Sam Zegas: Welcome to Deepgram’s Voice of the Future podcast, also known as Our Favorite Nerds. At Deepgram, we’re obsessed with voice, and this podcast is our exploration of the exciting emerging world of voice technologies. I’m your host, Sam Zeegis, VP of Operations at Deepgram.
And today, I’ll be talking with Darren Ascone: Escone, CEO at CoverNetworks and Matt Busygan, CIO. Welcome, guys.
Darren Ascone: Thank you. Hey. Thanks, Ryan. No, sir.
Sam Zegas: Awesome. That’s great to have a year. So this is our favorite nerds after all. So why don’t you tell us a bit about yourselves and specifically what kind of nerds are you?
Darren Ascone: So, Matt, if you don’t mind, I’ll go first. I think I would refer to myself as a closet nerd.
You know, the bulk of my life I’ve sort of been associated with athletics and that was a competitive division one wrestler. I’m a pretty experienced triathlete. So at the bulk of my life, I’ve always had this sort of comparison or association with athletics.
But deep down inside, I’m a massive technology fan. I’ve been a technology entrepreneur for over 25 years. I’m a programmer at heart no longer. They don’t allow me to touch code anymore.
And, you know, I’m a huge Star Wars, Marvel, multiverse, you name it. We are into it.
Totally geek out to that kind of stuff. So that’s kinda something that a lot of people don’t know about me.
Sam Zegas: That’s great.
Matthew Busigin: Darren is definitely a huge nerd, and it’s not so closeted.
We’re actually kind of both into everything. You know? There’s there’s a, I think that the thing that most characterizes nerdery in general. We call it nerdery.
And, you know, at some point in the office, there is a phrase, there is nerdery a foot, and that usually means that somebody’s getting into something.
I’m a software engineer.I also have a background in finance.
I did macroeconomic modeling before I got to Hover and I’ve tried to merge those worlds and, you know, the most number of times that I possibly can. But at this point, it’s mostly just software engineering and AI.
That’s why we’re here. Right?
Sam Zegas: Yeah. Well, some other thing that I know about both of you is that I would call you upstate New York nerds in a way because you have built a very passionate regional business up there that is really focused on that local community. So that’s gonna come up, I think, in the course of the conversation today.
Matthew Busigin: That is true. We’re Western New York boosters.
I didn’t, I’m actually not a Western New York native. And Darren’s actually an upstate native?
Darren Ascone: I’m a yeah. I’m a transplant for sure. I did high school in Albany, New York, which is upstate New York but I was born and raised in Staten Island, New York, which is outside the city.
You know, some people always say to us, oh when you tell people, when you travel, you’re from New York, they immediately think New York City. Oh my gosh, it’s so lucky you live in New York City. But, you know, when you say Buffalo, New York, they don’t understand that, you know, we’re a stone’s throw from Canada, and it’s as flat as can be here, and there’s no tall buildings. So yeah. I am a transplant as well, but we are huge, huge, huge proponents of Buffalo, New York.
The bulk of Hover or 30% of Hover Networks’ business is not-for-profit business where we provide telephony messaging, communication services to not for profits. We have been a massive proponent and give where you live since we started the company almost 14 years ago.
So, yes, it is fair to say that we are upstate or Western New York nerds because we do believe very heavily in this community, and this community has been very good to us.
Matthew Busigin: It’s a fabulous business community too. It’s a fabulous little technology community. You don’t really think about that you know, when you think about Buffalo, of course. And, you know, coming from, you know, the West Coast, you know, it probably appears quite quaint. I’m from Toronto.
So when I came to Buffalo, I quite honestly didn’t have a very large expectation for the tech community here.
And the first company that I worked for here was actually the company the last company that Darren started, which was Synacor, and I couldn’t believe how many smart people were here.
And all of those people have actually, you know, that that era of Synacorians has gone on to found, like, the next generation of tech companies in Buffalo and and and, you know, the 43 North Business Plan competition, which is the largest business plan competition in the United States, which is, of course, largely technology focused. So this is actually a wonderful place to live and a wonderful place to do business.
Sam Zegas: Well, you’ve really added yourselves as Upstate and Western New York nerds now because I can just see you light up when you talk about it. You talked a little bit about Hover in that description, but why don’t you go a little bit deeper?
