Klaviyo, Inc. (KVYO) Earnings Call Transcript & Summary

All right. Fantastic. We will go ahead and kick it off with Klaviyo.

Good.

AB, thank you for joining us. We're really excited to have you.

Yes.

So I know that you've always had a really specific vision of what you think Klaviyo can look like longer term. And it feels like this year, we've really started to unlock the next levels of that vision beyond all the success you've already had on the marketing side. So paint off the picture a little bit. How do you think Klaviyo is evolving? Or in what ways do you think Klaviyo is evolving? And how does that fit into where you think the company can be 2, 5, 10 years from now?

Yes. So, when we started, the genesis of Klaviyo was I'm a big runner and I've built this website to help people find 5Ks and marathons and stuff like that. And I worked with all the little race organizers. I remember doing that thing manually like go to each one, hey, you should listen to my website, you should listen to what I say. And I was like, boy, this is really annoying. I wish I could clone myself and just like be it everywhere to all of these folks at once. And that is what the driving ethos of Klaviyo is. I really do think. We ought to be able to scale people like either individual humans or teams, companies and just make them infinitely accessible at scale all the time. So we talked about empowering the world's businesses, the world's creators to own their destiny and the way to do that is to help them be themselves at Internet scale. So be authentically you one-on-one. In the early days of Klaviyo, our approach for that was we said, okay, well, if we're going to mimic the way that humans act and operate. We built like a really good brain for that. And so that became like the database that we built. And so stores indexes data in real time, makes available in real time and stores indexes all these interesting ways, the ways that like we actually think. And then we layered on top of that a bunch of like tools that actually allow you to communicate. So it was like marketing and messaging. But the key thing that always kind of irritated me about that was you still needed somebody to log in and sort of set up all the rules and like modify them all the time. So somebody had to like build a marketing campaign or build a marketing automation and then come in and check on it a week later and make some edits and review the data. I was like, that's not really -- I mean it's the one version of scaling ourselves. I mean you can deliver this experience to millions of people concurrently. That's kind of the lesser version, like wouldn't it be awesome if you could kind of just self-optimize. I think -- obviously, messaging is just one facet of like the customer experience, and that's great. It's very proactive or like we're sort of missing the other pieces of it. The last year, we've said we want to attack -- we want to improve on 2 dimensions. The first is, I think this era of like SaaS software, really, we think of it like it's tools, you got to log in, you got to -- you need to set it up, configure it, be the person that's using it. I think it's just over. It will take a little bit of time for people to adapt. But -- we've already talked to a lot of our customers who are like, yes, I'm bandwidth constrained. Look, if you can do a good job, like, yes, I'll check it before it goes out, but like I can delegate that work to you. That's why what we're doing on marketing is I think the future of that is really an autonomy layer on top. It's like, yes, we'll develop a marketing strategy for you. We'll figure out all the campaigns and automations you should be running. We'll draft them all out. We'll make them awesome. We'll set them up and all you have to do is go in and tick the box, tell us that you approve. And by the way, when you get tired of clicking all those boxes because we're already doing -- always doing a great job, we'll just turn on autopilot mode and off you go. Like that is going to happen. I think, to the marketing side of the customer experience. So we're solving for that there. And then I think this autonomy layer, it's coming to like all different facets of the way that, say, a business interacts with its consumers. So messaging is one. It's the proactive side. Hey, everybody is not always -- they haven't always open up your app. They're not always on your website. You're not always top of mind. So how do you know when to reach out to folks so they don't have to come pull and look for you. The flip side is like when somebody has a question, and it's not always like -- I think it's not just customer service. It's not just break fix. But when somebody needs some advice and they want to have a real-time conversation like a chat, and this could be over text, it could be over -- it could be audio, actually, in the future, I think it will be video as well. Any of these modalities, like what do you offer them? That's the customer agent products that we launched to our private beta program in June, and we'll soon be releasing more broadly. And then finally, we've done some work on what does it mean to personalize the experience, maybe when you're not chatting with somebody, but you're like in somebody's store, so to speak, right? You're on their website. And that's our Customer Hub product. They basically say like, yes, tell us -- remind us who you are, like we probably know who you are, but just not send a case, we can show you all your important information. And then when you're on somebody's website, like, yes, we're going to rewrite the whole thing for you so that it's tailored to you. And this could be a website, it could be a mobile app in the future, any other digital service. I think that's kind of it, right? I think this is how we take the ethos of what a business or a brand is and make it ever present. And I think that should be -- it should be self-configuring. It should be learning on the fly and improving. And we have metrics, whether it's engagement or revenue or LTV that we can optimize for and obviously, somebody can set, and this should be possible. And I think this should exist certainly for every consumer business, where internally, we have been saying is like, look, if you're Boeing and you're selling airplanes, then yes, you probably shouldn't use Klaviyo. But if you're an airline and you have millions of passengers, then yes, you should definitely be using Klaviyo, right? So I think for consumer businesses, this is like -- this is obvious. So we need to bring it to all of those customers, large and small around the world. By the way, one of the things that's been really interesting to me is enterprises love this idea because one of the things we've learned that was maybe a little surprising, but maybe not so much, a couple of Klaviyo scale might see some these patterns even internally and they're kind of fight against them. They love this idea of AI playing more of a role in the generation and delivery because they care so much about compliance and approval chains. When I talked to -- such a major global retailer like a sports travel brand. And they were like, yes, this would be awesome if you guys could generate this stuff because we could plug in our brand guidelines and guardrails and make sure that everything that our hundreds of marketing teams are doing is consistent. So yes, how fast can you deliver this? This is like a major unlock for us. It will speed up that, right? We'll make sure we're maintaining compliance. So I think all of that's going to exist. And I think this era of software, just purely as Software as a Service, just as you got to log in, you got to configure it, you got to do it all yourself. I think that era is over. And we're definitely building to be the CRM first for consumer businesses, and then we'll work on everything else that is really AI-driven, both on the generation and then the end customer experience delivery.

