I think you founded a company called BladeLogic-
Correct
... back in 2007, 2008. So you've been through distributed computing, cloud, et cetera, so it's very rare to see. So do you have a view on LLMs, whether it's Microsoft's or Google's? How does this kind of shake out from your perspective, and what does it mean for MongoDB as to which partner do you choose? Does it even matter which LLM you need to bet on? What are your views?
Yeah. My belief is that the world will go into a fragmented set. Fragmented sounds pejorative, but a wide set of foundational models for a variety of reasons. One, no one's gonna want to feel comfortable using only proprietary models because they don't want to be reliant on any one vendor. So you're gonna see this advent of a lot of open source models, and you're seeing a lot of companies, you know, essentially produce more and more open source models. In fact, some of our customers, we have one customer who's working with a partner to basically train an open source model for the pharmaceutical industry, because this pharmaceutical customer wants to use these foundational models to be able to start thinking of how to, you know, generate new molecules and ultimately new drugs to address a certain set of diseases.
Mm.
So you're gonna see, you know, I think, very use case specific foundational models. I think that's the way the world is gonna go. Obviously, it may gravitate to a small set that become most popular, but I think that's the way the world's gonna go.
Is there something about the document model that makes vector search a natural adjacency that you could credibly take on?
Well, I think that what's proven with the Document model, that it's incredibly flexible. So, you know, when you talk to developers who use MongoDB, they can add features very, very quickly.
Mm-hmm.
They don't have to go through something what's called the ORM process. ORM stands for Object-Relational Mapping, which is mapping code in your programming language-
Mm
to data sitting in tables. If you just think a simple example, if I'm thinking of, like, customer data, that one object could be spanning across multiple tables, tables on a SQL database: name, location, history, et cetera. Where in MongoDB, it's just managed as another object in an entry in a document.
Mm-hmm.
And so it becomes a very powerful way to work with data, which is why MongoDB is so popular, right?
Mm-hmm.
That's why developers have flocked to MongoDB, because we make their life so much easier.
Yeah.
I think with vector search, the variability of the types of data and the unstructured nature of the data. Again, vectors are indexes, but the data associated with vectors ultimately is very unstructured.
Yeah.
That's well suited for documents.
Got it. I have one more for you, and then we'll go to Michael. We wanna talk about this thing called consumption, which will be a heavy topic, and then cash flows. So you got enough meat on the bone-
All good
... coming your way. So Dev, on streaming, so when you announced the streaming capability, it just made me wonder, is this the beginning of a turning point where the core technology platform can do so many things? I mean, vector search is one, maybe there's streaming, and who knows, other capabilities that can be added on as, as adjacencies. In the streaming market, where do you see the white space? What is not being done well? Is this even a big market in the first place? What got you interested, and what is the, what is the thing that you're going for with this?
Yeah. So it's important for people to understand there's really three types of data categories. There's the operational data, the OLTP data, there's the OLAP data, or what's called the warehouse data, and then there's the data movement, right? And so the modern version of data movement is these Kafka queues, right? And so, Confluent has built a business, you know, on top of building a proprietary business on top of the Kafka technology. That, I think over time, will ultimately get commoditized. But what's really interesting is not so much the data movement, because there's plenty of plumbing to do that. What's really interesting is, how can I process data while it's moving?
Mm-hmm.
So I can get insights faster, so then consequently, I can make decisions more quickly-
Mm
or take actions more quickly.
Yeah.
Stream processing is the new interesting thing. So it's not the streaming plumbing, but the processing of the streaming data.
Mm-hmm.
The Confluent folks made an acquisition of a company called Immerok-
Yeah
... which is in based on a technology called Apache Flink. There's another type of technology called ksqlDB. We looked at both technologies, and we were somewhat underwhelmed because they are very rigid schemas. They make a developer's life very difficult. And we felt we were well set up for three reasons: one, it's very developer-centric; two, the data's all in JSON; and three, the variety and variability of the data moving across is well set up for the document model. So that's why we felt... And we talked to a bunch of customers, and there was a lot of receptivity for us to doing this, which is why we announced this capability, and obviously, we've been working on it for a while. And the reaction and the demand to get access to our private preview was, you know, enormous.
So we're really excited about the customer interest that we're seeing. And the meta trend that you may be asking, so what's the big deal? The meta trend here is event-driven, real-time applications is kind of become more and more of the norm. So being able to get insights and take actions on those, you know, events, and automate that into your application is critical for as you, as you run your business.
Yeah. So your realization was that Kafka and Flink were not exactly flexible, not developer-friendly?
Flink, yes.
Flink, okay.
Yeah.
Yeah.
Our stream processing does run on Kafka, but it's a different type of processing engine than Flink.
Did you develop that organically?
Yeah.
Okay. So that's the secret sauce?
Yeah, because we're using our own ingredients, right?
Yeah.
We're using the document model-
Yeah
and all the understanding and knowledge we have with JSON, working with JSON data
Yeah
and creating that streaming engine.