What does Hover Networks do? And how would you describe it to someone who might not be familiar with your space at all?
Darren Ascone: Yeah. We’re a cloud-based telephony provider. You know, most people in our world are gonna know the RingCentrals, the Nextivas, the 8x8s of the world. That is what we do.
We’ve shifted from a long time ago being just a pure telephony provider and only providing essentially you know, dial tone and IVRs and voice mailboxes to more of a communications platform, which has just been, you know, an adaptation and a shift in the way people are doing business now, including pre-pandemic. I mean, we saw the changes happening then post-pandemic. It is certainl –
Matthew Busigin: It’s an accelerant.
Darren Ascone: –accelerated. Yeah. You know, there has been some boy scout juice thrown on the fire for sure to force us into evolving and building the company. But unlike the large big box providers, we are a very boutique concierge type provider.
For years and years and years when we’re pitching potential customers when we’re talking to community members, I always use the analogy if you’re familiar with the American Express credit card in most specifically the Black Card. The Black Card is a very unique card. Only few people get it. And the service you get with that card is unparalleled. So you get you get a full time concierge. If you have a problem, you call the same person at American Express, you’re not dialing into any crazy number and getting a different person every time. They know you. They know your profile. That is the same thing about Hover Networks. So when you call Hover Networks, you know, we’re a 22-person team based in Western New York. We’re in all 50 states and 9 countries, but you get the same team that has sold you, provisioned you, onboarded you, and supported you every single time.
The team, the way we’re set up in our office, is built so that everybody can interact with everybody at all times. So everybody sort of knows about all of the customers, especially during the onboarding process. There’s enough buzz in the way that we provision people that when you call in and have a question people are very, very familiar. And I think, you know, we have worked very, very, very hard to get there.
And our, I think our Google reviews are proof in the pudding, but that’s what really makes us unique. We’re not, you know, we’re not just a phone provider, we’re not just a cloud-based phone provider.
We are a boutique concierge phone service provider, and we really stress the service.
Matthew Busigin: And that really extends to all facets of the service to be provided. It’s not just the onboarding process or the support but we also do quite a bit of boutique concierge, custom software integration. We do quite a bit of work with making our phone system play well with the other pieces of software and services that they have. And, you know, it’s it’s very difficult. I suspect to get the CIO of Nextiva on the phone so that they could help you, you know, connect up your dental CRM.
Sam Zegas: Yeah. It’s an interesting point. It seems as though the niche that you’ve really found a home in is one in which you’re responding to pain points around the responsiveness and the customizability of the solution that you’re you’re putting out. Did your understanding of this pain point evolve over time? I think I mean, imagine it did, but what do you really see as the main pain points for people who were using big box alternatives to you?
Matthew Busigin: Well, they’re cookie cutter. And if it works for you, then that’s great. But if it doesn’t work for you, there’s not a ton of flexibility to you, you’re going to be the one who has to be flexible.
Sam Zegas: Mhmm.
Matthew Busigin: Whereas what we found is that when we’re the ones that can provide the flexibility that that removes the pain when you’re asking the customer to be flexible. That’s a pain point for the customer.
Sam Zegas: Yeah. Makes sense.
Darren Ascone: Yeah. I agree a hundred percent. I mean, I think just the ability for us, you know, one of our core values in the hover networks core values that we review every year as a team is agility.
So there’s no task that when presented to us that we ever just say, it’s never a hard no. I guess that’s what I’m looking to say. When somebody calls us up, we always give it, you know, it’s full attention. We investigate. And if it is something that we can do, you know, honestly is a huge part in our business, especially with the customers we deal with.
If it is something that we can do, whether it’s tomorrow or if it’s gonna be 3 months, we’re honest with the customer. And I think that it’s really, really helped us build these long-term relationships with these customers. And it’s it makes it difficult. You know, our philosophy as an organization and my philosophy as a leader is the more we can do to embed ourselves in the customer into what they’re doing and helping them accomplish their goals, the harder it is for them to leave us long-term.
So, you know, we joke and say the more hooks you can put in them the harder it is to leave. And there’s a lot of truth to that. Once they have spent all of this time getting their off colored or not off colored. That’s not the word I’m looking for.
