We'll get started. Thanks , everyone, for coming. Derrick Wood, senior research analyst at TD Cowen. Susan [Extel], appreciate your support. We've got VP of IR, Rodney Nelson of Twilio. Thanks for coming, Rodney.
Thanks for having me.
I am going to give you a softball question to start with, I think. It may just lay the foundation here, but the stock has had just great performance this year. Obviously, you guys have gone through a big evolution.
Yeah
over the last few years, but there are a lot of things. There's been really impressive growth acceleration in the business. How would you set the table in terms of what's really been working and behind the scenes on the organic growth acceleration?
It is just that. It's been called a three-year journey to get to the point that we're at today as a business. We were, I would argue, one of the first software companies that took our medicine after everything that happened in COVID. We made a lot of really difficult decisions to inject financial discipline into the business. Headcount is down roughly 40% from the peak. It's been about flat for two years now. We've been operating at about 5,500 employees since the start of 2024. You now have a team that's largely been in place from top to bottom, with Khozema taking over as CEO at the start of 2024, bringing in a couple of really polished and really outstanding senior leaders to run marketing, run product, and run sales.
Pairing Khozema with Aidan as our CFO over the last several years has been a very dynamic duo in the business. I think most importantly, we inject a lot of discipline, rigor, and focus into many different areas of the business, and a lot of that started on the P&L when looking at margins. Obviously, margins have come up considerably over the last couple of years, going from a business that was roughly break-even to now one that is on the doorstep of 20% operating margins and something similar on the free cash flow side. Here recently, I think where you see the discipline and focus manifesting in the business is in the innovation velocity. We've really doubled down on the areas where we have a right to win, particularly in infrastructure.
We effectively abandoned our ambitions in the app layer that we were pursuing many years ago. The net result of that is a much more efficient and a much more productive product organization. The capstone of a lot of that work here recently came at our Signal conference a couple of weeks ago. Along the way, we've managed to re-accelerate the messaging business, which is our largest business. We've seen some emerging tailwinds in the voice business, which has catalyzed growth there and is now growing at its fastest rate in five years. You're getting a much healthier and much more balanced version of Twilio today than I think you've ever seen. We've augmented that with a pretty meaningful shift in capital allocation being much more tilted towards capital returns, which has amplified the financial performance of the business over the last few years.
It's been a lot of things, not just any one thing. This is, I would argue, the strongest Twilio's ever been in the 10 years we've been public.
Nice. Way to lay it out there. Khozema talks about this renaissance in voice and that obviously we've seen a big acceleration there. Can you just give us a sense of what's behind that? It sounds like a lot of the startups there, I think we, and I've seen other people position you guys as like the AWS of voice for these startups. You're the platform that they're building on. Is it all just the AI natives? What are you seeing from the ISV channel? Give us a little more of a bottom-up look at what's driving voice right now.
I think what's so great about our business is our core platform; our longest-running product is actually our voice API. When Jeff started the company 18 years ago, the very first product was our programmable voice API. We've quite literally been in this business for almost two decades. The one thing that is true for an AI native, that's also true for an enterprise, and that's also true for a more traditional-looking ISV is that none of them are in the communications business. If they want AI agents to do the bidding of what was historically served by a customer service agent in a contact center somewhere, at a very minimum, you still need the connectivity layer, and that's where our Super Network comes back into play. If you have customers all over the world, you need the full breadth and depth of 4,800 carriers globally.
You need to be compliant in all 180 different countries that we operate in. The rules and regs change constantly. You have to be on the lookout for fraud and robocalling. All of those capabilities are things that we've been building and compounding for the last 18 years. They're also really gnarly and, frankly, unsexy problems that nobody wants to rebuild and recreate the wheel on. We've taken something that is absolutely critical and vital and incredibly hard to do at a global scale with reliability. Even though it looks undifferentiated to our customers, it's mission-critical to how they solve these problems. That's just as true for the AI natives as it is for customers that have worked with us for 10, 15, almost 20 years.
