SoundHound AI, Inc. (SOUN) Earnings Call Transcript & Summary

Well, thanks, everyone, for joining. I have the pleasure of being here today with Keyvan Mohajer and Nitesh Sharan from SoundHound. Keyvan is the founder of the company. And obviously, there's been a lot of interesting things going on in the AI lately. And so we'd love to kind of hear how things have been emerging at SoundHound from starting in 2000. And then Nitesh will talk a little bit about the opportunities you see ahead as the CFO.

So let's start, Keyvan with -- it's the year 2000, you're getting your PhD at Stanford, you're thinking about opportunities. What was the genesis of SoundHound? Where did this all come from? And kind of walk us through that there to today because I think it's helpful for people to understand that.

Yes, thank you for having me. So fun fact, I actually went to VCs in 2004, 20 years ago, and I pitched them that voice AI mass adoption will happen in 20 years. Most of them declined to find us. But -- so 20 years is too long, but the fact that it's happening now is so incredible. So to answer your question, I did my undergrad in engineering. I finished top of my class, but I also started 3 companies when I was 19, 20 and 21. So by the time I graduated from undergrad, I had this itch to be an entrepreneur. And -- but I didn't want to be a certain entrepreneur and start little things. I wanted to make -- be a technical founder to a technology company that would make a big impact in the world and spend decades of my life in it. So with that thinking, I decided to go grad school. So I got my PhD at Stanford in engineering also. And when I started my PhD, I was looking for that, what would be that next big change that I could work on, and I turned to science fiction. What do they have in Star Trek, for example, that we don't have. There are great product ideas in Star Trek, advice for entrepreneurs. But some of them are a little far-fetched, like spaceships that go faster than lights, teleportation devices and holodecks and replicators. But the one that was less obvious was voice AI, like they talked to robots and computers and they had a conversation and asked them to get things done and get information. And I thought this is going to happen for sure, and it's going to happen in my lifetime. And I want it to be a part of that transformation. So with that thinking, I chose my PhD thesis to be in speech recognition and machine learning, then founded SoundHound in a dorm room at Stanford with my co-founders, spent 10 years in Stealth R&D to build the voice AI technology, unveiled it in 2015. And today, we power millions of cars and TVs, and we are in thousands of businesses and hundreds of large enterprise brands. And again, we are so fortunate that we are here.

Yes. And what's so important about voice? And why is it the killer app for information and AI? What is it about voice that made it so appealing to you?

Our business strategy is on top of 2 predictions. One is that speaking is the preferred way to interact with devices. And AI customer service will be as necessary for every business as WiFi and electricity. So you sign up for a website, you sign up for electricity and you sign up for AI customer service. And these are the 2 pillars of our business. In Pillar 1, we power devices like cars and TVs and IoT devices. In Pillar 2, we power AI customer service for businesses. And voice really unlocks the power of generative AI in these 2 pillars, and generative AI unlocks the power of voice also. So it's like the timing is so perfect. And it creates this intersection that is very rare. The intersection of adoption demand and technology readiness, right? So we have the opposite examples where in autonomous driving, the demand is there. People want autonomous driving, but technology is still catching up to that demand. And then you have virtual reality where technology is amazing. I mean, Oculus, I tried it many years ago, and it was amazing. But still, you don't see the adoption, the mass adoption. But voice and conversational AI, you have the intersection with it. Finally, the technology is keeping up to that science fiction promise and people are adopting it.

So what about when we think of other big natural language processing agents, Siri, Alexa? How are you different? And how are -- what's your advantage when you think about that?