Non-off-the-shelf dental CRM, you know, some custom application that they have built over years. it’s not just gonna work out of the box with a big box provider. So if we can integrate into that, the likelihood of them ever seeking another solution I mean, at the end of the day, telephony is telephony. Right? Phones or phones are gonna ring. There’s gonna be a voice mailbox. You’re gonna hit some buttons to get to certain destinations. It’s all of the other stuff that most people don’t see. which I think makes a cloud telephony provider valuable.
Matthew Busigin: Yeah. Then there’s the commodity side of things. And what’s in the commoditized side of things and what’s been in the value-add side of things, I think, has, you know, shifts over time. Like, there’s a certain gravity whereby, you know, the the the value-add drips into the commoditized–
Sam Zegas: Sure.
Matthew Busigin: –and I think that what’s happened over time is we’ve seen a lot of that volume, that value-add volume drip into the the the commoditized bucket. And now we’re seeing the like, for the value-add side, we’re seeing a whole brand of new types of features and and and sub-applications. And, of course, that’s usually where Deepgram comes in.
Sam Zegas: Sure. Yeah. Let’s jump in that direction. I’d love to hear what you see in that value-add bucket right now. And what are some of the things that are moving into that more commodified, commoditized category?
Matthew Busigin: So, I mean, Deepgram itself, I think, is probably one of the singular responsible forces for the commoditization of ASR.
Sam Zegas: That’s true.
Matthew Busigin: It is true. Before Deepgram, I think Graham called me when, you know, I was first acquainted with you guys. I think I couldn’t believe how good your pricing was compared to what we were paying for Google to do transcriptions of voicemails, and it was very expensive. And Google is a very difficult company to do business with, you know, from an SMB perspective.
You know, there’s no one you could, you can’t call anybody at Google ever. You can’t email anybody at Google ever. Their support systems are very obtuse and it feels to me like they’re largely designed around suppressing people from reaching employees at Google.
Of course, we immediately go to Deepgram and, you know, we have access to you guys– we have wonderful access to you guys now and your partners and this. And so that’s combined with the democratization of and, you know, via the pricing, of high-quality ASR. I think that that’s been industry-wide trend that’s being driven, of course. I think everybody is now including just transcription as part of their offering, whether it’s an add-on or whether it comes, you know, embedded in the price of a premium user or what have you.
But, of course, the next stage of that after doing transcriptions of, you know, visual voicemail and doing call transcriptions, which is excellent. That’s a lot of value is being able to do analytical work on that. And there’s just like there’s this gold mine of data here that we are, I think, just starting to really begin to tap where, you know, that I think in a few years, this is going to be very much more clear than it is right now.
I think in a few years, this is going to be a scenario where it’s just expected that every call that you make on your corporate phone system is going to have quite a bit of intelligence applied to it that includes ASR and then post-processing afterwards.
Sam Zegas: Yeah. It’s a very interesting set of points you make. We definitely see that trend toward commoditization of basic speech-to-text as, you know, as you would expect over time, many of the different providers who give that sort of service or produce models that are products that you can buy, they are improving, and you would want them to improve. And that’s the way that the competitive evolution of the market is headed.
That said, we’re really right at the beginning of a frontier of so much deep understanding, quote-unquote, insight, you know, the ability for machines to listen to this conversation and say, here’s a quick summarization of what they said that isn’t just a recapitulation of actual sentences that we produced or–
Matthew Busigin: –It has to understand, it’s actually generated with an understand– a semantic understanding of the text. And you can and right now, this is unstructured data.
Darren Ascone: And you think about the efficiencies that this provides, you know, we’ve used the word commoditization a couple times now.
You know, there’s this race to 0 in our business, right, where all of a sudden a premium user for $29.99 back then, a premium user is $17.90 now, and you get more than you got 2 years ago–
Sam Zegas: Right.
Darren Ascone: –for less money. Right?
Darren Ascone: –It’s the race to 0.
Matthew Busigin: It’s the tech commoditization cycle right there in a nutshell.
