Five9, Inc. (FIVN) Earnings Call Transcript & Summary

Everybody, thank you for joining. My name is Matt Stotler. I'm the analyst here covers Five9. For a list of our disclosures, please visit our website at williamblair.com. Maybe we'll start off by just going down the line, please introduce yourselves, your roles at Five9, and then we'll hand over to Barry for a little presentation.

Barry Zwarenstein, CFO.

I'm Jonathan Rosenberg, Chief Technology Officer and Head of AI, don't normally attend investor conferences, but somehow people wanted to talk about artificial intelligence.

And I'm Dan Burkland, President and CRO.

And before we get started today, welcome, everybody. Just wanted to [ say ] that we will be making forward-looking statements about events and trends that could affect the industry, the company and our operations, including future growth, product development and the like, AI and automation, of course. Actual results may differ materially. We have no obligation to update that. And things that could cause such a variance please refer to our securities and exchange filings and 10-Q, 10-K under the caption Risk Factors and other areas. Thank you.

Go that number. Yes. So if you like to start off, I think -- would you like for a presentation?

Okay. I'll start off with just giving a quick brief overview. For those who are not familiar with our story, what Five9 does is we provide a platform to help companies handle all of their interactions -- their customer interactions. So think of that as regardless of whether they're coming in through voice, chat, e-mail, SMS, social and whatnot, they land on our platform. . And then we take care of those by handling them in a variety of different ways. And that's been one of the big things in our industry as of [ late ] and of the investment community is how is that going to evolve and change over time. Historically, we would accept, let's use voice calls as an example. And on the platform, identify who the caller is, identify what their intent is for their call and then assess among the entire population of resources to handle that transaction or interaction. And determine who's the best skilled and currently available and we would route that interaction to the agent and also reach into the CRM or other databases and other sources of data to deliver that corresponding data to that agent simultaneously, and they would complete that transaction. So we were kind of that interaction management engine. Started out in the early days many years ago with just voice. We added chats and e-mails so that we could assess what those were and be able to determine how to invest route and have those handled by agents as well. And what's happened with the advent of AI is it's allowed us to now take in these interactions and determine how much of the interaction can we handle in a, what we call, self-service or in an automated fashion. So that we don't have to tie up the expensive -- the most expensive resource in any contact center is that human labor. And so sometimes, it's just the beginning of the call, where we're just deriving intent, we're identifying the caller, we're authenticating perhaps with voice authentication, voice biometrics. We may be asking a few pertinent questions so that when we get it to the agent, we've already collected a lot of data and information, and we can cut off, time off the beginning of the call. We've always been working as an industry to work on improving the efficiency and improving the amount of time that the human labor has to spend with a call because that's oftentimes 5 or 10x more expensive than applying software to automate certain functions. We can even go all the way to the extreme in some cases, fully automated customer interaction where they can talk to a bot that IBA, as we call it, an intelligent virtual agent, handle the entire transaction, if it's something that's highly repetitive and a typical question. We can also do that with where we can have the chat, bot handle that interaction and complete it. Many interactions have a combination of automation and the human element, as I mentioned. And we also have tools that will help us, listen into a conversation, transcribe it in real-time, fetch information to coach an agent, help the agent to be more effective and efficient at their job, all designed in that first premise was -- is to improve the efficiency in the contact center operation. The thing to remember about Five9 is we are the interaction platform, handles those interactions. And as technology and innovation becomes a bigger piece of that pie, we have to sell them more software in order to handle those interactions. And the more this automation occurs, the higher our ARPU or our revenue per customer increases and we get more wallet share. So we've been looking forward to this revolution of automating more and more transactions because we benefit from that. It increases our TAM. It increases our ARPU, as we mentioned. And that's what we're here today to explain that to you and hopefully answer any questions.

