Twilio Inc. ($TWLO)

All right. We will go ahead and get started. Thanks, everyone, for joining us. Khozema, thank you for coming.

Thanks for having me.

Khozema is the CEO of Twilio. Before we get started, a few disclosures. My name is Arjun Bhatia. I am the analyst here at William Blair covering Twilio. For a full list of disclosures, you can go to williamblair.com.

Okay. I'm excited for this conversation. Let's dive in. Obviously, as you know, I think the topic of the day is AI. There's been a lot of sort of debate about does AI disrupt incumbents, does it change the competitive landscape? And I want to sort of maybe start by phrasing that conversation in the context of Twilio. What are the sort of the moats that you have in the business that maybe don't make you at risk of AI disruption. We'll get into how it benefits your platform. But talk just about the moats on the platform that are hard to replicate even though AI has made it easier to code and build things. What's -- where does Twilio sort of defend itself in that world?

Yes. I mean our moat is really difficult to overcome because in many ways, it's physical, right? So sitting underneath our business is what we call the Super Network. And what that basically amounts to is that we've established 4,800 interconnections across 180 countries around the world. And those are negotiated contracts. They are physical connections between infrastructure, they are a set of compliance hurdles that we've had to climb over with every single one of those interconnections before you even get to the point where you're onboarding customers. And so you just -- you can't AI your way there. You can't prompt engineer your way there. Like that is a truly kind of physical, very deeply negotiated constructed moat.

That's with the carriers? Sorry.

That's with the carriers. And then on top of that, what sort of reinforces that moat going forward is the more and more that we layer data into the overall infrastructure, that allows us to create another layer of moat where we've established relationships now with our customers that are also very hard to overcome.

Okay. What -- in this world, so you have these 4,800 interconnects with carriers all around the world, so your customers can reach their customers wherever they are and communicate with them. If you're a carrier, what sort of stops a carrier from bolting an API onto their platform and say, Hey, we're Twilio now basically, in other words?

They could, but it wouldn't really help, right? Because the reality is, is that the reason that we have 4,800 across that 180 countries is that no customer wants to reach just a certain constituent of consumers through the one telco, right? So let's just take the U.S. In the U.S., you have like the big 3 that we all know, but there's like a number of other smaller players that have consumer exposure as well. So let's say that you signed up a contract with AT&T. That might help you for the AT&T subscriber base. It wouldn't help you with the 2 other carriers, which you would also need to be able to reach their consumers. It wouldn't help you with the other kind of smaller players that are in the U.S. environment, doesn't help you in Canada, it doesn't help you in Mexico, and it doesn't even get you close to any of the other countries around the world where those interconnections are required as well. And so establishing one, like that's easy. establishing 4,800, that's very, very challenging. And again, still has to be done one by one to be able to ultimately develop that what we call the Super Network.

Yes. And it's not code that you're saying is the barrier to build. It's literally having these relationships and...

Yes. I mean there's a ton of code on top of that, as you know. But yes, there's -- sitting underneath that is like physical infrastructure.

Okay. Let's -- and then I want to maybe -- we'll come back to AI for sure, but I want to just talk about the business and sort of maybe set the lay of the land for folks here. I think when you took over as CEO, Twilio was growing maybe 7% organic, high single digits, call it, There was a trough, and you've sort of accelerated growth well beyond that point now. So what has sort of changed at Twilio over the years that you've gone from this mid-high single-digit growth to now you're doing mid-teens, and there's obviously a lot of sort of momentum behind the business still.