Darren Ascone: Yep. So from our perspective, from Hover Networks’ perspective, we have always had the goal, and we are striving towards being able to bring what I would consider Fortune 100 feature sets down to small to medium-sized business. And more on the small side small, you know, dental practices locally that did not know if, you know, formally they had to have their afternoon receptionists go and try to listen to recorded calls to try and pick out which ones were good. Is the new person doing a good job? Or are they not doing a good job? the ability for us to transcribe, analyze, and deliver a port that said, “Sally had 4 good calls today and 8 bad ones.”
Now, Is it perfect in today’s world? No. The technology is getting better day by day. But just this sheer fact that let me go listen to the 8 bad ones. And wow, if 7 of them are bad, I feel pretty confident about the model that we’re running right now and that these, you know, we’ve now just saved this small dental office all of this time and made them better as a practice, whether it be by providing more training, having more people sit in on their calls, or whatever it may be.
The ability for us to be able to do this at a price that Western New York, you know, Western New York is not New York City, Western New York, no Buffalo, New York is not Los Angeles. People are extremely price conscious in this in this region. For us to be able to bring big city technology to a smaller town or a smaller metro area has been a huge advantage to us.
Sam Zegas: Yeah. What you mentioned there is sort of call analytics, the ability to listen to recordings, and then provide some sort of quality score, performance score, or coaching tips whether it’s something that happens asynchronously or live even in real-time. Is that where is that a use case you’re really interested in in the future roadmap? Are there other use cases you’re interested in?
Darren Ascone: Matt can go first. I mean, I will just say because I know Matt’s answer’s will be a lot longer than my answer will be, and it’ll be a lot more detail. But the answer is 100% yes. We are headed down that road. We have stuff out there now that we are working with some of our larger companies, our larger clients that we deal with. It is not made available across the board. You know, we’ve got, I don’t know, almost 1200 customers. So, you know, it’s not available to everybody. But it is in production. It is being used today.
And so yeah. The answer is yes. Matt, and I’m sure you wanna expand on that.
Matthew Busigin: This is something I’ve done a little bit of thinking about–
Sam Zegas: Oh dear.
Matthew Busigin: –I think there are several main value propositions that are on our roadmap. So we actually, today, have already done, so what we’ve done is we’ve leveraged large language models to take the output that comes out of Deepgram and then–
So the evolution of language models, you know, obviously, you guys are you know, in the deep learning space. You guys understand this extremely well. But you’re the medium that you’re dealing with is different.
Previously, you know, you would be a few shot or even one shot for the use, you had to provide examples serially and, you know, question answer, question answer, or prompt completion, prompt response, prompt response.
And large language models in the past 8 to 9 months have actually just moved to 0 shot and where you can just ask it to do something. You can just tell it to do something. Ask it what you want. Here here is a transcript. Here are the pieces of information that I want. Oh and by the way, I want that as a JSON data file.
And it will do it for you. We’ll just give it to you. So that’s–
Sam Zegas: Let me actually break in here just in case people who are listening are not familiar with some of these terms.
2 points. 1 you said, working in a different medium. I think you’re talking about us working in the audio-to-text medium. And you’re talking about something that’s a different medium because it is text input and text output. And in that realm, there is this phenomenon that’s is generating a lot of research interest lately about these large language models, these huge corpus models that are able to predict and analyze text in certain ways. Sorry to break in there, but go back go ahead.
Matthew Busigin: Well, and, you know, predict and analyze, like, I think is a bit of an understatement when it comes to their capabilities wher by, you know, again, like, we you can literally give it a complete call transcription and then ask things like, obviously, you can summarize that, you know, that’s something you’ve been able to do, you know, since 2017, 2018 with language models that were in the millions of parameters of size, you know, starting from the BERT models that that that that Google released.
But now we’re able to surface not just summarizations, but actually, we can do transaction tracking, complete transaction tracking, every detail of every transaction. And we’ve run this against one of our customers is in the automotive space, in the automotive auction space. And we are actually able for instance, to listen to have Deepgram do ASR on the call recordings, and then we can take those call recordings. And we can get the list of transactions that were detailed on that call, the state that–they pardon me. The state that they were in the price that was negotiated, whether the transaction had been finalized or not, and we’re just asking the model for it. We’re, like, what we’re not training to do this. We’re not doing language model training. We’re just literally asking GPT 3 or, you know, similarly sized model to to to present us this information.