I think what's interesting with the AI natives in particular is that the best use of their resources is to build what's differentiated in their product stack for their customers to solve, say, customer service AI. They have to have an enterprise-grade communication stack from day one to even have a shot at winning that business. What we've seen with the AI natives is they not only rely on Twilio but also come to us through our self-serve channel, so they're voting with their feet and their dollars to use Twilio as a provider of choice. They're also then leveraging our reputation for quality and reliability in those RFPs to say, You're getting best-of-breed AI capabilities up here, and you're getting best-of-breed communications capabilities down here, and that's because it's coming from Twilio.
In just about every case, we are providing the telephony infrastructure to actually power that call back and forth between an agent and a human. Increasingly, we're also selling through many software add-ons, some of which we've sold for a very long time, some of which are newer to handle more features up the stack, if you will, while staying in the infrastructure layer. That could be orchestrating the models themselves, the speech-to-text and text-to-speech chain. That would be our Customer Relay product. That could be streaming real-time intelligence off of that call. That would be our Customer Intelligence product. A variety of other features in between.
A lot of the momentum is happening in the voice business. Because customer service is such a rich opportunity, but there is an emerging opportunity in messaging as well that we can speak to a little bit if desired.
When you guys talk about voice being the new entry point for new customers, is that mostly from the AI natives? That's a lot of the times when they will start?
Yeah. I think that's become the new front door, if you will, in self-serve, and a lot of that is AI natives, but it's also for just new businesses that are building themselves in the AI era. They maybe think a little bit less about, How do I go and procure all these different packaged software solutions? Gee, I have all these tools. I have Cloud Code, I have Codex, and I have Vercel. I have all these things that allow me to just build exactly what I want with relative ease, and communications is one of those things that I can now plug into natively, whether that's coming directly to Twilio's new console, which we just launched at Signal, or accessing these tools through coding platforms. It's never been easier to get up and running with Twilio's stack.
It is absolutely the AI natives that are coming through the front door, and that front door is voice, but it's just as much the AI-enabled business, if you will, that's entering.
Right
The door is there as well.
Can you give us a sense of what Customer Relay is and what that ecosystem looks like, and speech-to-text and text-to-speech, and when do you guys win? When do you guys partner? Who are the partners that may be alternatives to using your own?
Yeah. I think it's important for us to always recognize who we are, and then that dictates what we build and what we partner on and what we may leave to others. In our case, we don't want to be a model building company. You can see that in the text-to-speech and speech-to-text landscape. You're starting to see that in the speech-to-speech landscape. You've already seen that at the LLM frontier lab level. The model layer fragments pretty quickly, and we've seen that at just about every single layer of the stack, and that creates complexity for our customers.
Rather than pour a lot of CapEx dollars into the ground to consume a bunch of GPUs to train a model, we would rather partner with a variety of different ecosystem players, give our customers as much choice as possible, but do it through a single product where you don't have to then go out and procure maybe five, six, or seven different vendors to do everything from speech-to-text, text-to-speech, the LLM, the speech recognition, the term detection, the interruption handling, and oh by the way, the connectivity. You can do all that through a single provider in Twilio. The added benefit is you get better input costs because you're not paying rack rates to the individual model providers. You're also getting the ability to pick and choose which models you want to leverage.
If you want to switch from model A to model B because model B is maybe now just a hair faster or has a bit better speech recognition, that's a very simple configuration change on a massive re-architecture of a software platform. It gives our customers much greater flexibility than being overly committed to any one model provider.
Yeah. You play the orchestration layer into the models, whether it's an LLM or a speech model, and drive the connectivity there.
Exactly.
The use cases, you talk about customer service. That's, I think, one everyone's familiar with. Are you seeing outbound sales and marketing voice AI? What are you seeing in actual call centers? Where is the strongest adoption from a use case perspective?
I think what's interesting is you're seeing emergent use cases in just about every industry. We've talked a lot about how in healthcare it's now much more efficient and a much better experience for the patient to have an AI handling billing questions. The bills are often very confusing. There's a lot of words on there that the average person is not going to understand, myself included. It might require a patient to ask a question three, four, five different times just to really fully understand the answer. If the person or if the thing on the other side of that call is a human, a certain degree of frustration is going to build. They have a cycle time they have to be managed against. With AI, none of that exists. It can be endlessly empathetic. It can be endlessly sympathetic.
It'll answer the same question multiple times if need be. That call actually winds up taking a little bit longer, but it still costs our customer a fraction of what it used to cost them to solve that patient's problem much more effectively, even if the call now runs seven minutes instead of three.