Yes. I mean so many different things have changed. I would say, if I were to kind of say it in a nutshell, like we're just running the business a lot better, to be honest. Like -- and I think it really encompasses 3 different dimensions. Like in terms of the financial discipline of the company, like we're very smart about our OpEx envelope. We've taken a number of actions to keep our head count about the same to reduce our stock-based compensation, to get ourselves to a point of significant cash generation. So I'd start there, like doing that allows us to kind of operate from a position of strength. The second thing is that on the growth side, we've gotten a lot more clinical about where we're going to focus our growth efforts. And so maybe said simply, like it's been more of a self-help story than it's been one of like market tailwind. On the self-serve side, which has kind of been the roots of the company, like product-led growth, we've done a ton of work behind the scenes to ensure that for every successive customer that onboards, the process is made easier and easier for them. And so all of that friction that I described earlier in terms of compliance hurdles and onboarding, like we've tried to implement a series of protocols that a customer can attach themselves to an API. We templatize a lot of the things that they have to go through. And then once onboarded, they can connect to pretty awesome technology. And then that runs the gamut, though, from not just self-serve customers all the way over to ISV customers, which is sort of a different side of the continuum. These tend to be mega centers. We've done a lot of work for them as well because these guys have a number of established sub accounts. We've also made it easier to onboard those sub accounts, which are basically businesses they serve by white labeling us. And so we get pulled through, but onboarding all of those guys is complex as well. We made that process simpler. And then finally, on the innovation side, we've actually moved faster by focusing on less, right? So let's say, a couple of years ago, we would probably have focused on like 100 things all at once, all with equal intensity. And now like we pick like what are the 7-ish high conviction projects, which we think a couple of those will yield really outsized benefits down the road. And I'd say it's the totality of those things that's led to some of our recent success.

Okay. Very interesting. And I think if I sort of drill down into one of the areas where you have been seeing a lot of success, specifically, it's voice, and there's all the rage about voice AI and how AI is going to change the future of voice, and we're kind of in this sort of renaissance and your growth sort of reflects that, right? I think the voice growth is 20% last quarter, so outpacing the business overall and far above where it used to be. So what is happening with voice broadly? Because there are these tailwinds, and we sort of hear it everywhere of customers even talking about we want to change the way we communicate with customers through voice. What are you seeing in the business?

Yes. So there's a lot there. So I mean, I think, first of all, like just a couple of years ago, like we were actually calling for what we believed would be a renaissance of voice, like we were using that language. I'm not sure people believed us at the time. But that's a language that we were using because we did anticipate like as AI took off, the place that it would take off first is in voice. Like voice is more natural, like that's the way that we engage with each other. It's the way that you and I are engaging right now. It would be much easier -- it's much easier to have this conversation over voice than it is over text message because you get the emotive capabilities. You know when something is going wrong, you know when someone is angry or upset. And so voice lends itself to workloads that work very well in that respect. The honest truth is that while voice is a catalyst for what we're seeing -- voice AI is a catalyst for what we're seeing in terms of some of the growth that we're seeing in the overall voice channel. The reality is that it's still -- voice AI is still a pretty small percentage of what we're seeing in voice overall. And voice, while it's the second biggest channel at Twilio, it is still a relatively small channel relative to the entirety of our business. So why am I saying all that? The point is that the AI contribution in terms of voice is still pretty muted. It's not 0, but it's still pretty muted. And so what we're excited about is that as we look out and as AI really starts to take off, like a lot of that volume should be on the comp.

And what is the -- what do your customers sort of need to do to implement voice AI or scale up usage for you to see this revenue come in the next 2, 3, 4 years?

I think we're just like in the early stages of the buying cycle, right? Like I mean, I'd kind of maybe turn the question back on you. Like the number of times that you've interacted with a voice AI agent in the last 1 year, I'm going to guess, is like probably less than 5 times, right? -- it's probably the same for every single person in this room. Now the reality is the technology is there. It does work. And when it uses a context layer underneath it, like a segment, for example, like it actually does solve customer problems very rapidly. It reduces costs. It generates more revenue. But I think the adoption cycle by enterprises, in particular, is a little bit slower than maybe sometimes AIs are written about. I think the AI natives are going there very quickly. The enterprises tend to pay the bills, right? And I think you've got 2 things happening on the enterprise side. This is maybe crude, but like on the regulated side, okay? So think financial services, insurance, health care, I think there's a lot of experimentation, but very, very slow adoption, okay? Where we are seeing heavy experimentation and more adoption tends to be on the unregulated side. So where the stakes tend to be lower. So think food service, e-commerce, retail. And I do think you'll start to see it take off a little bit faster there. My own personal perspective is that in 3 years, I would be stunned if the first point of contact for any one of us in a service event isn't first a voice AI agent, not an IVR, but a voice AI agent that's actually able to converse with us in a very natural way.