And so I think that you know, that’s obviously a very idiosyncratic a critic example of this use, but I think that that the future is going to be, you know, obviously, some generalization of that whereby, you know, there’ll be a number of things that you can unearth structured data out of an unstructured data pile.
Sam Zegas: Mhmm.
Matthew Busigin: But I think that different customers are ultimately going to have different specific to their business, different questions that they wanna ask. Like, so one of the questions that we had from one of our largest clients is they wanted to know what their first call resolution rate is.
And you know what?
There are actually very few ways of measuring that and pretty much the only way to reliably measure that is by asking a large language model to read the call and say, was the customer’s, was is the customer, did the customer get a single resolution call in this interaction.
And so we’re able to provide that. And that actually relies a very significant understanding of the text and a very significant understanding of the conversation and the dialogue and there’s clearly these large language models have been trained actually on a very large amount of straight up literature and dialogue.
Sam Zegas: Mhmm.
Matthew Busigin: And so it actually understands the modes of communication that humans do, especially in written form, and is able to come up with if you would have told me that a that a computer could do that, that specific task. Tell me whether there was a first call resolution on this call just by giving me the wave file and, you know, after doing some post-processing, you, you know, obviously, the Deepgram step and then putting it through a large language model that it would be able to tell me whether that caller’s issue was resolved or not. I would have told you not in my lifetime.
Sam Zegas: Yeah. And an interesting way that I think this research is going to evolve is what you’re talking about with this large language model that takes in text and is able to provide you with certain sorts of insights based on all of, you know, this huge set of parameters that it has been trained with.
It is probably understanding certain semantic units that, you know, we don’t have great insight into how it understands those units, but it is sort of saying, you know, like, when you’re talking about analyzing a transaction, there are probably semantic units that it’s looking forward to be able to put that sort of give you that sort of output.
What’s gonna be really interesting is as companies like Deepgram and others are developing more semantic tagging as part of the input layer to that. And as those semantic tags can be more customized to the specific needs of a company or a set of users, there will be more and more inputs for that sort of summarization and insight to take place. And I think that over the next few years, as that ecosystem of data evolves, we’ll see very interesting developments on the, excuse me, on the understanding front.
Matthew Busigin: I’m very, very curious to see so, whether that analysis is done on, you know, the audio side, you know, tagging on the audio side or whether there’s a post where the delineation between, you know, you know, the Deepgram side of that or the language model post-processing side of that. You know, it’s going to be a real interesting thing to watch.
And I’m not convinced that the correct place to do all of that is in the input layer. I’m not completely convinced to that. I think that there’s so much value that you can unearth from, you know, the post-processing side of things that might be a pretty persistent place to want to stick that kind of intelligence gathering mechanisms in. I’m just guessing.
Sam Zegas: No. I do agree. I do agree with that. And when we look at that sort of research. It is a post-processing step. Typically, it’s something that we’ve already you know, we’ve processed the audio, we’ve put together you know, we have a text that we’re working with and then it’s an NLP function, you know, something that is focused on a text input and figuring out how we can create meaningful semantic units.
Like, for example, is this a patient identification number? You know, is this a Social Security number? Here, we’re talking about a case file, things like that, which are very dependent on the specific use case of an individual customer or maybe a specific user as opposed to something that is just like a very general use model, which can certainly provide a lot of very powerful insights, but maybe less honed toward the specific need of the user at the end user.
Darren Ascone: I think one of the most unique things that we have found has been of great interest to us is the ability to know which party is responsible for follow-up. You know, as we’ve talked about, first call resolution, and whether or not they’re achieving that metric or that that KPI that they’re trying to get to as an organization.
For us knowing that which party is supposed to be doing the follow-up is a huge insight because it it allows a manager to sort of look at it from again, we know the science isn’t perfect yet, but if you can look at it from a 30,000-foot level and go, wow, every phone call that we get internally ends in follow-up or results in us having to follow-up or the vast majority of them. Clearly, they need to fix that. Right? Because then now you’re doubling the amount of phone calls. So not only–
Matthew Busigin: Sounds like you’re–
Darren Ascone: –inbound now, you’ve got to follow-up and make an outbound call. That means more bodies internally. So that metric alone just having the model be able to tell us who’s responsible for follow-up has been interesting.