You're also seeing there have been a lot of service -level businesses and home service companies, many of whom consume our capabilities through an ISV, where they are getting inbound appointment requests in the off hours when nobody is around to actually take that call. They're effectively finding revenue by having AI agents serve those shoulder periods where maybe a call comes in to a plumber at 8:00 P.M. They're not monitoring their systems, and they're not answering their phone, but the AI agent can take the call. That's found revenue. That's an appointment they otherwise would not have gotten. You are also starting to see a bit more outbound as well. I'd say it's predominantly sales, a little bit less marketing.
Anywhere that you can have a conversation and leverage AI and memory and intelligence to drive a more intelligent conversation, whether it's inbound or outbound, we're beginning to see experimentation in all those different dimensions.
I know you guys don't guide to this.
It's a good pretext for a question.
Yeah. In terms of what could be elements underneath the model that can keep growth accelerating or just see new catalysts into the model around AI, whether it's expanding use cases, moving more into the enterprise, or just what you talked about on what sounds like increased minute usage .
Yeah
More new customers at the front door, or what you announced at Signal under the persistent memory, and there's upsell, cross-sell there. That's a lot to throw out there.
Yeah. I think maybe as a starting point, at least through the lens of AI, it is still exceptionally early. I've called three different hotels in the last, call it, 72 hours. Two are part of a major chain. One is a boutique. I was met with zero AI. These were very basic customer service questions. I was literally scraping around trying to find a bill to submit my expenses. That is tailor-made for AI. All you need to do is plug into the billing system, have some understanding that my phone number is probably associated with that bill somewhere, and you can trigger that instantaneously in the context of that conversation. Yet, I was put on hold for +5 minutes in every single instance. Even there, where it's a fairly rudimentary use case, we're not yet seeing the proliferation of this technology yet.
I think that does create some secular tailwind that I think can benefit not just the voice business but, frankly, all of the channels. Because one thing that we firmly believe is if you're going to roll out AI and have more intelligent conversations with your customers, you have to recognize that those conversations happen everywhere. Personally, I would much rather text that hotel and get my bill that way than have to spend time on the phone, whether it's a human or otherwise. We do see more customers, whether they're AI natives or our direct customers, increasingly looking at how they can leverage all of these channels in sync with one another, and that's where elements like Customer Memory, Customer Intelligence, and orchestration come into play.
In the old world, if you were building with Twilio, you'd have to write glue code to staple all of these channels together. That's effectively what Orchestrator now does out of the box. You'd have to have some agent or some platform that's giving you that real-time insight into what's happening in that call. That's been Customer Intelligence for Twilio for the last couple of years, but that product has continuously evolved to include more channels and become real -time. You need some layer that's going to reinform the next communication of who I am to that business.
If I have a phone number on file with United, every time I call, they should have some idea of why I'm calling, what flight it's related to, what issue could be a part of that call, and effectively head off the reason for calling before they even pick up the phone. That is all possible with Memory because we can extract those insights off of that call in real time, reconcile it back to a profile, de-duplicate observations that we see, and maybe eliminate previous observations that had been filed in that profile. You always have the most current, most informed view of who that consumer is to then deliver a relevant and contextual response to whatever the next use case may be, sales, or service.
Yeah
marketing, identity, et cetera.
That sounds like a big deal. What was new that really came out this year on memory capabilities? You just said it.
Yeah
to stitch a lot together, and now you've got much easier access to that memory to orchestrate that?
Yeah
What's been the big advancement?
As I'm sure folks in the room remember, we bought a business called Segment in 2020, and the vision of what the communications plus Segment capabilities were supposed to look like is effectively what Memory is now today. The challenge was, see above, if you wanted to use Segment in concert with maybe a custom source system over here and all the disparate channels that Twilio would provide, there was custom pipelining that would probably have to be done to that custom system. There was stitching of the channels together. Those channels did not live in a single pane of glass. There was a separate system for SendGrid and a separate system for messaging and voice. Segment was a separate platform altogether.