And we won't know necessarily.

Well, I think we'll always disclose it, and customers will, too. But I think we will not be able to tell the difference versus a human. And the technology is already there.

And what -- like in this world, so this future of where there are voice AI agents, what is the benefit for your customers on the unregulated side, if it's a retail store, like what is their ROI of going down this path? Why are they doing voice AI?

Yes. I mean, so everybody immediately kind of goes to cost. And I would say that's like actually the third benefit that you get out of it. So to be sure, definitely spending a handful of cents on an interaction versus paying a human agent for an interaction, like there is a huge difference in cost there, but that's really the third benefit. I'll maybe use an example to make the point, okay? So like in the food service industry, right? Like think about like a big day like an event, okay, like World Cup is coming. So during the World Cup, my guess is there'll be a lot of pizza ordered, okay? So during that kind of an event, like what happens, right? Like believe it or not, about 35% to 40% of all orders do not happen on web forms. They take place over the phone. okay? So what actually happens, though, is 20 calls get placed, 19 of them immediately hit an IVR menu, you get through a few -- the menu is not there to guide you. It's actually there to stall you, okay? It's there to stall you until they can find time to attach a human agent, all right? Now of that 19, probably a handful of folks drop off because they're like, I'm not going to wait more than a couple of minutes, and the business loses those customers, okay? Now fast forward to the world in which it's all AI iterated. So now all 20 of those can be handled simultaneously because the AI can infinitely scale. But that's -- so there's like a revenue event there for the customer that becomes pretty interesting. But the more interesting part of it actually is for our customers what they're able to do is like in the context of that 1 out of 20 that gets through, what has to happen because the person in the store, she's got to get through the other 19. She rifles through those calls as fast as she possibly can because she's got to get to the next one. That leads to kind of a crappy consumer experience. And there's no ability to upsell the consumer, okay? Now just imagine so that what has to happen in 60 seconds or less, even if you just double it to like 120 seconds or extend it to 3 minutes, the consumer gets a better experience. The AI agent on the other side, especially if they're using a context layer, they can revenue upsell the customer -- the consumer. So they can actually now make more money, right? So the revenue benefit in terms of upsell, that's benefit number one. Benefit number two is the additional revenue benefit by being able to infinitely scale. Now that's a little bit more complicated because it basically transfers the bottleneck from the IVR to the supply chain, if you will. And then the third benefit is the cost that we talked about. And so the ROI here is actually pretty profound.

Yes. That's interesting. And I mean, you get to the other 19 people that were on hold also.

Exactly.

What -- and I kind of want to maybe flush out what Twilio's role in this is going to be because you're obviously the voice infrastructure. So consumption-based model, everything is going to flow through you, but you also have some of these software like add-ons that you've introduced into the platform on the voice side, conversation relay intelligence. What role are those playing in this sort of -- in this example?

