All Podcasts

VOTF Episode 6: Purge the Pleasantries! How AI is Summarizing the Important Stuff.

Host and Guests


Sam Zegas

Sam Zegas is life-long language aficionado, with years of study in linguistics and foreign languages, and now he can add “podcast host” to his resume. He holds an MBA from Harvard Business School and an MPP from the Harvard Kennedy School of Government.


Josh Schachter, CEO & Founder, UpdateAI

Josh Schachter is the Founder and CEO of UpdateAI, a conversational intelligence on the mission to ensure that every meeting SaaS teams have with their customers is valuable. Josh is a serial entrepreneur who most recently built corporate venture-backed startups with Boston Consulting Group’s Digital Ventures division. He has an MBA from UCLA Anderson School of Management and an engineering degree from Columbia University in New York.

Sergii Kondratiuk

Sergii Kondratiuk, Head of Data Science, UpdateAI

Sergii, a PhD in Data Science, from Kyiv, Ukraine. Worked in Machine Learning and Natural Language Processing projects for 10 years. Passionate about delivering value to users with scientific approach.

Training AI to understand what matters to humans engaged in conversation might be one of the coolest things companies are doing when using ASR to power speech understanding. And now, the promise of a robotic secretary that allows you to focus on your counterpart while it does the distracting tasks of notetaking and summarization is not that far off! Tune in to hear how innovative companies like are helping to cut down your to-do list by removing the need to create a to-do list.

Read Full Transcript

Sam Zegas: Welcome to Deepgram’s Voice of the Future Podcast, AKA, ‘Our Favorite Nerds’.

At Deepgram, we’re obsessed with voice, and this podcast is our exploration of the exciting emerging world of voice technology.

I’m your host today, Sam Zegas, VP of Operations at Deepgram. Our guest is Josh Schachter, the Founder and CEO of UpdateAI, and Sergii Kondratiuk, Head of Data Science. Josh and Sergei welcome to the show.

Josh Schachter: Thanks so much for having us, Sam. 

Sergii Kondratiuk: Hello.

Sam Zegas: Glad you’re here. So this is ‘Our Favorite Nerds’ podcast after all, and we always start with a question that helps you introduce yourself to our listeners a little bit. Why don’t you go ahead and tell us a bit about what kind of nerds you are?

Josh Schachter: I guess I’ll kick us off. So, I got into technology through product management and my first role in product was at eHarmony, the dating site if you remember back in the day when it was, you know, one of the top sites. Now I don’t know if it’s still around even. 

But when I arrived in 2011, on my team, I was the only one that had a BlackBerry, and everybody had iPhones back then. And I just remember being completely mocked from the entire organization that I was the only one with Blackberry. 

So I think it took a month and then I had to run out to the Apple store and finally, you know, join the cult to get up with the technology. But that’s a memory of within the tech nerd world, I was considered a nerd.

Sergii Kondratiuk: That’s great. No. Okay. Sounds good.

I guess I’m more nerdy in more technical stuff, typical nerd stuff, like customizing hardware and software. That’s why I prefer Android because it’s so customizable, and I’m using it from the earlier versions like 2-point something.

And also, I’m nerdy about scientific stuff. So I have a PhD in data science and my thesis is about gesture detection, specifically for Ukranian Sign Language for deaf people. So that’s kind of my hobby and my nerdy stuff.

Sam Zegas: That’s really interesting. When you say Ukranian Sign Language, I guess that’s different than, for example, American Sign Language which some people might be familiar with.

Sergii Kondratiuk: Yeah. By the way, most of the people, they think that there is some universal sign language, but, actually, every country that have their own language, they also have their own sign language. So there is American Sign Language, Ukranian, French, and so on.

Sam Zegas: Yeah. As an interesting aside, from myself as someone with a background in linguistics, the relationships between sign languages that exist in the world are actually pretty complicated. 

For example, American Sign Language is closer to French Sign Language than it is to British Sign Language, which has just created some weird overlaps there. But anyway, that’s a better digression. 

Josh Schachter: Do you want me to leave the room? I’ll let you guys here. We can just have, like, a sign language– 

Sam Zegas: I couldn’t help myself. 

Josh Schachter: I didn’t realize I was in the presence of such sign language knowledge. This is great. I’m learning something.

Sam Zegas: Let’s get back around to UpdateAI. So for our listeners who are not familiar with you, why don’t you give us a little bit of a synopsis of what you guys do?

Josh Schachter: Absolutely. 

So we are a platform for customer success, and we are eliminating busy work that CSMs have when they’re on calls on Zoom with their customers.

So what exactly does that mean? Well, Sergei and the team here have developed our own data science that detects action items in real time in conversations on Zoom. 

We are Zoom apps platform, and we served you your items on a silver platter right after your call so you can follow up with everything that your customer asked you for.

Sam Zegas: Really interesting. So it’s listening into calls. it’s understanding the contents of calls and then it’s recommending next steps or it’s trying to summarize notes or what exactly happens? What does the user see after a call?

Josh Schachter: Yeah. So, you know, you invite our recorder joins the room just like lots of applications have recorded these days and in real time through Deepgram’s transcription, we are transcribing. We’re capturing the audio.