And I'll take it from here. So like I said, I'm Jonathan Rosenberg, I'm the CTO and Head of AI. And you don't have a CTO show up at a conference without showing an architecture picture, right? So here's a diagram of sort of what our software does, just taking a little bit deeper than what Dan just said. So it starts here on the left with what we in the industry called channels, which are different ways a customer can contact the brand, whether that's picking up the phone or making a phone call or receiving one, going to the web page and there's like guys ever sees little chat buttons on the bottom right hand of the page, cut that. So we provide that software, for example, that does that little widget on the bottom right. A full website that has content dedicated interaction, setting an e-mail, more and more SMS is becoming mainstream for communicating with brands, WhatsApp, whatever. All these different channels, and so -- and there's a lot of work in delivering software that handles all those channels. And so we bring in those channels into our platform, and that's the second part here. And what this platform does, is it handles that interaction in whatever way we have to handle it. It makes routing decisions. It does reporting and analytics. It decides what agent to send it to. It shows the supervisors what's going on live during the day, so they can manage their staff and so on and so forth. And I'll get into a little bit more about what's in there, but there's a lot of stuff in there that handles that interaction. Big part of that is dealing with third-party data sources. And this is sort of one the things I think if you're not familiar with the contact center industry, that's in the not obvious category. But when you call up the contact center, almost all the software needs to access internal application systems. So if you call up like FedEx, and you want to know where is my package and a voice spot answers that, it can only answer that if you can plug into FedEx's databases and systems that have voice botch, knows where the package is right now. So typically, the bigger the company, the more such integrations we plug into dozens often. And so our platform integrates with those third-party data sources. So to make it real, and this is -- there's a lot in here. There's no quiz at the end. Don't worry. And what I'm going to do to sort of give you a little bit more detail is actually like take a typical customer flow through our platform to help you understand what happens. So let's say, why don't you pick up the phone, and what you're going to do is -- they place a call, so that's going to light up our voice network. We have a global voice network that has points of presence all over the world. We're actually a regulated carrier in the U.S. in order for us to do all of this. We have a huge amount of phone numbers that exist in our platform, for example. And so that voice call comes in through our global voice network with all the security compliance there. And what it does is it lands on the sort of the beating heart of our contact center, which we call the core ACD or routing engine. This is the engine that receives that interaction and figures out what do I do now? So for example, it may make a decision that this call needs to go to an agent. Well, which agent does it go to? Well, a live agent, for example, I've got lots of live agents, which one is available right now, which one has the right skill to handle this call or it might decide it needs to go to a bot, is an interesting example. You might have a case where a high-value customer calls in, and you want to put them to a live agent. Everyone else is going to be handled by a bot for their particular use case. So this is the kind of logic that our routing engine does. In this particular use case example, let's say this call gets routed to a virtual agent. We call it an IVA, as Dan said, and it's the thing that says, "Oh, thank you for calling. How can I help you today? And it's an automation that does all of that, it greets a customer and ask what they do and they want to fulfill it. To do that, we, as part of our strategy, plug into what we call third-party AI engines. Five9 has never been in the business of building our own machine learning models. We take other vendors, models or engines, and we apply them into our use case. So in the case of a virtual agent, there's actually a lot of stuff we have to do. We need synthetic voice. That's the thing that speaks to you. Thank you for calling, how can I help you today. We use Google, we use LuminBox. We actually recently signed a deal. This is really interesting. We found a company that does dubbing for films, right? And a lot of times when they dub them, they use -- they want to use an AI to generate the voice. So they hire an actor. The actor records like 20 minutes, 20 hours or something of dialogue in that language. And then once that's done, their AI model can generate really natural sounding voice in that person's voice. This has been used -- we makes a -- who was the guy who is in -- a real genius my gosh. The guy was in real genius who had throat cancer. Do you guys know what I'm talking about? Nobody knows what I'm talking about? [ Valcoma ]. So [ Valcoma ], for example, as he got the throat cancer, like he wasn't able to speak anymore. Companies like this go collected material of what he had spoken before, and they were able to generate synthetic voice that sounds like him. So this country, a company called, WellSaid, fascinating company, that's what they do. We plug them into the contact center now so that you can get this kind of really quality voices and they sell different voices. That's a naming stuff. So there's synthetic voice, there's speech recognition to understand what's being spoken and converting it to text. There's natural language processing that takes that text and then convert it to understanding and meaning that lets us take action. And then there's things like biometrics, like authenticating who you are based on your voice print. So our platform, this is what it does, is it knows how to plug in all these different components in the right way and interact with the customer. All right. So that's where we read the customer. Thank you for calling. The customer said, "Yes, I'm really annoyed. My package hasn't showed up like, when is it going to be here, right? That's what the customer said. Okay. So then what we need to do is we need to fulfill that inquiry. We've just determined that they want to do package tracking. How do we do it? Well, we have to figure out who this customer is. So we plug into customer relationship management products like Salesforce, one of our biggest ones. We have tons of integrations into sales force that has to be actually customized for each of our customers have unique sales force implementations. We pull out the customer data and we might go to some other system, tracking system, maybe it's a marketing system that knows something about promotions that should be spoken to this customer, lots and lots of these third-party data sources. Our platform can be configured with a drag-and-drop visual designer to allow such integrations to be built. All right. So plugs into all those. In this particular use case, maybe not a package tracking one, but maybe it's something where this customer received an offer for a new product or service, and they're calling about that. We want to put them on the phone with a live agent is the best way to close new business is to connect them with a human, not a bot. So now that we determine that they want to buy a product, say, we transfer them to a live agent and that's part of what our system does. Finds the right live agent, connected to them, bridges the call in, we have a UI that sits in the browser that the agent looks at that can hang up, transfer, play announcements, all that kind of stuff on the call. Now that what we do is during the call itself, we want to assist the agent in doing their job more effectively. So we have a product called Agent Assist that sits on the agent desktop. It's an AI module that's listening or receiving the transcript of the call and based on that, it provides guidance to the agent. So if the customer says, "Oh, yes, that gold package is too expensive. It tells the agent, oh, inform the agent that competitive offerings are often twice as much, and this is really a good choice, right? So that kind of guidance is presented to the agent. We also have a feature called summary. Where at the end of the call, we automatically summarize the call, and we take those notes and we put them in a case system so that next time you call back, you don't have to repeat yourself to the agent, which many of you probably have had the experience where you need to do that. So those get engaged on a live call. So here we apply AI into the live call. And then we might take certain interaction. So for example, in this case, where the customer is buying a new product, we might want to automatically send that customer an e-mail at the end of the call that has like a summary of the new product description and send them a text, taking them for their call and giving them a link where they can find out more information about the product. So our workflow automation solution allows customers to configure these type of workflows that interact with a customer here through e-mail and SMS. And then finally, at the end of the call, we do reporting on this. Helping the customer know, hey, did you know that 23% of your calls into your sales organization are completing with a closed win, but 77% are not. Like that's super important information for the brand to know about and our reporting our analytics, our real-time dashboarding give customers ways to get an insight and visibility into what's happening. So this is all the stuff that happens under the hood inside of our platform that is engaged in handling all these different types of interactions. And the main thing, I think, to take away from this is, one, Five9 is agnostic of the underlying engine, large language models, technologies like Open AI and chat GBT, we integrate and plug into them, the better it is, the better it is for us. We're a consumer of that technology. And the second is, there's a huge amount of stuff that happens on top of those large language models for us to actually do the job that we do for our customers. So with that, I'm going to hand it off to, I guess, Barry, for...