Yes. Okay. So let's play back that same example, but entirely through the lens of Twilio, okay? So what happens now is that instead of the 1 of 19, you've got all 20 happening simultaneously. That's 20 revenue events that we now have access to that we didn't have previously. And you might say, okay, but some of those we're going to get through eventually, yes, but not in the same cycle time, right? Because what used to happen in an hour just went way up because of the infinite scaling properties of AI. So we get more revenue -- more voice minutes, okay, in Twilio vernacular as a result of the infinite scaling, number one. Number two, you get more voice minutes when a call goes from 1 minute to 2 minutes or 3 minutes. So that's kind of the second revenue opportunity. And then the third revenue opportunity is the price uplift on the voice interaction because none of these voice AI agents works unless they have a context layer underneath, where there's some sort of a data layer that's able to access a consumer's profile in a data warehouse and then able to activate it back to them to be able to solve a customer problem, to be able to upsell them, whatever the case may be. And so we get additional price there. So we win 3x, but the beauty of this interaction is that do we make more revenue? To be sure, but not at the expense of the customer who's getting way more ROI and also the consumer who's getting a better experience. So everybody wins.

Right. Interesting. Okay. We could probably talk about voice AI for a long time, but I do want to talk about other parts of the business. Actually, maybe before we go there, one last question on voice. Just you mentioned it's still a small part of the business. What -- in your mind, kind of as you're thinking of the company over the next several years, is Twilio going to be voice first? Like is this going to become like a majority of your revenue stream? Like how do you think about what that mix looks like in the future?

I think that we're at the beginning of an AI super cycle, obviously. And I do think voice will become a major beneficiary. But I also think there's like a little bit of a demographic gap in terms of like who prefers voice versus who prefers other channels, okay? And so I'll just use myself and my kids, right? Like I'm 52, like I love voice, okay? Like I love interacting with voice agents, like I think the technology works, as I said. And I don't mind synchronous communications, okay? Like my kids can't stand the thought of like being on the phone with someone like in a synchronized way. They want to be able to deal with it on their own time. And the beauty of like what we offer and why I think it will bleed over into multichannel as well as perhaps text only is it's asynchronous. And yes, it lacks like some of the emotive capabilities, but you can deal with that like in a voice context and then actually complete the work, we can see like through our conversation orchestrator, we can seamlessly transfer that over to another channel where whoever the consumer is, can deal with it on their own time through whatever channel preference they have. So I'm not sure, right? Like I don't think I have to have a view on it necessarily. I think we win no matter what so long as this AI super cycle takes place, all of these channels are going to be consumed in a much more profound way. And as the market leader, like we have a lot to gain.

Yes. Okay. All right. Now let's switch to messaging because that is right now the biggest part of your business. And you've also seen on messaging, the growth rate accelerate. So this is a key driver of the overall top line. So what's been happening in that business? What's driving the volume increases? And where is the momentum coming from there?

Yes. So messaging is more or less untouched by AI, like it's not 0, but it's pretty close, I would say. So we haven't really seen the effect of AI there at all. I think by and large, it's just broad-based strength in the business, right? Like it's not really -- I mean, it's not a stable macro environment, so I certainly can't point to that. But I think it's just strength that we're seeing across the board. I mean we analyze the business by channel. We analyze it by industry, we analyze it by geography. And by and large, across every one of those vectors, like we're seeing strength in the business. So that feels pretty good. And I think a big part of that is that self-help story that I talked about earlier, where we put a lot of work into the PLG side of the business, and we went even bigger on that at SIGNAL, where we kind of revamped our one console experience to make it easier, not just for our customer to avail themselves of messaging, but any channel all at the same time, including data. And I think that's going to allow us to continue the pace that we've got in messaging.

What is the role that you see some of these add-on, like premium add-on capabilities in the messaging ecosystem? Like what role are they playing? It's like Verify is a big one, and I think 2-factor authentication is a key use case that you have -- that you serve. How much of an uplift are those driving? And how critical is that?

Yes. I mean I think Verify has been growing at a very fast rate for a while now. I mean I think we have opportunities, for example, like with branded messaging, like I think that's just starting to take off. It certainly reinforces trust in the ecosystem. So that one is pretty exciting. I think we have additional opportunities in terms of utilizing conversation memory in the context of messaging only where you're utilizing knowledge that you have about every one of the transactions that you've had with the consumer to be able to perpetuate your relationship through -- could be messaging, it could be voice, it could be e-mail, it could be all 3 at once. But I think all of those, like it really opens the door, and we're excited about it.