So several advantages. One is the core technology. So our core DNA is technology innovation. We have the core technology in-house. There are very few companies that have this stack of technology like your own speech recognition, your own national language understanding and so on. And we've always been, in our opinion, ahead of others. We were the first, for example, back in 2015 to show very complex and compound queries that really delighted our audience, and we continue to innovate. And owning the technology also allows us to make our products better as opposed to being an API user of another platform. Second is data. We -- there are data companies that are now worth tens of -- over $10 billion, but we have more data than probably in the top 5, top 10 companies in terms of data. We've been live in production for almost 20 years. Conversations and voices in kind of millions of users in dozens of languages. And that data is priceless in my opinion. Then I would say the ecosystem of our partners and customers and channel partners and investors that really want us to succeed. We are in millions of devices. We are in thousands of businesses, and those integrations are really hard to achieve. And the last thing I would say is our business strategy. We are in an area of focus that is too hard for new players to catch up. It just takes a long time and effort and resources. And it's too unwise for the big tech to focus on, right? And that gives us another advantage.

And by the way, just to add why -- to your question on like why voice is different. If you just think of human history, tens of thousands of years, we naturally have most increasing -- communicated via voice. You have to learn how to type on a keyboard. You have to learn how to text really fast with your thumb, but a kid learns how to speak very early on. So the technology has now become ubiquitous and pervasive and it's easy to communicate through natural conversations. It just unleashes whole new workloads of what you can get done with the human technology interface.

So Nitesh, when you think about the verticals that you're going after, one of the big ones is restaurants, obviously, quick serve. You guys are constantly coming out with press releases around that. What's the advantage from a financial perspective to the businesses? What's the advantage of using voice AI versus their traditional ways of doing business?

Yes, it's multifold. We've gotten traction over the last couple of years. And at first, just sort of post pandemic, there was labor shortage. They just didn't have -- and even now, there's challenges of getting the resourcing. We had an example of a restaurant that we kicked off first in the Midwest and then we're scaling towards the Southwest. Well, last year, when Taylor Swift was going through, which -- I guess her Eras Tour just ended this week. So she went in St. Louis. And after the concert, there were just hoards of customers that came to White Castle, one of our earliest drive-through partners. And the staff subsequently -- they only had 2 or 3 people that are working there and their AI system they call Julia. They were just raving about how Julia was such a great resource to help take the order, so they could do food preparation or handle other customer service inquiries. So it really was a dependent worker that never gets tired, always on 24/7, you can train it to upsell, and it will never forget to upsell. So just to mitigate labor shortage. There's obviously a cost benefit. Again, you're not -- don't have retraining and so forth. You don't have just the cost of acquisition of new talent. So there's a cost benefit. And ultimately, there's a cost -- sorry, and ultimately, we're shifting and we're seeing more and more data on revenue upsell. So now it's becoming a strategic pillar where we see we have a large major QSR in the chicken sandwich side that's seeing ticket prices uplift 10%, 20%, $20 ticket prices going to $20, $23. And that completely changes the architecture for a restaurant to say, okay, not only is this good consistent acquisition of talent, cost arbitrage, but it's also revenue generating as well. So this is one of those things. It's early days still. We're getting a lot of traction, as you noted, a lot of new customers we're signing up. But this is one of those I think you just don't turn back from. When you get it, there's a little bit of upfront work, but then it scales very well. So that's what we are in.

And you talk about the large Torchy's tacos or White Castle, a lot of these bigger deployments. What about the smaller kind of restaurants that are 2,3,4? Can they deploy your solution? Or is it only something you can do at scale?

Yes. So we actually -- we have built AI solutions for -- you can be a single location plumber and barber shop and can have your own AI customer service in a matter of minutes, we call that smart answering and self-service. So really, you go sign up and you enter your website your phone number, answer a few questions, and you can customize it as much as you want. And it learns as people use it. And we really believe in that's like $100 billion TAM that is not addressed. And then we go all the way to the top 50 enterprise brands that if you automate even a small percentage of their volume, the saving is in the millions. So we are actually in the broad.

One of the ways we early on got traction was do integration with point-of-sales partners. We integrated with Square,Toast, Olo, Oracle MICROS Simphony. And so that allowed us to sort of penetrate into those smaller restaurant franchises at least. When you get to larger QSRs, they oftentimes have custom point-of-sales system. So we have our own gateway to kind of integrate quickly.

Got it. And as we think about the transition from pure voice processing into what -- where you went with Amelia, can you talk a little bit about that and how that extended and expanded your breadth?