So let's start on this idea of the era of SaaS. I think since 2022, arguably, we started to see consolidation in the front office stack, where there's a lot of buying in 2021. So my question for you is, why do you think now is the time where you see an incremental shift in the landscape? And from a competition standpoint, this is a question that we've been debating all conference for the last several months now. There's incremental competition from the AI natives, potentially incremental competition from the frontier models. So a lot in there. Why now? And what about the extra competition?

Yes, yes. I think you got to kind of parse it apart. So like if you just -- if you go purely to this like, hey, we're going to use the software to deliver these experiences, there's many services, there's many different formats, different channels, et cetera. Look, this consolidation story there is rooted in a pretty basic thing, which is everybody wants the experience to be driven off the same brain, so to speak, same data set. And it's really frustrating when these different pieces of software, either one don't connect back, each have their disparate sort of data systems, which -- or they're just lacking of brain, they all have to connect back to something and then the latency gets really bad. So I think this idea that like all of the different various marketing channels are going to converge. And then -- well, by the way, these basic things of like if I send somebody a message, send them a text message or an e-mail, and then they open up their phone or they open up the website and it's inconsistent. It's just a terrible experience. And it shouldn't be that way. And it's like, well, why is that? It's usually because there's not one back end. So we obviously solved that. I think that's why you see a lot of consolidation there, plus the normal things of like, look, people want to learn 5 different systems. They want -- one, they want it to be well designed, all this kind of stuff. The AI native stuff has been very interesting. Like I do think there's -- I spend most of my time -- I'm most excited about the companies that are AI natives that are new. The same way that like when the web started out, it was very fun watching the companies that were like trying to figure out the web, but like it was hard and it was messy. I mean I remember going to a lot of like the developer conferences back in the early 2000s, where there's only 30 people in a room messing with the latest and greatest technology about how to build an interactive website and, wow, this is great. It's not really there yet, but you can tell it's going to be. This has to be entirely the same feel. But it's fun hanging out with the AI native companies because they're thinking and dreaming that way. The challenge for a lot of the incumbents is that some of them are like, yes, we'll see if it happens, right? I don't know, give another 5 years, we'll see if the go. That's like that's born. I'd be like going and talking to people who are building software for CDs, right? So what's interesting, though, is I've actually been kind of frankly, unimpressed like in the marketing space. The customer service is a lot more interesting. I mean a lot more interested there -- but I've been unimpressed by like people going for this, building better tools to do content creation. Like look, we've got a project internally as part of defining your marketing strategy. I think every modern brand should also be -- in part, should be a media company. Like, we're working with this outdoor retailer, and they sell all sorts of camping gear, hiking gear, climbing gear, all kind of stuff. And like why aren't you also the go-to destination to go find great climbs or great camping sites. Why don't you both? And they're like, "Oh, we never really thought about it. We just don't have time to do it." I'm like, but you can. All that information is there now. You can provide more value to your customers than you're doing before. Like well, that's a great idea, which we did not execute. Like well, now you can do that. So that's the kind of stuff that I was hoping, you'd see more