Does it impact your margin profile, your gross margin profile?

Yes. So all of the software add-ons are margin accretive to messaging. Messaging tends to be -- it's a huge part of our business, as you pointed out. It tends to be kind of a structural drag on gross margins. We're not honestly that worried about it, like we're more or less more focused on the gross profit growth profile of our business. And so we'll take what comes in terms of the gross margin characteristics so long as we're disciplined on the price side.

Okay. Got it. And then you mentioned just earlier this kind of big platform relaunch you did at SIGNAL just last month. What were the sort of the key changes that you made? And as we're thinking of -- investors are thinking about what the business outcome should be from this relaunch, what should we watch for?

I mean I think the outcome to answer that question first, should be ultimately accelerated growth. The kind of the impetus behind doing it was -- I mean, you followed the company for a long time, so you probably know this, like I mean, like I've worked at Twilio for 8 years, I probably have like 50 different accounts just because I'm constantly like playing around with the console. And like for the life of me, I -- as someone that worked there, couldn't figure out how to like onboard myself with messaging and voice at the same time. And I'm using those 2 channels because those are ones that we organically built. Like that's shameful, right? And so like there's only so long you can have that persist. And so we did a total revamp of the console to make it super simple, including all of our acquired capabilities for you to onboard any channel, multiple channels at the same time, to be able to ingest data at the same time, to be able to use the entirety of that conversation suite that we talked about at SIGNAL, and it's -- the way that it works is once you're in the console, even if you know nothing about communications and the only thing that you do know is you have a set of use cases that you need to launch, we will help you through an AI assist such that your use cases will drive the things that we tee up for you and then we make them super simple to use. In fact, we allow you with a set of credits to be able to play around in a sandbox, so you can try before you buy. And I would say, so far, like it's going great.

This is like a freemium motion essentially to...

Limited. It's boxed.

Interesting..

It's like a sandbox.

Yes. Okay. And what was the process before was sort of very fragmented.

The process before is, I mean, you had -- it was very fragmented. Not only was it fragmented, like you had to be sort of an expert engineer to avail yourself of multiple channels. And then even if you could, we still sent you a bill that had total fragmentation in it, right? So 3 different if you were using Twilio Segment, Email and our voice channels. And not a great customer experience.

As a part of sort of broader platform launch or maybe in general even, how do you think about your pricing power because you have this sort of unique asset, as you talked about with the Super Network and you're adding these new capabilities on top, these new add-ons that you're obviously charging. But how do you think about core sort of messaging or voice pricing? And does that evolve as the platform becomes better and better and better?

I mean I think we have some degree of pricing power, and we want to be careful about it. Like we certainly don't want to gouge our customers, but we also want to be appropriately compensated for our technology. And so very regularly, and it kind of varies based upon geo, like we raise our list prices like pretty frequently. Now a lot of our contracts are negotiated, and so it takes a little bit of time for that to ripple through the revenue base, but that's a pretty regular motion. I think what's more interesting about all of these console capabilities is our ability to price and package when customers are using multiple of our products. And so we made that hard for them as well, right? So to offer any kind of a discount when you were using multiple products or for you to be able to consume multiple things at once and get the benefit of that, like we made that super hard for you. So in addition to all of the stuff that customers can see, there's a fair amount of work that we did in our back office, too, to completely revamp our billing system. And that's underway, I would say, we keep launching new packages and prices every quarter or so, and we'd expect that work to be done by next year.

Okay. And then on the just context side, that's become a more and more important story, like you're obviously not just the communications infrastructure, you're also serving up ways to make the engagement -- customer engagement more relevant. What -- and maybe part of this is segment, part of this is other sort of enhancements you've made to the platform. But what are those capabilities sort of you're introducing into this customer value prop? And where does sort of pricing for that fit in? Are you just kind of hoping increased consumption drives? Are those priced separately? How do you think about that piece?