Thanks, Jonathan. So a real brief, will take less than 2 or 3 minutes. So don't worry about that. Consistent revenue growth, this last quarter, we grew 20% year-over-year, 5% above the consensus. When you think about Five9 revenue, think of it in 2 buckets. The one is where we have new logos that Dan and the team bring in, they go live. The other is an installed base and how that grows. . We have extremely good visibility on the new logo side, big backlog. We actually had a record in Q1 in terms of see turnups less so in this macro environment with the cross currents on the installed base side. But still a pretty reasonable visibility of all assuming the economy stays approximately where it's at. One of the key things of Five9 is moving up into the enterprise, defined as approximately more than $120,000 per year in annual revenue -- recurring revenue. We went public, we were at 61%. This was 9 years ago. Now we had 86%. I'm going to go into a little bit more detail over here. The reason that you have that sort of camel's hump in the middle is that from COVID, everybody wanted to go into the cloud and working from remotely rather than in big contact centers. And then recently, we also had some macro headwinds. We've especially gone way up in the market. When we went public, we had 3 companies that had more than $1 million in ARR. Now it's 161 at the end of the fourth quarter, and it comprise of currently of about a little over 50% of our revenue. Not shown over here, and Dan will talk about this, I'm showing the Q&A, is 2 of the -- probably the largest contact center in the cloud deals, a packaging company -- a package delivery company and a health care company, respectively, about 40 -- excuse me, $50 million, $40 million ACV. So people really starting to take notice about the success that Five9 has had in moving right way up the stack. These customers have got excellent retention rate that we typically done in line in the entire platform in the entire company. We expand over time. And we have a general decline right now in the retention rate despite that because of the fact that we have the installed base that 1/2 of our revenue growth year-over-year that is facing some headwinds due to the macro. This shows the evolution of both our revenue in blue and the profitability from the red to the green. Q1 is always a weaker quarter for us as you can see from each year. Longer term, we expect to get to an EBITDA of 23%, which is where we were basically in the fourth quarter of 2018, we've been making some investments for these mega deals that we've been talking about across the world in terms of cloud operations and professional services. Finally, on cash flow, very strong and continuing cash flow. The reduction you saw in Q1 2022, that is primarily due to some earn-outs that we had for acquisition we made and the Zoom transaction, which was not consummated. With that, I'll turn it over to you.