of, but I just frankly haven't seen that much of it. So anyway, we're going to get deep in there. Hopefully -- I mean, hopefully, there will be more companies that are kind of focused on that in the next couple of years. The customer agent, the service side, I think it is more interesting. But that's where actually -- we're just very focused on this idea that I think you're thinking too narrow if it's purely about customer service. I think this is -- every business is going to provide some AI that is going to be like the concierge, the representative to that business, and it's going to solve all sorts of problems. And we have these like fascinating stories when we turn our agent on for our customers. And you look at -- we have this whole classifier that will look at the kinds of conversations people are having. And it's like it's crazy. I mean there's basic things of, hey, I need to return this product or I have a question about sizing or whatever. We see people like I've seen conversations with some like high-end some fashion brands. I've seen people like plan entire like, well, we're having a wedding, and I'm wondering what the bridesmaid should wear, but also I have questions about what my mother-in-law, like if it's holy stuff, this whole conversation going on, and they're just querying away at this like product catalyst to try to understand what's there. I mean these are kinds of conversations that I think are going to become the norm. You already see this happening with like the larger LLM. So I think it's fascinating who's going to do that well. And obviously, we're benchmarking like crazy, our models, our end agent, how good is it at retrieving the right information and actually answering questions. And what we found is, at least in a retail and e-commerce context, because of the access to data we have, both about the business, product inventory, catalog as well as that consumer, their data set, we can just provide better answers, right? So I think AI is the most interesting companies. We spend a lot of time there. And then yes, I know something -- often times people ask, hey, do you see this collapsing of the user interface where does everything devolve to like ChatGPT or Gemini? To give you a little bit of analogy, like I always felt like even in retail and commerce, it felt a little simplistic to me that the entire world was going to collapse down to like Walmart and Amazon. It just felt like that's possible, but like you're not really going to have -- there's no room for different experiences. I mean, in theory, it's possible. But in practice, it just doesn't seem likely. So I do think for a lot of productivity-based things, and I think we probably find personally, I certainly do. I think a lot of the LLMs just for a wide variety of like personal productivity tasks. And so things that are more adjacent to that, yes, it makes sense. Those will get pulled in. I think as you get more domain-specific and you're getting deeper into it, I have a hard time -- I mean it's possible, maybe we'll have to see how it rolls out. But if you're thinking about like, say, we give you a marketing strategy and like -- and here's a campaign we think you should run and then we want you to review it. It's hard for me to believe that entire interface and then the editor and all kind of stuff will fit inside of that box. So I think we're still a little bit like the explosion of web apps. I think we're going to find there's some patterns that will be discovered of like good ways to interact with LLMs. And I think we're going to find there's also some specific to each application or at least some of these high utility, high-value domain-specific applications where the UX needs to be a little different and we want to be really great about, too.

Yes, fantastic. All right. I want to talk a little bit about CRM, meaning Klaviyo's expansion from marketing to sales. And historically, we've seen a little bit of a nuance here between B2C and B2B. So talk to us a little bit about where you feel you're strong today from a sales feature and functionality standpoint and where you need to go to be best-in-class.