Sure. Yes. And maybe it's a good moment to kind of just talk through our business model and architecture. So as Keyvan noted, we voice power products like cars, TVs, IoT devices, and that's a royalty business for us. And as Keyvan mentioned, we're in millions and millions of cars. We've seen a lot of opportunity to the second key assumption, Keyvan had about, voice enabling, customer service. And so restaurants [indiscernible], but we absolutely saw this as similar to how Amazon entered with books, but they had a long-term vision of e-commerce. For us, it was a very similar vein of we wanted to enter with restaurants. It suited very well for the sort of feature and facets of our technology where you need precision. You don't just order any old pizza, you want pepperoni and sausage. And our technology worked for that type of precision. So restaurants made a lot of sense as the [ first bore, ] but we absolutely saw it as a disruption across all customer service. And Amelia allowed us to accelerate that. So to have now enterprise customers, deep money center banks and yes, banking, financial services, health care, insurance, retail, hospitality, they had those customer bases. And there was a real complement on the technology side where a lot of those customers, as we are diligent saying, really wanted voice AI as the next horizon. And so that complement has been really beneficial. So yes, we've now extended into an array of different industries, and we're just sort of again at the precipice of massive disruption that we're driving here.

And when you talk about your presence in automotive and you talk about your presence in restaurants as an example, is there a marriage between those 2? Is there an opportunity? And then what does that look like?

Yes. So that's our third pillar, I'd say. So in Pillar 1, we power devices like cars and TVs and IoT devices. In Pillar 2, we power AI customer service. And then we had this vision years ago that we should connect them, right? So if you're driving your car on the road and you're already talking to your car to control the air conditioning and radio and navigation and then you want to order food at a drive-through that we also power. Why do you actually have to go in the drive-through and go in line and wait to get to the AI kiosk? You can talk to your car. And this agent can talk to that agent. So while you're talking to your car, 20 minutes before you get there, you can discover restaurants that take order and you can place an order. And you finally have the scale to make this happen. The vision we've communicated for years, and it was a matter of being in having enough coverage brands nationally and also enough cars and TVs. And we finally have that. And that we expect will generate the third pillar of our business that will also create a flywheel effect because, first of all, it creates value for all parties involved, right? So the most important one is the driver, right, the convenience of ordering in advance. And beyond just food, you can make appointments, make reservations, buy groceries and so on. The merchants like restaurants get new leads, and we monetize this lead generation moments and then we share that revenue with the carmakers. So they also generate revenue on a recurring basis in most cases for the first time. So that should create a flywheel effect where more merchants will sign up because of the lead generation opportunity. More devices will use us because we monetize for them. And more users will use it because there's more that they can do.

And has that already come to fruition? Or is that still in its initial stages as you think about the lead gen model and all that?

We actually -- we have announced that we will unveiled how it works actually with a live demo at CES this year in January.

So right around the corner. That's great. And then Polaris, the -- your large language model, can you talk about that a little bit? Or how we should think about that in the broader? I think you announced it in Q3, just how that relates to the business and where that's going?

So SoundHound is a technology company. We build a lot of tech in-house. We use technology innovation outside of our 4 walls, but there's a lot of innovation that happens in-house. And we have a lot of data. We have a lot of know-how. So Polaris is our own foundation model that is multimodal and multilingual. So audio input, text input, audio output, text output, visual output in some cases. And we are already seeing massive beat of other models in terms of accuracy and speed and cost. And we have gone live now in 1/3 of our restaurant AI agents. And it's markedly better than what we had used before. And it gives us opportunity to do a lot more. One key difference I wanted to explained about, for example, what SoundHound does and what companies like OpenAI do. OpenAI is a great company and we use some other models, and they've really pioneered. But they -- when they put something out there, it can be amazing 70% of the time, and it could not -- it could pay out maybe 30% of the time. And that's okay for their audience because their audience are there to see a glimpse of the future, right? So when it doesn't work, it's okay. It's going to get fixed eventually, right? But our audience is not forgiving, right? So you can't be the AI customer service of a large enterprise brand and be right 70% of the time and hallucinate or really bad outcome 30% of the time. And I think a lot of other players are suffering because of that because they are trying to -- they see this opportunity of AI customer service as one of the biggest opportunities of generative AI. And you can't -- it's hard to go from those demos to actually go live. But that's something we do really well. We can reduce hallucinations by orders of magnitude, make them negligible. Having our own models, our own data and decades of experience really helps us.