Matthew Busigin: So and that of course can be aggregated and analyzed by large language models for business analysis in of itself. And we’ve done that. We’ve actually done that.
And it’s been correct. It’s and it’s been extremely enlightening.
Darren Ascone: So this may paint some more color to it back to the first question of what kind of nerds we are and who’s the bigger nerd. And I think that the people now know that this might not be me and it might be at least in my window, now the one below me.
Sam Zegas: I actually wanted to move back to a comment you made very early in the conversation about how a significant chunk of your customer base is nonprofit. I’m curious if there’s a difference in the sorts of needs that they have, the sorts of use case patterns that you see over there.
Darren Ascone: So I think that we’ve done well in that. Again, I mentioned earlier, and it is something we’re very proud of. It is part of our core values in giving where we live. So we’re very passionate about being able to provide not only, I’ll say, reduced service fees for or not for profits, but annual monetary giving.
We’re very good about participating, signing up annually into, you know, whatever they’re giving packages for the year, where we’ll commit a certain amount of funds for the year with these people to help their organizations, you know, stay in business and move their organizations forward.
We know, you know, a lot of their funding has been cut over the years with all the things in the world that have gone crazy. We have found to hone in on your exact question. We have found that if you’re not aware, people are very passionate about their telephones working 247/365.
Right? People think that they need the phone on the desk. This thing with a wire on it to work. A lot of people forget they’ve got one of these too. I mean, there are one of these in their hand that also typically work. But if the phone on the desk isn’t working, you know, I’ll even go back 5,7,8 years ago. people would call and scream that the phone doesn’t work in the event that there was some sort of Internet outage, whatever have you.
We have found that not for profits, the have a bigger calling. They’re typically organizations that understand we are doing our best to help them be their best.
Sam Zegas: Mhmm.
Darren Ascone: So early on in our careers, we found that they were great customers because they were compassionate they were caring, and they have a bigger calling. If something went sideways or somebody’s voice mailbox wasn’t set up, they don’t call up in scream and say, Johnny’s voice mailbox wasn’t set up. They say, hey, by the way, we’ve noticed that x y z wasn’t completed. Could you guys take care of that? Oh absolutely. So I think that we found early on in our, you know, in the infancy of our business growth that they made good customers for us early on because they were compassionate when we were knocking the rust off and figuring out how to do things 14 years ago.
And we’ve grown a reputation of being flexible, giving, and understanding to their needs. I mean, as you can imagine, global pandemic hits, their funding gets cut off for whatever reason. Customer with a hundred telephones, we know there’s not a hundred people in the office working they can’t pay for a hundred telephones that month.
Sam Zegas: Right.
Darren Ascone: That’s okay. You know, there was no they were scared to call us and say, listen, we can’t pay our bill, and we said, that’s okay. We didn’t think you were going to. You know, and that happened that wasn’t with every everybody. Some made good on it after the pandemic, and they got their funding, and they said, you know, after the pandemic, you know, whatever that means. You know, a post heart of the pandemic. And other people said, we can give you this much money towards that bill. or we just said, you know what, keep it as a donation.
So I think we just found, you know, you hear Buffalo oftentimes, you know, called the city of good neighbors. We believe in that, you know, that is that we’re known for that. It’s Buffalo’s a big town. It’s not a real city. A lot of people know that. It’s 1 degree of separation almost always. So, you know, no real, I’ll say technological fit for us in terms of not for profits. I think it’s more built around values.
Sam Zegas: Yeah. And the fact that you are willing to be a proactive, an intensive partner probably helps a type of organization that may not have really big, technically empowered team in-house.
Matthew Busigin: Absolutely. And that’s typical of of of our customer base in, you know, whereby we’re certainly providing well, we provide quite a bit of more general IT consulting than we don’t do that. We don’t pay for it. We don’t charge for that. But just by virtue of where we sit, you know, in the orbit of our customers, they’ve seen us as domain experts and we can certainly satisfy quite a bit more of, you know, those demands than they’re going to get from, you know, their Nextiva rep.