You can buy them separately if you want to. But I think ultimately, like what we really want to be able to do is infuse data enrichment into every one of our channels. And look, if you're an AI user, the best way to drive down cost of the LLMs is to add a context layer because now the LLMs are referencing the data that is super specific to what you need to get done versus the data set at large. And so token consumption goes way down as a result. And consumer outcomes go way up because it's referencing the specific attributes, specific experience, specific purchase history of you. So it's just a better all-round experience for our customers and then ultimately, their consumers. As I said, like we intend to price it in the way that customers want to buy it. And so it could be separate SKUs. I think likely it will be packaged and priced in that way more than anything, but it's a super compelling offering. And I think whether it's persistence, whether it's memory, whether it's orchestration, like all of these things allow our customers to ultimately create a much, much better experience for theirs.

Okay. In the time that we have left, I want to turn to another topic, which is just profitability and some of the improvements on operations that you've talked about before. I mean, I think a core part of my thesis on Twilio, obviously, the top line growth is big, but you're also gaining efficiencies and you're driving margins and op income and free cash flow growth higher. What are you doing internally from just an operations perspective that is allowing you to scale margins? And how much more sort of room do we have on the profitability front for that to continue its trajectory?

I mean we certainly intend to continue producing operating leverage over time. Some of that will come from volume leverage. Some of that will come from cost efficiencies. Like on the cost side, what I would say about that is like over the last couple of years, our headcount has basically been flat, while the company's growth has materially reaccelerated. Last year, actually, like our OpEx, as you know, fell, right? So I wouldn't anticipate like that's necessarily a permanent feature. But I don't think that we have to add a lot of headcount to pursue any of our future ambitions either. I think it will tick up a little bit, like so I don't want to pretend like it's not going to grow at all, but we are expecting operating leverage going forward. And look, like there's a lot that kind of gets written about like the gross margin characteristics of the company. I think so long as we continue to grow the gross profit line at a sporty rate, we can drive operating leverage as a result. We can continue growing into new areas by reinvesting some of the profits that we have, but also saving money with a variety of initiatives inside the company, using AI in some cases, getting -- retiring tech debt in other cases, increasing our exposure to different parts of the world in terms of our workforce. We're already remote first. That provides us a real strategic advantage. And in doing so, also reduce materially, we have so far, and we continue to do this, reduce our stock-based compensation, which we think is a great outcome for investors.

What are you doing internally with AI? How are you using it across the organization?

There's 2 things that we've gone big on, and then there's a variety of others. So the 2 big ones, I would say, are both with respect to customer service as well as inbound sales, like virtually all of that now is handled by virtual agents, okay? Now the form factor is a little bit -- or the channel, I should say, is a little bit different, right? So in both of those instances, it could happen over e-mail, it could happen over voice, it could happen even over text. So it varies a little bit based on the channel that a customer would utilize. But there's kind of a twofer in the benefit. Like the customer service side, like our agent there like is generally able to solve the customer's problem without them even realizing they're interacting with an AI agent, even though we disclose it. On the inbound sales side, what's cool there is, is that not only do we get the productivity benefit in terms of lowered cost, but more importantly, the digital sales reps that we have, they can now attach them to -- themselves to an actual qualified lead versus something random that came in. The other area is on the engineering side. I mean, we obviously are using Claude code or maybe it's not obvious we're using Claude, but we are using coding tools. And Claude code is the one that we've employed. We use Gemini more broadly for the entirety of the workforce. We use some functionally specific tools. I would say on the engineering side, we've definitely improved velocity. I wouldn't say it's reduced cost much just because like we're not prepared to push production code that hasn't gone through a pretty heavy review. And then in the G&A functions, I think it's very early days, but I do expect productivity there over time.

All right. Fascinating conversation. That's all the time we have. Khozema, thank you so much.

Great. Thanks for having me.