So we've been talking about real-time databases for a number of years, right? Including Vivek's company, TIBCO, from the late 1990s. And I asked this question of Jay Kreps as well. So what's different this time? So can this really be a big market? 'Cause he had been, Vivek had been talking about it for 20 years back, you know, like, "We're gonna take over the relational market. Real-time is where it's at." What has changed that has gotten you interested in this market?
Yeah, I would say, to me, it's less about, you know, I described those three categories. I am not convinced the middle category, the data movement category, is a category unto itself. I believe that it'll ultimately be subsumed into a larger platform, and I believe it'll be subsumed into an operational platform because it, it's all about processing and working with data-
Mm-hmm
for real time to building applications. Now, there'll always be pipelines to, or, you know, warehouses to move data, and so that data- people can run reports-
Yeah
and do other things, you know, downstream. But from a real-time point of view, I think it'll be my belief long term, will be subsumed into a larger operational platform.
Got it.
I would just add-
Yeah
... you know, the result, this is all the result of being pulled by our customers, right? And sort of following developers and following what customers need. To the earlier comments, we're in this huge market, we have very low market share. It's not because we need to sort of access some additional TAM or anything else like that, it's really just responding to where our customers are taking us.
So Michael, on consumption, so you guys were the first ones to call out slowing down of consumption in April 2022. As you exited the quarter... what is your observation of consumption trends? Are we at a point where we've bottomed and things are stabilizing, maybe even you see a bit of a bounce? Because clearly, as Jan Hatzius, our Chief Economist, has been saying for quite some time, we're not. The probability of a recession is lower than what we thought before. We probably are headed for a soft landing, and I'm gonna call it software landing, so nobody else can use that on the sales side. It's a Goldman Sachs thing. So, if that's the case, then shouldn't consumption growth start to pick up?
Yes. So let me describe what we've seen. And we've tried to call out all the trends that we've seen, including, you know, back in, at the end of Q1, of last year. And so what we've seen is, starting with Q2 of last year, slower growth of existing relationships or workloads. The new business environment has remained robust for us in terms of winning new workloads. But as I mentioned, we don't just, you know, win an account once and then all the other workloads flow. It's sort of a workload-by-workload dynamic. So, the winning of new workloads, we've been able to successfully navigate the macroeconomic environment. The sales teams have done a good job.
We continue to be sort of, you know, mission critical and kind of, you know, top of the must-have versus nice-to-have list. And so, we've been able to navigate that, you know, very effectively. We're really pleased with the results there. In terms of the existing workloads, though, what we've seen is we've seen, starting with Q2 of last year, slower growth of those existing workloads. And really what that translates to is the underlying query activity, right? The underlying reads and writes. We have this very tight value linkage between our customer usage, the value that they get out, and their, and their, their bill. And what we've seen is just slower growth in the underlying database activity, and therefore, slower growth in those existing workloads.
In our investor day in June, we, you know, helped try and provide some incremental visibility in that, and sort of showed a chart that went back over several quarters. And what that showed is sort of in the several quarters prior to macro, you know, there's certainly some seasonal, you know, fluctuation, but, you know, you could kind of draw a line, as we did, sort of the average of what the week-over-week growth rates look like.
Mm-hmm.
And it was at a, you know, meaningfully more elevated rate than that same kind of, you know, starting Q2 and beyond average. We've seen that generally stabilized within a range. There's certainly seasonal puts and takes-
Mm-hmm
... to that. Q2—the other thing you can see in that chart is sort of Q2 is, like, slightly seasonally lower on a week-over-week basis, relative to Q1. Momentary parenthetical, not to confuse people who are focused on the Q1 being seasonally low because of raw numbers of days.
Yeah.
That's an absolute comment, but what we tried to do in this chart is sort of normalize for all that by looking simply at the week-over-week.
Mm-hmm.
On a week-over-week basis, you can see that Q2 from that chart that we presented in June is slightly weaker, and so we saw that play out.
Mm.
I'd say we've been at a stable range. I don't know that I can call it a bottom, but certainly we haven't seen, you know, deterioration. At the same time, we haven't seen acceleration.
Yeah.
I think those are the high-level comments.
Matt wants to ask you a cash flow question.
Mm-hmm.
Well, first, let's get to Dev, and then the cash flow question-
All right. Okay, perfect.
... because I think it ties in nicely. But you talked about new workloads being the unit of competition for MongoDB, right? And you recently announced another iteration to your go-to-market for Atlas. Can you talk about kind of the importance of that change, how that, how you see that accelerating new workload acquisition? And then we can tie that in with Michael and kind of the disincentivizing of, you know, upfront commitments and how that will impact cash flow.
Yeah, the key thing is, again, it's important, we may sound like a stuck record, our unit of competition is the workload, not the customer, because in every organization, any large organization, you'll have MongoDB workloads sitting next to Oracle workloads and maybe even SQL Server workloads. So we have to win app by app or workload by workload. So we've reoriented the whole company on the product side, offering more and more capabilities, so we're more attractive for more workloads. And on the sales side and go-to-market side, basically orienting our go-to-market organization to just acquire new workloads. Now, as you can imagine, in a traditional software kind of philosophy, you pay a salesperson to go get someone to commit to you to doing some business, and that was our approach.