There was a lot of maintenance code, effectively, that had to be written to even get to the point where you could begin streaming real-time insights and reconciling them to a profile. We took a lot of the DNA from Segment, and that's what belies what's going on in the Customer Memory product today. Instead of going through an onerous multi-week, multi-month sales cycle to buy Segment, implement it, and then do all of that work, Customer Memory is simply an API, and it's natively integrated into all the channels in our new console, so that the second you turn it on, which is literally just effectively a click of a button to turn it on, you can begin streaming insights into Memory, capturing them in a profile, and going, even if you're from a cold start.
That's a far cry from a multi-month sales cycle, probably a multi-month implementation, and some additional custom work to get to that point. Effectively what we did was take the best components of Segment and make it accessible as an API that is natively integrated into all of the channels.
The APIs into customer data sources.
Yes
their own data sources.
Exactly. if you want to take that memory—
Yeah
We have APIs to ingest enterprise knowledge, but then if you also want to take the data that lives in Customer Memory and pipe it back out to source systems, that is available to you as well.
That's awesome.
That's frankly a feature of our platform.
Yeah.
We wanted to build these things in such a way that you don't have to rip or replace a single system if you don't want to have these tools available to you. You can begin deploying them and gaining insights from them from day one.
Again, that's part of the value prop : we are a neutral platform that can be interoperable and play well with any model provider, any source system, any data lake, and any data warehouse. That whole idea is to make the platform accessible; that's also where our Twilio Agent Connect SDK comes into play. You can deploy that in any ecosystem. It's self-hosted; it's a very simple SDK. You can pipe agents into the platform from there.
Where does LLM sit in your stack? Where are you actually leveraging AI internally for users to be able to maybe execute multi-channel orchestration better or build campaigns or something like that?
Yeah.
Yeah.
Yeah. We're of the bring -your-own-LLM-
Yeah
Mind you, we want to play well with everyone. We don't try to funnel a customer in the direction of a single model. There's actually only a small number of products where the P&L of the product, if you will, would be burned by an LLM. Customer intelligence comes to mind.
Yeah.
That product is effectively monetized on ingestion, and then custom operators would have running over the top of that call. If you're a dealer and you have an AI agent taking inbound calls for test drives, 95% of the words on that call probably don't mean a whole heck of a lot in the context of that business. If I say make, model, trim, and year, I want those insights extracted and reconciled back to my profile. If I say that I have four kids , I'm talking about a mid-size model. That should be a trigger to the salesperson.
Like, hey, maybe test drive the full size with the third row that can seat seven because you may need it someday. All of those things should be able to be captured in real time to inform that next engagement, whether it's the follow-up text that I'm going to get or the in-person communication I'm going to have on-site with the dealership. That is a product where you would have some burden of an LLM that gets embedded in the cost, though.
The real value to us is those generative custom operators and then the piping back to memory.
The orchestration, by the way, between those different communications channels. We don't actually have a lot of product where we're just creating a wrapper around an LLM by design. Like that's not the value that we want to provide. We are consumers of tokens in our engineering organization, of course, but there's not a lot of that embedded in the business today.
Are you monetizing API calls in that memory layer?
Yeah, that's effectively the pricing mechanism for Memory.
API calls
It's effectively API calls. Intelligence, again, is kind of ingestion and then operators running over the top of it. Orchestration is again on more of an API call basis, somewhat nominally priced. Again, the whole intention there is we don't want to preclude people from driving more volume on the platform. If you want to send that follow-up communication via text and then engage that consumer in more of a two-way communication, but that's orchestrated based on what was said in a service call, we don't want to preclude that from happening. The economics are probably richest in memory and intelligence.
If that also comes with the added benefit of more volume. You've really hit a home run with the customer.
The new products that were announced, I mean, do you see them as being any kind of material catalyst in the next 12 months? Or are they more enablers as opposed to direct monetizers?
Well, I think for starters, we don't really have outsized expectations for them this year. I mean, they went GA in early May. We're halfway through the year, effectively. They're not really a material contributor to the guidance that we have out there for the balance of the year. These are products that we've been talking to our customers about for the better part of two years. I do think that made this product development life cycle particularly effective, as we were constantly gathering feedback from our customers along the way of, number one, where do you want us to innovate? Number two, what do you want it to look like? There were some additions that we made along the way.