But when it comes to nonprofits, like, yeah, the technological side of things are isn’t so demanding except for things like idiosyncratic whereby, you know, for instance, one of our largest nonprofit customer. I think our largest nonprofit customer has hundreds and hundreds of analog cordless phones. You know, so that that that’s kind of a technological problem of on its own in trying to make that particular situation not only work, but also, you know, functionally, but also economically. in, you know, the a modern VoIP context.
Darren Ascone: And I think too, like, we have a large as we’re thinking technology, I’m trying to think if we’ve done special things for people, pro bono, just to help them. We’ve got a large legal not-for-profit where we do post phone call multi-language surveys and store the data for them and deliver it to them so they can understand you know, you know, they do a lot of immigration work and helping people, you know, get situated in the states.
Matthew Busigin: –And legal intake automated attendance.
Darren Ascone: –Yeah. And so we provide that for them, Pro bono. Like, you can’t go to a big box provider and get that off the shelves. you know, and it’s gonna cost enough for profit a lot of money to have an outside development shop do that for them. We have the technical expertise.
We know for a fact they can’t get that anywhere else. I mean, they could, but it would cost an arm and a leg. And, you know, that’s why we love them and they love us.
Sam Zegas: Yeah. Actually, Darren, you’re showing the phone there on camera and Matt, you’re comment about the customer that has a whole lot of cordless hardware phones. It reminded me that you guys recently launched a softphone application. And I’m curious, like, why did you put that on the road map and what value does that bring to users?
Darren Ascone: So that is a great question. So we might have been a little bit late to the game in releasing a softphone, you know, that we used to use have our customers go download off-the-shelf softphones.
Post-pandemic, we have found that customer needs are changing. I think anybody in our industry, anybody in cloud services, UCaaS space is finding that you know, the workplace is changing. More and more people are working from home. More and more people are working off of common area WiFi signals. So we have found that there was a need for a better solution for our customers. So most recently, we have launched our what we call, the Hover Buzz Suite which is a desktop softphone natively built into our application. Also built as an SDK that can be called out and plugged into any of our customers’ applications if they’d like to stick it into their own CRM. That is possible as well.
And we built that because we found that the off-the-shelf softphone products that were available to our customers were not I’m just gonna say as sophisticated or reliable to meet the demands of our customers. Our customers, because we’ve done so much customization for them over the years, didn’t feel that that was scratching that itch. So what we have built is what we consider a best of breed.
Again, I said we came late to the game. I think that probably helped us. We’ve had the ability to go and analyze all of our competition softphone applications. And I think we’ve taken the best from all of them. And then, in addition, we are primarily a Polycom shop. Most of our customers have Polycom handsets on their desktop.
So we have been able to take all of the Polycom feature sets that our customers are used to and transition that onto the desktop. That was important to us for 2 reasons.
Number one customers, we knew we were gonna demand it. Right? They’re so used to having this Polycom with the certain buttons in certain places and certain views.
Number two because of supply chain shortages that I think everybody in any sort of hardware space has dealt with, the specific Polycoms that we like to use that we auto-provision were becoming more and more difficult to get our hands on. At the end of 2021, we made a huge software purchase where we filled our warehouse with as many Polycoms that we could get that we like to put on our customers’ desktops. We knew that that had about a year and a half lifespan in our business that we could fulfill customer orders without having to buy another Polycom.
That being said, we do not want to get caught with our hands behind our back and not be able to get Polycom. So we need we don’t want to be held hostage to a piece of hardware anymore. And decided that if we could solve this with software, which clearly we’ve been able to do, the world is shifting that way, our customer base is shifting that way. And I think over time, we’re gonna start to see, you know, less of these on the desks. I don’t think they’re gonna go away.
You know, I hear that all the time.
I speak to lots of different groups. People always say is the desk phone, the office phone going away. And I think the hard answer is no. I think there are still going to be some people. You know, we still have law firms that have the secretary make the call for the attorney and say, “Mr. Jones, so and so is on the phone for you,” and that person picks up the phone. I don’t think that goes away in 2 years. I don’t think that goes away in 5 years.
I think we see less I think we see hybrid models where people have a couple of desk phones in the office and they have their laptops with their softphones, and they come back and they hot desk into a physical phone if they wanna throw on a headset or what have you. But I don’t think they go away. But I do think that the the softphone application that we have built is, you know, arguably second to none. I know I’m biased, but we’ve spent, you know, almost 2 years now building it. So I am we’re super proud of it, super happy with it, and customer adoption is been unbelievable.