All right. I appreciate that over, and thank you very much. I think we'll start by spending a little bit of time on AI, just because it's obviously topic du jour and a great interest in the contact center space. Maybe we'll start with structurally, when you look at the type of interactions that are happening in the contact center today, what portion of those do you think ultimately become automated over time?

Yes. So that's a really hard question to know because it varies dramatically by brand and by industry. So some customers that we have, like the whole reason they have a contact center is like high-touch feel closures. And that answer there is going to be 0% automation. We have other customers. For example, we have a customer that does like ride bookings for a patient transportation, and like they would love to get it -- they love it -- all of them could be automated, which is still tough to do. So it varies dramatically. There were -- and part of it too is sometimes you want to talk to a person and there's a dimension of it that's human emotion. I mean you all hear ever sit in the audience like they were call a contact Center and press 0 because you want to get to an agent because you want to speak to someone live, please raise your hand. Okay. Not you. Apparently, you're the only one, you're the only one that have never done that. Everyone else, like once -- so there are still cases where people want to talk to a human, as they're angry, they're upset. And those are likely to be automated by...

Right. And there's a lot of questions about what happens for contact center providers like Five9. Let's say, if you do automate more interactions, there's a need for, let's say, fewer live agents or at least less live agent capacity, right? That's one of the questions out there. So first, I'd love to get your view on I think there's 15 million, 16 million agents in the market today, how does that trend over time? And then how do you monetize AI? And what does that mean if more of that gets automated for Five9 over time?

Okay. Yes. So if you look at it -- as more automation occurred -- when we looked in my example earlier, Golf comes in and it's handled the traditional way by a human agent, great. We get about $200 per seat per month for human-tended transactions. As we add to Jonathan's example, he just said, automation to that sequence, whether it's automation being added to assist and help the agent be more effective and more productive, okay? We charge more for that. Software that we sell to them on top of the $200 per seat. As we sell things like auto summaries, where we can summarize a call very quickly in a matter of GPT being able to do that work for them rather than have them type in all their case notes, okay? We charge for that, and offset that because the ability for us to charge. Remember, we're offsetting a very high price, sometimes $4,000 a month, human labor to do that typing versus the software. So we can charge a premium for that. All the way to the full self-service, either IVA, the virtual agent or a digital agent where you can do a chat. That's fully back and forth, bidirectional and complete the transaction. So we go from today's pricing all the way up to 400 plus from a monetization standpoint. And so the more that gets automated, the more wallet share we receive and the better it is for our TAM and for our revenue per customer. So over time, to Jonathan's point, it's impossible to predict the exact percentage and how that pendulum swings. We know it's going to swing over to more and more automation. But customers -- our customers, companies and brands are balancing that human element, that high-touch concierge service, delivering a great customer experience that gets loyalty and NPS scores and therefore, get spread about delivering great customer experience versus offloading to automation. Most of the brands I work with on a daily basis say, it's not an either or we're not going to push people to automate or not going to do just concierge service, we want to deliver the high touch. But we also want to give that same consumer the option, they prefer to go get self-service very quickly and easily because that's what their preference is at that point in time. The same consumer may use both vehicles, actually uses multiple vehicles, not just self-service or human, but we use on our smartphone, we use websites to do things like reservations and [ aspect ] resets. So there's a lot of self-service already happening that one could argue, gee, hasn't that reduced the number of agent, 5 agent seats dramatically when we use the websites and the apps. No, it does. More options we give consumers the more they're apt interact with the brand. And actually, the interactions go up, and that's where we win because our platform is handling all those interactions, not just voice and not just human-enabled ones.

I mean, a simple way to think about it is Five9 doesn't sell human agents, but we do sell virtual agents. More human agents are replaced by virtual agents, the better. We just don't think it's going to be 100%. But if it was 100%, woohoo.

[indiscernible] Yes.

What do you think the impact of generative AI is specifically and in particular, when you look at chatbots and the chatbot market.

Yes. So it's a revolutionary technology, make no mistake about that, if you haven't all actually played with it yourself like you should go do that. We think it's going to make these things far more capable and they've been able to be before. I think that's a lot of people have used a voice bot or a chatbot, often get frustrated. Sometimes they don't even try. You just press 0 or press agent type agent right away. But for those that have tried sometimes it's frustrating. It doesn't quite understand language. It can't probably do the right things. So we believe that large language models will make the technology higher quality, improve effectiveness and thus increase the percentage of interactions, which don't need to overflow to an agent for those reasons. They won't handle the ones where it's about a [indiscernible] or high touch, but the ones that are -- well, it should have been able to handle it just didn't work. Those will now continue to be handled by the chatbot and that's woohoo for us. The second thing that's a really big deal is it reduces the time to implement and deploy this technology. Traditionally, in the prior generation, you had to like train the models. You probably heard about that. You have to train model on data. With large language models, that's not really necessary anymore, that used to take via time-consuming process. We had to do each customer one at a time that reduces that, which reduces our time to value and improves the customers' time to realize their benefit. And it also expands the market for AI applications because it required all this model training and tuning. Traditionally, the best highest quality bots were deployed only at the largest contact centers that can afford it. Now if you're an 1 agent contact center or 10,000 agent contact center, you'll be able to have a capable body at your disposal. So that increases our addressable market for these assets.