The challenge, though, when you do workload by workload, is it's very hard for a customer to predict what exactly the usage or the consumption of that workload will be over any period of time. So we were naturally forcing customers to commitments because salespeople operate by the way they're compensated, and customers say: "Well, if you give me a better deal, you know, in terms of discount, I may commit to a bigger number." And there was this awkward kind of tension in the discussion. So over the last three years, we've been slowly reducing the emphasis on commitments, so it's been a journey. And this past year, we stopped paying our salespeople on even one-year commitments, because our retention rates are very high. So I said: "Why, you know, why force the customer even to a one-year commitment and just get them to deploy the workload?
And as they see it grow, if they want incremental discounts, they'll come to us and saying, "Hey, my app is growing quickly. I want a better price." And then we'll say, "Let's talk about a longer term commitment." So that reduction or that removal of that friction, based on the change in comp plans, has really helped us accelerate the acquisition of new workloads. But again, I want to say, it's just—it's like a slope of line, it's not some inflection point. And, we're really pleased by the velocity of workloads that we're acquiring. Most of them are, you know, start very small because they're new workloads, so it takes time for them to—they don't really have much impact in the quarter, in the current quarter. But over the, you know, the years, we should see that impact really show up in the pipeline.
I imagine the incentive mechanism for sales reps now is that they participate in the underlying application growth?
There are some incentives in terms of the quality of the workloads as well as the number of workloads, so we try and balance, you know, P and Q.
Michael, on the cash flow transition and what that may look like, and also maybe just some context around kind of Atlas pay-as-you-go versus upfront commitments today.
Yeah, so a few different things. This is, as Dev mentioned, been sort of a multi-year journey. I think it shows up particularly in the Q2 numbers, if you look at sort of the delta between the non-GAAP op income and the operating cash flow. Part of that is just Q2 is seasonally a low quarter from a collection standpoint. You can see that by the ending Q1 AR balance. But certainly, you know, this dynamic that we're talking about, you know, also factors in. We shared on the call, if you think about Atlas year-over-year growth rate was 38%. Dollars committed upfront was actually down 15%, right?
So that sort of helps sort of put some numbers or order of magnitude around the dynamic, you know, that we're describing here. We also had shared, I think it was in Q1 of last year, to also help people understand this dynamic, that around 80% of Atlas doesn't flow through deferreds, right?
Right.
So if you think about, there are people who, yes, will still enter into, you know, an upfront paid commitment, usually, as Dev said, now driven going forward by them pushing for that, as opposed to us, you know, the sales rep having an incentive to do that. Side footnote that also still is incrementally helpful from a discounting standpoint, right? When that balance in the negotiation shifts to the customer asking for the commitment rather than the salesperson having for the commitment, you don't suddenly gravitate towards the, the max and the discounting matrix, and so you see sort of improved discounting discipline there. You also have, obviously plenty of customers who are, you know, pay-as-you-go, sort of, you know, sign a, you know, contract, will be billed monthly, invoicing in arrears. Obviously, that's what the self-serve basis is.
So again, this has been part of sort of a multi-year journey, but it particularly shows up in Q2. The last thing I'd say is, you know, we're not all the way through the transition, but we're certainly not at the beginning of this. So, to the extent that you see a meaningful divergence, you know, in the amount committed upfront, that will still be a factor, and I would expect that would play into next year. But at some point, we'll sort of normalize or stabilize, and then we'll kind of return to traditional relationships. I think the last thing that's worth mentioning, in case it isn't obvious, I'll just make it explicit: we still collect the same amount of cash, you know, over the course of the year.
It's really just a timing factor, but want to call that out.
A quick lightning round in the 55 seconds that I have. I have three things. You can feel free to be brief, yes, no, something. Early fee for calendar 2024 budgets? Number two, would you consider accelerating your hiring if the environment stabilizes? And, three, how do you price vector search? I mean, do you get... Is there a separate SKU or,
I'm sorry, I didn't catch the first question.
The calendar year view for calendar 2024 budgets.
Customer, customer budgets?
Customer budgets. Yeah, yeah, yeah.
My sense will be somewhat the same as this year. I don't see... I mean, obviously, AI is getting a lot of hype. I personally believe there's gonna be a little bit of value to spare. I always believe people overestimate the impact of a new technology in the short term, but underestimate long term. And so I think you're seeing a little bit of the, you know, a little bit of the hype cycle going on right now. But I think we're seeing large enterprises, but they're being very thoughtful and methodical about how they think about AI, so I think that's gonna be more of a long-term phenomenon.
Mm-hmm.
Overall, I think budget onloads will be the same. Your second one?
Very, very quickly on the lightning round.
Yeah.
On hiring, yes, our hiring, as, as reflected in the guidance, is back-end weighted. We're obviously digesting a bunch of investments from fiscal 2023, so have plenty of hiring in the back half of this year. And then vector search will show up in Atlas.
To be announced?
No, will show up in Atlas consumption.
Okay, got it.
As opposed to a separate SKU.
Perfect. Thank you so much, gentlemen. Amazing presentation. Thank you for showing up, and,