I mean, when we think about Twilio Agent Connect or some of the innovations that we launched in partnership with the coding platforms at Signal, like those were maybe not on V1 of the roadmap, but we had to, not pivot, but make adjustments on the fly as the AI ecosystem continues to evolve around us, and those were very well received by our customers. On top of that, we were very intentional about bringing partners into the beta for a lot of these new products so that they would be fully enabled out of the chute to kind of talk about how you could use these products, what the prevailing use cases are, which has been a point of emphasis for Thomas and his team over the last couple of years, is to really lean back into that partner ecosystem.
Who were those?
I mean—
Did they come on stage on the...?
We had a number of them at Signal.
Day two?
They were there on day two. They actually had some incredible sessions that could basically show you, in the space of a single prompt in Claude , going from a cold start with Twilio, no account, no API key, and after 30 minutes of Claude doing its thing, you had working Twilio products and working Twilio use cases, which is pretty incredible. We've never necessarily had the GSIs as a big part of the story, simply because APIs don't require a lot of heavy lifting. There are some incredible boutique partners that we've been working with for a long time who participate in that beta and are kind of enabled to deliver these new innovations, too.
Okay. They were service partners?
Yes.
Nice.
Yes, absolutely.
Okay. Well, I think we've got to talk about the messaging business again.
Of course.
I mean, that also has accelerated. Has that been because of the synergy with voice, or what else is behind that?
I mean, a little bit. We try to position the platform as conversational and omnichannel by default, and that's exactly the experience you have in the new console. There's no longer this fragmented, disparate, one -channel lives over here, one channel lives behind this login over there. You now can build in a much more conversational fashion in the new console because all the channels are right there available to you. That's more on the come because the console was just launched a handful of weeks ago. Look, we've seen kind of remarkable resilience in the messaging business, and it's not coming from any one place. Our top five or six industries, which make up the bulk of the business, are all seeing really healthy rates of growth. Tech, financial services, healthcare, professional services, retail and e-commerce, advertising, and marketing. I mean, they're all performing very well.
Geographically, the U.S. continues to hum along. International, even though the messaging business is going through some tough comps internationally right now, continues to perform pretty well. Then, I think we've seen pretty good resilience and durability by use case. Verify, if you want to use that as a proxy for the authentication business, that continues to be one of our faster-growing add-ons. It's also one of the larger add-ons that we sell that continues to see really healthy adoption in a variety of different industries, not just regulated verticals like FinServe and healthcare. The marketing use cases continue to hum along. Notifications are pretty steady. The one new thing, or maybe two things, that we saw in Q1 that stood out, we did see a much stronger seasonal recovery. Q1 is typically a seasonally slower period for us, especially coming off that holiday shopping season.
The bounce back that we saw in Q1 was a bit stronger and a bit faster than we typically see, which was great to see. Again, not necessarily isolated into any one area. We have also begun to see the AI natives leaning into the messaging channel as well. Now, that's mostly for traditional Twilio use cases, notifications, and 2FA. Way out at the margin, there is some experimentation beginning with two-way use cases. That's not really a growth driver today. That is, again, if you're going to have conversational experiences with your consumers. A good number of folks want to communicate via the messaging channel, and so that's something that we think is a possibility over time, especially with these newer features that we just launched.
Does RCS and WhatsApp, do they move the needle, or are they pretty small?
They're still pretty small. Both are growing very quickly. They both pale in comparison to the size of SMS still today. You're still making some trade-off every time you move from SMS to those channels. They're not as ubiquitous. WhatsApp, in particular, you have to find markets where you have good agreement between businesses and consumers that yes, we should all be here. In most markets, everybody wants to use WhatsApp for peer-to-peer. In a lot of markets, businesses would love to use WhatsApp as a business messaging tool. In very few markets do both of them want each other there. Brazil is probably one of those exceptions where there's an incredible ecosystem around WhatsApp for business messaging and business calling. That business has done well.
RCS, it's a similar story, but it is carrier-enabled, and it comes with a lot of added benefits that have been native to WhatsApp for some time: branding capabilities, richer media. I do think some of those things, if you're going to scale two-way use cases, are table stakes. I'm much more likely to text back and forth with a business that I recognize with a logo and a name than a random short code or a random toll-free number. It's still very early days. The carrier support continues to come online. Handset support: if you're on a very old iPhone, you still can't receive RCS messages, so there's a hardware cycle element to it. We are beginning to see those volumes grow pretty materially, albeit off of a very small base.