Got it. And just out of curiosity, are we thinking of this as a domestic business? Or is it a multi-language capability here? And how do you -- how are you growing internationally?

We are -- we support dozens of languages. We are already live with customers mostly in Pillar 1 in dozens of languages. We power cars, for example, Stellantis brands in Europe. We are in India in multiple languages. We are in Asia. But Pillar 2, a big concentration of that is U.S., but you can see the TAM will increase as we go to other languages.

And a lot of the -- I imagine a lot of the restaurant is U.S. because the drive-through availability in commonplace...

We are actually live in 3 continents in drive-through. We are live with customers with -- we are working with 30% of the top 20 QSR brands. A lot of them are live already. Some of them are live, but quietly live just to test and pilot. But we are in 3 of the top 4 pizza, we're expanding with Chipotle, Casey's, just announced Church's Chicken, Torchy's taco. And we are over 20 car brands and over 200 enterprise brands, including 7 of the top 10 financial institutions.

So -- and Nitesh, you think about the financials are starting to catch up with the opportunity. What are you most excited about from -- what you could say about kind of the opportunity ahead? And is it the monetization that could come off of the lead generation business? What is it that you think about?

Yes. I mean I'll start to extend off the last question, the TAM is massive. We talk about $140 billion TAM, but that might be understating it in terms of the ultimate commerce opportunities that we can address. And even though it is a global opportunity, if you look at just the U.S. and just take drive-throughs, although we're doing phone ordering with several Chipotle, Jersey Mike's, et cetera. But just on the drive-through, there's nearly 300,000 drive-throughs. And by the way, strategically for restaurants is to have more convenience and pick up and go. So that's actually a growing part of that sector. If you do the math on scale, times, price, you'd get $1 billion revenue plus per year easily for us. And we have a running start competitively. We do believe there are some other players. There's some legacy players that we don't believe have the tech to compete. And then there's some other larger players that are trying to come in, but it's just hard. So we have a real strong running start deep competitive moat in a massive TAM. So just that one vector is a massive opportunity. We're small or early stage public company. We want to catalyze and do successful things and then go to the next successful thing. But in the grand scheme of things, I'll go back to an earlier point, like I think I'll start with the vision, right? We do believe voice is natural conversations will unlock completely new pathways of how humans interact with technology. This is the next major horizon. This is the GenAI era. Natural language conversation is a primary interface and we believe voice AI is a killer app. But we do believe in customer service, in particular, going beyond restaurants into financial services, health care. Think about the accessibility opportunities for pervasiveness of a personal tutor or -- so this broader, what we're calling Agentic AI revolution is absolutely here. So we're super bullish for us. We need to be limited and sort of focused on the key things that we can address in the near term and then move on to the next. So -- I mean, that's what we've been trying to do. And we know the opportunity is so massive. It's growing so fast. We have to be agile, aggressive and thoughtful. So we've been doing this organically. We've been growing historically over 50% CAGR consistently, and we've amplified that with some M&A as well this year, which we allow -- we think allows us to accelerate that customer journey and customer adoption.

There's -- I read an interesting article about the use of AI in training. Can you talk about this a little bit, especially with training people of seasonal employees, things like that because it seems to be an interesting little facet of what's going on inside of Amelia where people are really getting a lot of value out of that? So that would be interesting if you could talk about that a little bit.