Yes. And so well -- so we built Klaviyo to focus on B2C use cases first, frankly, in large part because there will be less friction, right? I mean it was one thing, if I had to convince a salesperson, it's like, hey, so we actually believe the entire customer experience should be delivered through software. It's kind of a tough pitch, right, given the job they're in. Now I think what's happened -- what we found is marketing, I think it often is just automated sales, just sales at scale. It's a pretty good analogy, right? So we found it's like, yes, Klaviyo is obvious in a B2C context. But then what happens is we're finding now is through AI, you get these like really high-value conversations, Hey, I have this one consumer, but they spend a lot. They're a very frequent customer. They're a VIP. And I really -- I almost want to provide them a more tailored experience. I think what we're finding Klaviyo interoperates well with CRM. So we can escalate to an account manager or a person or a human for now. This is actually -- one of the things I think is interesting about the customer agent we're building. So we started out with text, so chat modality and then audio. I think in short order, probably over the next year, we'll have to see how some of the foundational models develop. I think we'll be able to offer video as a modality as well. And then you're sort of saying like what's the difference here? I think a lot of businesses, we're going to find it's like, what's their value add. It's sort of like the data set that they have that an LLM can think and train off of. And then it's like what is the brand and personality that you want to provide? How do you want to interoperate with the world. Once we get down that path of providing, I think, the kind of modalities that people expect that are more sales modalities because we're already doing that with marketing. I mean you could have a sales conversation with our customer. But if you think about more traditional sales where it's like, yes, I'm looking at somebody in the face. I think we're going to get to the point where, okay, we now can provide that kind of experience. And then it's going to boil down to the depth. How good is your agents at understanding context behind the scenes. And that's actually something we're using our customer agent on Klaviyo's own data set. So it's fairly complex, right? We have this fairly complex -- people ask very specific questions about, hey, I built this market automation, it's not functioning this way. By the way, I have another question about this other feature over here that just came out, like should I use that or not, it's applicable to my business. And we need our agents to be able to handle that. So we're already getting into some of these more complex queries that we tend to think like, only humans can answer those. But I think in pretty short order, we're going to find that maybe we can't -- maybe it can't be an augmentation or a replacement for all of these, but I think we're going to find it, gets pretty good pretty quickly. And that's something we're very bullish on. So I think there's a natural progression for us from doing marketing at scale, service and then that starts to bleed into things that might look like more traditional sales use cases.

Yes, fantastic. Yes. And then we spoke a little bit about this at the beginning, but we've talked a lot about Klaviyo's infrastructure over the years, and you talk about the speed and also having the completeness as well. How do you do that? And how is that kind of different than how you see others in the space try and solve that problem?

Yes. So this concept of if you want to replicate a person, you'd better build some data structures that are as good and as fast as this guy, I think it's very important to us. Our approach has been this is we look at it as like, look, the way you store data, there's not one sort of database data format indexing scheme that works. Instead, what our data platform does is when you push data into it, structured data, unstructured data, we actually index it in a variety of different ways. And we have a couple of routers that will look at the data and into it, how it's likely to get used downstream. So just as an example, like when you -- when a consumer takes an action, they buy something. They have a conversation with our customer agent. They browse a website. They go visit an event that you ran. We take that data and we store it in a bunch of different ways so that you can access facts about that exact information. Hey, what does this person buy in this transaction? You can look that up like in milliseconds. We also then go aggregate that with other things that they did. Hey, help me understand the attribution of what kinds of things this person responds to. For instance, maybe they prefer getting messaging over e-mail versus text messaging or maybe, hey, it was a result of that customer agent query. So we index some of the attribution data. And then we aggregate it up with other data we have about that end consumer. Hey, how many transactions have they had recently? And can I query that also like in very low latency? And then obviously, larger across the entire business. Hey, how is this customer similar or different to other customers in my business? So because we do this kind of multiplexing, what it means is then when it comes time to like query time, when it's time to think, right, and take an action, our database is very fast. And so that's our goal, is that behind the scenes, by the way, you could replicate what we're doing if you had a large, like, say, data infrastructure team. And I -- we pattern matched our data system, our data platform, the product we provide with a lot of the big at-scale consumer companies and their data infrastructure team they provide to themselves where they're their own customer. And oftentimes, we swap notes and they're like, wow, this is really great, like we could use a lot of the stuff that you're doing. I think over time, if we build our data platform, right, people should opt to use this versus roll their own, right, set of data warehouses and databases together. And it's a little bit outside of scope. I know we're mostly talking about software applications because infrastructure is so core to what we do. I often -- the other thing I really like looking at is who are the data warehousing, data lake, data tech companies? And how much are they thinking about this like multi-index sort of like multi-data backend future. There's some -- a couple of interesting projects there, but I think we're, if not the only company, one of the first companies to really try to commercialize that. And we haven't commercialized it directly. We do a little bit through our data platform, click the data platform SKU. We mostly just use it to power our applications. But I do think over time, that's another really interesting one is can we just expose this brain to our customers? I used to think that was going to be more important. I actually think now it's more likely that, that's going to be a competitive advantage for our AI because it just has a fast link to look up a lot of information that otherwise you'd have to go invest a whole bunch of infrastructure to basically to get the same results to get the same latency.