Yeah. You mentioned seasonality before. Can you remind us of other seasonal factors that tend to happen throughout the year in Q2, Q3, and Q4?
Q1 and Q4 have the most pronounced ones. Q1 tends to be seasonally slower, but it's because of what happens in Q4. Around the holiday shopping season, there's a lot of activity for obvious reasons. We tend to see pretty sharp acceleration in volumes around Cyber Week, so Black Friday, Cyber Monday. You usually sustain some degree of elevation through the Christmas holiday, and then towards the end of the year, you see volumes fall off a little bit as businesses and consumers go their separate ways, and then a recovery as you move through Q1. That's the most consistent seasonal pattern that we see. We get asked a lot about cyclical events. Politics comes up very frequently. That's been a smaller part of the business over the last several cycles. The bar that we set for customers to be on our platform is very high.
We have an acceptable use policy. You have to get opt-ins. If you are in the business of sending political messages and you follow all the rules, you can be a Twilio customer. There just seems to be a lot of traffic that doesn't want to follow those rules. We made a decision years ago that we would, number one, force customers to register all their messaging traffic, and then number two, hold them to that.
Yeah.
I wouldn't expect politics to be a material factor in this particular cycle. In part because the rest of the business continues to grow.
Yeah.
As a percent of the mix, it'll just get smaller over time.
Yeah. We do have midterms.
We do have midterms, for better or for worse.
Okay. On A2P, the new carrier fees, you guys historically haven't seen any kind of loss of demand with prices that go up that you guys pass through. Do you think that the same thing's happening with this cycle? Clearly, it's an issue.
Yeah. I think we've heard more from customers in this particular cycle than maybe in previous ones. Look, nobody likes a price increase, which effectively is what this is. Now, all of our customers recognize it comes from the carriers.
Yeah.
We've heard more noise about it. We haven't actually seen any behaviors change yet, though. We're also not naive to the fact that this is a supply/demand equation, and at some point, if prices go high enough, businesses, in particular small businesses, may have to think about how to deploy what are pretty rigid budgets as effectively as they possibly can. That's where having the breadth of channels that we have is so critical. If you want to maybe deploy some of those dollars to WhatsApp or to email, or if it's a use case that may have some benefits in moving to voice, we can obviously help you get there, and that's been a consistent line of communication between our sales teams and our customers. We know that potential is always out there, but again, see above.
That's why the breadth of the platform that we have is so critical, so that customers can find high ROI avenues to deploy this spend. I'd say it's maybe a bit more focused on marketing, which has more discretionary elements to it. Although customers use us for performance marketing with existing customers, typically. It's maybe a bit less relevant for account notifications in 2FA, where if that six-digit passcode is what's standing between me and transacting, it's a trivial price to pay to get me authenticated safely. Similarly, if United is sending me that text message saying, Hey, you were at gate C100. Now you're at gate C1. You better get moving, or otherwise, you're going to miss your flight. They can't run the risk of that not getting delivered, and messaging is still the most effective place to get there.
That is something that we hear, and we're pretty adamant about educating customers about the other channels that we can provide.
The new console should help, right?
I think the new console will help in a variety of ways. Number one, the ability to experiment in all of the channels is so much simpler.
Yeah.
If you're a brand new Twilio customer and you've never experienced the console before, you now have an overhauled trial credit system where you can deploy trial credits across all of the channels if you so choose. You can get comfortable with how you might orchestrate conversations and workflows between various different channels. The add-ons are all right there for you. We've simplified the error codes; if you run into a problem, an AI assistant is right there on the dashboard to tell you what that error code means and what the resolution steps are. There's no more; you're eight clicks away from resolving your problem because you have to go to the help center, engage an assistant, and maybe submit a ticket. That's all being done right there in the new console.
You also now have much better visibility into your usage telemetry and your billings, which was a frequent pain point. The ability to get up and running in multiple channels and consume our add-ons has never been easier in self-serve, and having visibility into what you're doing has also been leveled up dramatically.
All right. Great. Amazing. Great conversation. Thanks.
Of course.
We're out of time.
Thanks, Derrick Wood.
Thanks, everybody.