Yes, I'll actually touch on it a little bit on one of the products, I think we briefly mentioned, employee assist, which actually there's a similar thing in Amelia, they call Amelia answers. And think of this more broadly is like the ingestion of any content that you can quickly train and then be either a very helpful assistant or even better, ultimately more autonomous. But one application, I'll go to the restaurant application, again, where you can ingest an employee manual and train an employee how to fix a machine or clean a machine or make a beverage or all these types of things. We have a similar application where you can ingest an operating manual into the automobile. So now -- and the distinction between us and big tech is we're integrated with the mechanics and the electrical system and the automotive. And now if you have a light that shows up on a display or you have a flat tire or something, you can just communicate with your car to get that advice rather having to thumb through your glove compartment to find the page on the manual or to call into service. So these types of complementary capabilities, whether it's in training, onboarding, yes, I mean, just even for a user to be able to do all sorts of capabilities, the technology is there, and we're seeing more and more use cases.

And how close are you to the end customer gave on -- do you have a regular conversation with your customer and hear from their customers' customers? Or how are you working that feedback loop? And how are you guys adjusting the technology to be better for the customer experience?

Yes. So it's a wide range. So for small businesses, there's a lot of it is automated, and it doesn't learn from interactions. And then for our automotive partners, we -- there are constant surveys that we do have visibility into the interactions, and we can automatically detect if something is -- can be improved and we make those improvements. That's how our models have become better over time. But we also run surveys and get the results and react to it and share them with our customers and so on.

And you talk a lot about accuracy being important. What is the speed dynamic like? So I spent some time with TikTok looking at a bunch of White Castle drive-throughs, which is you guys are TikToks famous. And looking through some of that and watching the experience and the time it takes, is it meaningfully quicker? Is it about the same as human? What is the -- how do we think about that?

So speed is one of the items we pitch to our customers that the throughput is going to go up. And we have the like side-by-side demos, I think people -- others have created. One of the technologies we made was dynamic interaction. We pioneered that concept, where these concepts of turn taking, it doesn't always have to be like that when you talk to an AI, right? So for example, like with some of the brands like Alexa, for example, Alexa, you wait for it to acknowledge you and then you ask how is the weather, you wait for it to respond. And then you're going to follow up. You have to say Alexa again and how about tomorrow. And that's turn taking, it's slow and tedious and not natural. And our first version of drive-through was similar to that. So we -- the driver had to ask a question and we have to acknowledge what they said and we have to show them what we think they said and then they had to acknowledge it and they have to wait for each other and sometimes we interrupted each other. So we came up with solutions that called dynamic interaction. And first of all, the impact of that, I think, is almost like Apple Multitouch. So there was touch screen before and then Apple did this multitouch that just made everything else obsolete. That's how I think about dynamic interaction where there is no turn taking anymore. You just talk. You can talk to the person next to you, you talk to the AI, the AI listens to everything and decides where you're -- when you're talking to it and when it needs to take action. And when it takes action, sometimes it just updates the screen. Without speaking back to you, sometimes it speaks back to you. It just tries to be smart about making you more efficient and natural and that is dynamic interaction. There are really good demos online that you can watch.

And you can [indiscernible] to get you sharpen drive-through, we've seen data now with some of our partners where 72 seconds sort of typical orders now going down below 60 seconds. So from the throughput point on speed, we're already seeing commercial benefits. And then you can think of the visual point, Keyvan just raised. Instead of having to say, would you like to add some dessert with that, you can just flash up like a Sunday and just human behavior, you're like, oh, yes, maybe I'll get that Sunday. And it doesn't take any increased time, and we are seeing some early signs of ticket price increases, like I mentioned earlier.

That's awesome. And when you think about what's going to happen in the next 10 years, right, so we won't go 20 years, we'll go 10 years, what does SoundHound look like in 10 years? How is it being used?