And then as you embed more AI in your platform and you think about the potential for maybe commoditization in some of the AI functionality and your ability to charge for AI eventually, how do you see that playing out over time?

Yes. I'll give you an example, some of the things we're doing with marketing and embedding AI right now. We've taken kind of a dual approach. It's probably more biased towards bake things into our products to just to grow market share. I think there's such a moment here where you build these wow experiences that just so level up what people should come to expect. I mean we've got these interfaces inside Klaviyo now that like that we'll be sharing with all of our customers very soon. It's just dramatically different. It's very different when you log -- you can tell what SaaS software you log into, and there's no to-do list. It's just like choose your own adventure, you just click around everywhere because [indiscernible]. AI software is going to tell you, no, no, I need you to review these things because if you weren't here, I would just take these actions. So I think -- as we move towards like that -- yes, that new AI -- sorry, that new AI future, that's going to become default. We want to get that to everybody. We don't actually want to put that behind a paywall. Now what we are finding is there's some functionality that may be more applicable to our larger customers or we believe that some of the things that we're going to build, we actually have -- there's room and customers will understand price leverage over time. For example, our customer agent, we're deliberately pricing at a price point that we want to encourage people to get it out in front of consumers. We don't want you to hide it on a page where it feels very hard to get to. But we also know there's different types of conversations. There's some that are cheaper, and there are some that like are actually higher -- they're cheaper compute, cheaper to serve, but also lower value. There's also some that are higher value. So I expect over time, we'll see some pricing discrimination even on the cost of those conversations.

And then on the consumer side of things, I think how consumers interact with technology could look different, very different potentially 5 to 10 years from now. So how do you think about investing in your product stack to prepare for that when there's probably -- it feels like a lot of unknowns right now.

Yes. It's all about like this is a very fun time for our product engineering team, myself or the whole company because things go very fast. So I'll just give you an example from like, I think it was a week or 2 ago. So the models are improving so quickly and it's so changing internal functionality, we'll talk about the consumer side. There were some new models that made it image editing much easier. Was it an Achilles heel of a lot of AI models? Yes, yes, you'd say, hey, take this person and put them in a white shirt and it's like what totally changed for the person? Well, that problem is like they solved it. They almost solved it. It's like really, really good, right? You can imagine it has lots of applications for us where we're like, hey, we have these like staged product inventory where it's like, hey, here's -- we did this like photo shoot. Well, so we've actually -- we codenamed a product internally in photo shoot. We're like, well, why can you take that product and put on anything, right? Put on a model of your choosing, right, or put in any kind of scene. And literally, we took that model, and I think it shipped on a Tuesday. By Friday, we put it into our product. We watched the traffic over the weekends. And already, we had thousands of customers, right, on the weekend working through these things using us. So velocity, I think, is the key thing. And then I think as it relates to both -- I mean, consumers, whether it's our customers or end consumers. I think you have to iterate a lot on the types of experiences, like what that could look like. For instance, I'm very bullish on -- we have this new product we're shipping called Customer Hub. You go visit a website, not only we'll rewrite the web page, but it gives you kind of this pane of glass when you go visit a retailer that's not named Amazon, where you can see your entire experience with that customer. We're iterating on the little modules you put in there like crazy. One of the things we added the other week was we put in the ability that if you're a customer and you come to a website and you like just remind us who you are, we'll show you all of the discounts and promotions that you're eligible for that you might have missed because you missed that e-mail or text message. We're finding already people are going in there just to hunt for deals because they think they might have missed it. And brands love this because the whole reason they ran that promotion in the first place was to drive engagement. Now we've just given them another surface. But that's the thing that we just -- we've had to play with and see what the data says.

Your comment on the data warehousing companies is really interesting. Maybe just contextualize for us. We know that Salesforce is Data Cloud. We know that Snowflake talks about marketing as an end-use application. We know there's a CDP layer in there somewhere. Where do you think the value accrues? And you're talking a little bit about there's some displacement opportunity there, it sounds like with your multi-infrastructure end...