Matthew Busigin: I think the other thing that’s really valuable with moving customers to a softphone model is that we have a lot more control over the customer experience. So the customer experience out of Polycom is extremely rigid and limited in, you know, you know, the you don’t have a lot of degrees of freedom, you know, when when you’re creating a customer experience using any of the the the hardware, whether it’s Polycom or whether it’s Yealink.
Whereas if it’s software, well, software is infinitely mutable and it’s immediately mutable. you can customize it at any time and and to do anything that you would like. And I think that the way that we see the future certainly especially when it comes to real-time transcription is to have more and more information that is being summoned by, for instance, keywords that are being discussed, you know, and as we’re taking, for instance, deep Deepgram streaming ASR, you know, being able to pull up relevant knowledge base.
This is all futurism, of course being able to pull up relevant knowledge base entries, you know, if you’re you’re on the phone with the customer and, you know, they have a particular support in the line of porting query, which is not something which the the the customer agent is particularly well versed in.
Then if there’s a document somewhere on the, you know, the Internet or the knowledge base somewhere. Well, you know, you wanna go back to large language models. You know, we can do things like NLP, semantic, embedding vector search and actually be able to say, oh well, I’ll bet you this would be really useful for this call right now.
Sam Zegas: Yep. Yeah. You’re talking about a category that we see a lot of interest in which we call Live Agent Assist tools. Generally, the ability to what we would do is create a custom model that is able to recognize all of the unique terms and vocabulary that are relevant to a certain customer’s use case. can do that with the right data. We can do that in a very short period of time a week or two.
And then from there to launch an application that is dynamically in real time listening to the conversation and pulling up articles that would help to resolve that thing. It’s a really interesting kind of emerging piece of technology.
Darren Ascone: Yep. It changes the call center overnight. I mean, like and by overnight, I mean, in the coming years, as we’re seeing the call centers up today, they’re gonna be radically different, I believe. And I think we believe that this technology is gonna change the way people do business from a staffing perspective from a turn time perspective. You know, how fast can you get the customer the answer they want?
Matthew Busigin: –And and and resolution as well.
Darren Ascone: From a training perspective, how fast can we onboard a new employee? Well, it’s a lot easier than having two employees, somebody sitting there with the second said on listening to them, writing notes for them, or whispering, you know, using whisper barge technology to whisper to them from another office and say, here’s what you want to tell them.
How about there’s a screen right in front of them on the computer that is coaching them in real time through their customer support call? I mean, it’s amazing.
Sam Zegas: Yeah. It really is enablement and agent assistance is a there’s a huge amount of opportunity in the next few years to really improve the AI models that feed those sorts of products.
Very interesting to me to hear that the switch or the investment in the smartphone, which, congrats. It sounds like it’s going well. It is in part. It’s related to the pandemic, both from a sense of the supply chain limitations, but also from the need to adapt to a hybrid workspace or a workspace where people are working from home some of the time and maybe in an office physical office, another portion of the time. So it’s sign of the times.
Darren Ascone: Sign of the times.
Matthew Busigin: That’s gotta be the the the biggest macro trend in, you know, across the entirety of not just corporate America, but, you know, in fact, the working conditions of, you know, most of the developed world. Every information worker, you know, now pretty much has access to this hybrid approach. And that’s a big challenge for our customers. Our customers are are are that’s probably the number one challenge. that that our customers are experiencing right now is attempting to appropriately accommodate and empower people so that when they go home, they have at least as good a tool set as they have when they are at the office.
Sam Zegas: Yeah. Fascinating. Well, guys, this has been a really interesting chat. Thanks for all the thoughtfulness and thanks for working with Deepgram. We really appreciate you.
Darren Ascone: Very well.
Matthew Busigin: Likewise.
Darren Ascone: This has been awesome.
Sam Zegas: So to all our listeners out there, thanks very much for tuning in today. Come check us out for more information about Deepgram and about Hover Networks that’s hovernetworks.com. And of course, you can find us at deepgram.com and @DeepgramAI on all of our socials. So with that, wear out. Catch you next time.