If you got any of our products, again, I always ask myself what will happen for sure. right? So let's talk about drive-thru. You think like 3 years, you will go to a drive-thru and you will talk to that busy distracted tired person who's also making your food and doesn't want to talk to you? Like do you really think that's going to happen? So -- and when was the last time you actually had that experience, and it was so good that I'm going to go write a Yelp review for this restaurant because I just had a really great customer experience, right? So that kind of interaction will be automated by AI in hopefully, a lot faster than 3 years. But I think in 3 years, it's all going to be automated. And we like to be in places where the puck is going. And that's one of the areas I really believe in. I think the same thing about AI customer service. In the past, when you called a number -- a customer service number and you've got an automated system, we got frustrated, we got frustrated and press 0 to get to talk to a human and you just immediately want to switch, that's going to be -- that's going to have [indiscernible] where you actually prefer to talk to the AI, right? So first of all, there's no wait time for AI, right? So you don't get put on hold for 45 minutes. It always picks up the phone, AI is going to be more knowledgeable, more patient, more polite in many cases, and can do a lot more things. So I think every business is going to be AI. Again, AI for every business is going to be like WiFi and electricity. And we think SoundHound is going to be one of the dominant leaders or prominent leaders of that.

Fantastic. We have a couple of minutes. I wanted to open up to any questions if anybody had any questions for Keyvan or Nitesh.

You mentioned that the data you ingested -- I'm sorry, I'm just trying to clarify like the data that you're ingesting through the QSRs are your proprietary data or it becomes the QSR's proprietary data? How does that data gets kind of stored and translated?

So every deal is different, but we -- our strategy is to be friendly to all parts involved, while we extract the reasonable value from that data. So my opinion is that data is the end user's data. Like if you speak to a device, your voice is your data, right? But these products get a license to use your data, first of all, to serve you for what you just asked for, and then use that data to improve the models in a generic way to serve you and your others better. And that's how we've structured it and it works well. And we give a lot of control to our customers. So if you want to delete your data, if you want to let your users delete your data, if you want to delete the data by default right away, all of that is available. But because we are in so many devices and so many customers, the amount of data that we get is massive. And we are able to use it to improve our models. So we get voices, for example, from millions of users in noisy environment in lots of languages. One example is actually is very clear in my mind is the audio you get at a drive-thru is so noisy, right? Next to the freeway, the cars are passing, it's raining and just the audio quality even for human is hard to understand what -- but that's where AI actually is better than human. So our models -- because we get that data, we learn from it, we can actually handle data like that.

And I'd add on that I think this is one of our differentiating factors and even compared to the big tech because we are brand-centric. So we customize data, we customize privacy, we customize security. So we can do things that are sort of flexible and in service of the brand and their end customers. And that's just increasingly important as we move forward.

Great. Any other questions?

Is there any duration associated with the backlog [indiscernible] provided?

So we've -- I've said on the last call, generally about 6 to 7 years in range. And depending on the industry, the OEM, the auto side tends to run that -- some of our contracts with -- on the restaurant side, they actually are sometimes month-to-month, sometimes 1 year, sometimes 3 years. So it really depends on the industry. When you get into enterprise financial services and health care, those tend to be multiyear contracts.

Is the backlog contracted like they have to pay it?

There's a mix. So like generally, the backlog metric we convey is like signed customer contracts. So it's like a total contract value. But depending on the construct of the customer, sometimes there's projected volumes. So we often have like upfront work that's part of it. We might have even received the cash that just have to amortize for revenue recognition rules. There could be minimum commitments where no matter what, yes, it's bulletproof, we'll get the collection. But sometimes, there's also volume projections based in there as well.

And obviously, those volume projections don't happen?

We would -- yes, we would adjust it.

Well, great. I think we're out of time here. We've got one more. Go ahead. Yes.

[indiscernible] question. In terms of your IP, do you license that to big tech today? Or do you have plans to do that in the future? By big tech, I mean, companies like Meta, Amazon, Google, Apple, that sort of thing.

Well, yes, we have patents and we have IP that can be licensed. And in our history, we have had licensing of our technology to big tech players over time. So that has been in our history, whether it's something that we have -- we're able to announce specific names, probably not at the moment.

Great. Keyvan, Nitesh, thank you so much for being here. It's been an honor. And thanks, everybody, for attending.

Thank you.

Thank you.