For us, I think we think so much of the value is actually one of the reasons we got into marketing after building this kind of data platform first is I've always felt that like if you choose to be infrastructure, then you go wide, right, with probably lower margins, right? And that's a great business model, a lot of companies have done that. If you do applications, obviously, the more -- those tend to be higher margin, higher value because they're closer to the end outcome. And then it's just a question of how widely applicable are they. In general, we've looked at our brain, our data platform as a means to an end to power our applications. Our belief is that we should build the core applications that we think are going to be ubiquitous, everybody needs, but also we want to open it up to third-party developers. So we've actually done a lot in the last 12 months to invest in our developer program. I think we now have -- I think there's like hundreds of applications in our version of our App Store, our app directory. And we're investing a lot in getting people to build more there. Why? Because it's like, well, look, if we can't build it, we would love for somebody else to do it and at least do it on top of this shared data platform. And we know businesses want this because, look, I get requests all the time, part of the reason we got into our customer agent and customer service people say, why can't you unify this part of it? So we know we can't get to all of it, and we would -- actually would like people to build more applications. And this is also helping us as we branch out beyond retail and commerce. There's a lot of bespoke applications that we're finding in other industries or if I could just go do this with Klaviyo, this gives those folks access to do that.

Yes. Yes, makes sense. Okay. I want to ask you about the R&D headline from a couple of weeks ago. Tell us a little bit about what your framework was for doing the reduction in force and then how you're reinvesting those resources?

Sure. Yes. I mean -- I think it's -- we -- I have a lot of respect for everybody that signs up to put on, as we say, puts on the Klaviyo jersey. At the same time, we very much believe that we are -- drive very hard. I think we talked about careers at Klaviyo is a little bit like joining a professional sports franchise, like people don't spend 30 years in the NFL or the NBA or whatever. And so I think it's fine if it's like not everybody is, but we respect the work that folks do. This for us is like kind of normal course of business. I will say the one thing that's maybe a little different or something to make sure that we're thinking a lot about right now that's not entirely related, but there's some overlap is this idea of like I mentioned the AI native company is the most interesting. This idea of thinking AI first and LLM first is so important to us as we think about talent. I think in the SaaS era, you could get away with a little bit of like, yes, I've seen this before, let me take my playbook. I'm going to rerun it here. It will all be good. I think that's just totally broken. So we talk a lot about what are the advantages of scale and being a company where it's like, yes, you've got 100,000, 200,000 customers, partners, right? We've obviously built a business in size, cash flows. We have a ton of amazingly talented people, but there's some biases that come with that, too. And one of the big ones is, and I think about this every single day is what got us here probably is not right for the next era. And so one of the things that we're checking for both folks that work at Klaviyo now and folks as they come in is, not just are you into AI. We use analogy of like if you were in the late 1990s and say you're at Amazon or Google and you're like, hey, you want to go work there. I mean if you showed up for [ interview ], cool tell me about how you entered the Internet? But it is like, well, I've Amazon before. You in like tens of millions of other people, oh, I tried Google, right? I'm like that doesn't mean anything, right? What you want are the people that have these like side projects, like I built my own website. okay, now we're talking because that was hard and people figured it out. I think the exact -- you can imagine it's like all these things, if you were at Google and you're like, hey, I've run a data center, but I run -- I don't think Internet scale. It's like you're just totally the wrong fit for what we're doing. It might have been made a ton of sense in the '80s or the early '90s. It's just -- it doesn't make any sense anymore. We think about the same thing for AI. It's like if you're not the kind of person that's going and coding against LLMs or my gosh, now with all the tools, you don't have to -- coding has taken on a whole new world. It's not semicolons and braces anymore. Now it's just about can you type in text and iterating on that. If you're not the kind of person that's just proactively going and doing that, I don't think you're the right fit. It's a little tough and feel got to go figure that out themselves. But thankfully, I think we have thousands of people that are very leaned into this. So if you're proactively in AI, that's great. So there's a little bit of that, that we're also to support for.

Yes. Fascinating. So the other thing that we wanted to spend a little bit of time on is we've talked about the different product modules. But at the same time, you're also going from being primarily mid-market focused to going further up market. And there are plenty of companies in the SaaS world that have really struggled with that. So tell us a little bit more about what you think the limiting factor to the type of customer or the size of customer that you can [indiscernible]. And how do you continue that kind of one-size-fits most approach while also layering some of the best-in-breed functionality that you need to go up market?