Right. And when you mentioned data at some point, obviously, the foundation of AI is as long as good as the underlying data, right? And where Five9 sits as kind of a hub for interaction in a given business, it seems to be a very advantageous position for that. So what is the data strategy for Five9...

I'll touch on it and let Jonathan go deeper. But what he just mentioned was the data that we've positioned over the last few years was primarily there to train the model to understand spoken or and interpret what's being asked. That, the LLM models has really shrunk down to virtually nothing. Well, they -- it now comes out of the box with an ability to understand and interpret what's being asked. . What you cannot take away, though, is the integration, as you saw in the lower right-hand corner of the diagram that Jonathan showed is the integration to all the back-office systems that have all your information to answer the actual question, right? You're not going to call up a contact center and ask about the weather, that it can go buying from the Internet. You're going to ask it something specific about your account, your balance, your package or information, it needs to go get that information and speak it back to you. But the data itself was there to help the systems the -- has training data to learn how to understand and interpret spoken language. That part of it now has been shrunk tremendously.

Yes, basically, what you just said is no longer term. That simple. What you thought you knew about training AI model for natural language, just take that, roll it off and fees paper, cost in trash. It is no longer true that you need to have tons and tons of data to train a bespoke model. Now the reason for that is you need a huge amount of data to train the LLM. So thank you, OpenAI, Google and others who are doing it, they swallowed the entirety of the Internet to train these things. And once trained, it doesn't need further training in order to be applied to predict use case. What instead of training, you need to customize it for your particular use case that is done through a process called prompt engineering. And this is something that is going to be entering the Lingua prompt up that is going to come off the tongues of everyone from technologists to everybody, just like model training, everyone knew those words, growing technical, we're all going to be talking about prompt engineering. That's the way you do it. where you instruct the LLM to do something by literally telling it what you wanted to do describing the task in natural language is how you customize it. So that's -- so to do prompt engineering requires data integrations to third-party data sources like the customer database that you have all the context of the customer, you feed that to the LLM when you answer the question.

Right. And from a pricing perspective, right, obviously, you talked about charging 2x for a virtual agent IVA versus a live agency, very consumable for customers you need to understand as more value -- theoretically more value, more transactions are handled by technology versus live people, how do you see the pricing model for contact center, in general, shifting over time, right? Transaction-based, usage-based, what does that mean for the TAM?

Excellent question because what we've word seat, we all visualize a human. We've got to get away from that because we say virtual agents. It's hard -- but then you have other transactions that don't really have a per port or a per transaction basis. So we will absolutely see an evolution of the pricing model to move -- some of it may still be seat-based. But a lot of it will be either permanent based. We also have -- we have usage right now that's based on permitting and then you'll have per transaction based. So we're currently working on that. How do you best modify the pricing model in order to articulate clearly to customers so they can understand how this is laid out. What I will say is from a TAM and from a pricing perspective, we still -- we're back into those models and still receive the same margins and the same uplift to the spend they're placing with, not only Five9, but as an industry, we all recognize it's really value-based, right? The value that we're going to be delivering, we can charge that on a per transaction or per minute or per port. We're trying to figure out what's the easiest for customers to consume.

And the defensibility of our full amount of money we collect is based on the value we provide and also the complexity of what we do. What we do is not like the large language model that -- what we do is the whole platform I just shared with you. And in particular, all these channels that come into a customer, the voice, the chat, the email, whatever, that's like our platform that we can capture those channels. So the only way to even build a voice chat bot or an agent assist application is we do it or someone else does it, they have to sit on top of our platform to do it. So it's like the Apple model, like where if you want to have an app on your phone, you have to go through the Apples by App Store platform. That's the most egregious example, so I don't think that's not a perfect analogy, but it's a platform play that we have, and that allows us to capture and retain value.

All right. I think we're about time. So we'll leave it there. Plenty more discuss. We have a breakout right after this in Jenny B, 1 floor up. So please come by for additional questions. Thank you for joining. Thank you for being here.

Thank you.

Thank you very much.

Thank you, Matt.