Yes. So I -- so the first is like it's my ambition, it's our ambition that I think we should be working with businesses of all sizes. I have a soft place in my heart for every entrepreneur. I know how hard that is to build from the bottom up. And so we really want to help amplify those businesses. I know there's a lot of other great companies out there that want to do the same. I actually started my career working with the Fortune 50. And it's very fun when you're working with brands like McDonald's and Starbucks and [ CVS ]. These are great companies, right, huge scale. So we saw how those patterns look like, and I'm eager to bring those kind of like working with those businesses and some of like how to help them be successful at their scale with their complexity to Klaviyo. So we're very much invested in that. I remember a couple of years ago, I got this advice from somebody and I said like, well, how do you think we should go about it? Because like I have a little bit of experience here, but it's a couple of years old. And I got 2 pieces of advice from somebody I really respect. The said, one, they go find some deals that are very, very curious but have seen this before. They'll help you with some of the patterns and help you pattern match on this. And the second is like, don't wait. Just go get your butt kicked, right? So like just dig in there, take the mentality under you have to dig in with customers and just get in with there and help them solve some problems. I think in the enterprise, that's one of the things I learned back from my doing that before we started Klaviyo is you have to go in there and really understand what problems matter most, where are you going to start, how are you going to get in there and then you grow and expand from there. And so that's very much our mentality. And so as we think about Klaviyo, we're finding there's opportunities with our marketing stack where we're starting out in some regions, some business units and then working our way across the organization. We've gotten a lot better at that. I think with our customer agent, we've now actually given ourselves another angle. We're getting introductions now into the operating teams that are thinking through, hey, what does this look like to have an agent present that represents us. So anyway, there's a lot more to do there, but I think just like go idea, get your bucket. We've taken our bruises on that, and we continue to learn. But thankfully, I think if you go talk to our customers, we review every single one of them every month and make sure that they love us because everybody talks. And then on the building product side, look, I think about good engineering, I think, is like you build systems that can scale. You give people concepts that are abstract, they can work with. So we thought about Klaviyo, this idea of like what is the customer? We thought about that deeply. What does it mean to like represent them in all of their actions? What does it mean to do marketing at scale when you have thousands or millions of marketing campaigns? We've designed our system from the ground up to think that way. And then honestly, we actually almost like with our smaller customers, our entrepreneurs and SMBs, we almost try to reduce all that complexity down. We hide it behind a ton of UX. So it's actually -- it's all there, but they just can't see it. So it's great engineering, I think, and flexibility gives you what you need in the enterprise, give people the building blocks, the LEGO set and then we use the analogy of like, yes, then for the entrepreneurs and SMBs, either give them a book on how to build the pirate ship or whatever. And now with AI, I think we actually just build them to damn thing because a lot of them are like I don't even have time to build this thing, like just do it for me.

Yes. Yes. So I remember at the time of the IPO, you had this beautiful chart that had essentially a back-end infrastructure. And it was a bunch of different flavors of database technology. And the whole point on one of the points was you can swap in new technologies as they come down the curve to keep your back end best-in-class, dynamic, nimble, et cetera. Maybe share with us, is there one technical change you've made or one technical upgrade that you've made in the last 6 months that levels you up?

There certainly are new data stores. I mean people talked a lot about like vector database and things like that as it relates to AI. So I think that's very important for interest. Actually, one of the most interesting things with enterprise that we've learned is a lot of folks -- they're somewhat standardized at this point, like the kinds of different format. I think what's unique about Klaviyo is the fact that we can scale such a breadth of them, like ways to store data, and we're smart about how to sort things. What's actually more interesting is I've got a lot of enterprises, they're happy to put their data with you, but what they really want is a place where they can build models outside of Klaviyo because they look at the models themselves. And I'm not talking about LLMs here. I see like models where they're clustering their customers, they look at that as a big competitive advantage. And so one of the things that we try to upgrade is the ability that people can train directly on data sets in Klaviyo and do that quickly without having to import and export data all the time. I think that's going to be very key for us is those data science teams, they don't actually really -- the data they want to clean and normalize, but they really want the ability to go build and train and put their own intelligence on top.

Fantastic. I think this is a good place to leave it. Please join me in thanking AB for his time. AB, thank you.