We are doing it through multiple channels and streams so that it’s higher quality. And that’s why it was important for us to use Deepgram. We’re transcribing that audio in real-time. And then the data models that Sergii and the team have built are layered on top and they’re identifying the action items. If, you know, let’s say, it’s “I’ll schedule a call with you next week,” or “let’s follow up tomorrow, I’ll send you the contract.” 

Those types of things that, you know, happen all the time in meetings that we have with customers and we wanna make sure they never slip through the cracks. So we identify those in real-time. 

And then as soon as your meeting closes, and I mean, instantly we deliver to you through Slack and email a list of all the action items that were captured.

And we can point you back to the recap of the conversation, so you can see everything in context. And some other cool stuff within that other data science that Sergey and the team have built as well.

Sam Zegas: Interesting. So I think I got it. It’s about time-saving. It’s about efficiency. It’s probably also about reliability or accountability. You’ve got this record of things that need to come as a result of a conversation that was just completed.

Josh Schachter: Yeah. I mean, let’s, if I wanna extract it and make it, like, our you know, the the the mission at that level, what it’s about, really, is building trust and building relationships between companies, especially in the SaaS world and customers.

And in a conversation that you’re having with a customer, how do you, what are some of the ways that you build trust, and these are not exhaustive, but two of the ways that you build trust are by staying focused and keeping your eye contact with them and being present.

And the other way is, like you said, being accountable for any next steps or follow-ups that come out of that conversation, and being transparent too in that accountability. And so those are really the central tenants that upon which we built UpdateAI. 

The Zoom application that we have is, we chose Zoom as our launch partner because we were able to be fully integrated into the application, make it really easy so that we’re not intrusive, you know, during the conversation.

It’s behind the scenes. It’s embedded. And then we wanna make sure that immediately you can return to your customer, the action items so that, like you said, for sure, ability. 

Sam Zegas: So the pain points that you’re trying to address, I imagine, like, CS teams are overwhelmed or, I guess, part of my question here is, why focus on customer success as opposed to a different part of the value chain? 

Josh Schachter: Yeah. It’s a great question. We could have focused on other parts and we will eventually. But when we came across customer success, we just felt this energy and this buzz about this function that is growing so quickly. And yet, at the same time, it’s underserved.

You know, we recently embarked in an outbound campaign, an email campaign to see CS leaders. And we tested different subject lines and the subject line that got the highest open rate for our emails was ‘Why do sales teams have all the toys?’

So we knew we hit a chord there. And, you know, when we first started, when we incorporated UpdateAI and really learned about what we were gonna be, we did a lot of design thinking hundreds literally of interviews with customer success leaders and others.

And when we came across customer success, this is a group that is so sacred to the health of any SaaS organization. 

Sam Zegas: Mhmm. 

Josh Schachter: And in that sacredness it’s measurable too. Right? You talk about it, what CS does measuring against net retention and those types of metrics. So we said, okay, great. Here’s a group that’s growing quickly. 

You can measure their success, and it’s important in the organization, and yet they are underserved. and they have this burden of back-to-back calls with customers and lots of multi-asking, which is fatiguing them and actually causing attrition while the industry is trying to grow. 

So we felt like that was the perfect beachhead for us to enter.

And then, you know, we’ll grow from there. Yeah. That really resonates with me. Deepgram could not do what it does without a really strong customer success team. So if you’re listening guys, kudos to the Deepgram success team. And they do have a strong team. And I wanna give a shout-out to my CSM at Deepgram, which is Marshall. 

He’s been with us since the very beginning, and he’s been awesome. He’s truly been such a rock of our relationship. 

Sam Zegas: I’m glad to hear it. Marshall’s great. 

Sergii, I’ve got a question for you here. So what is difficult about what you’re doing from a data science perspective? You know, you’re processing language, you’re trying to turn that into human-readable outputs. Why is this hard from a data perspective? 

Sergii Kondratiuk: Yeah. So it’s I think it’s it’s hard from first of all to give a proper definition what is an action item. 

If you ask, like, 10 different people, they will probably give you a fuzzy, like, concept, what is an action item? But if you ask them to show specific examples, they would probably give you different a different subset– 

Sam Zegas: Mhmm. 

Sergii Kondratiuk: –though overlapping mostly, but they will have some different understanding what is in specifics, what is an action item.

So the hard part here is to collect relevant data for us. So we’re in the csm domain. So this narrows the error in which we try to detect those action items for now. 

But then when we were really rigorous in building a taxonomy, a proper, a very deep understanding of what is an action item and what makes any sentence a specific sentence an action item.

We also have linguists on the team. So they really enhance the data science part. We’re not just trying to blindly train some ML models because that will probably not work in the domain, especially with the limited data amount but we are combining ML models with a taxonomy like what is an action item. 

And that was really hard to build a taxonomy from scratch, basically.

Sam Zegas: So once you have the taxonomy, are you using machine learning models to apply labels and then generate outputs that way? Or is there a different approach?

Sergii Kondratiuk: So we build it taxonomy because we, when we label our data first to train m ML models afterwards, we need to label them consistently so that the model won’t be confused in future. 

And that’s really important because if you feed inconsistent labels, then the model will be, it will be harder for the model to define those hidden patterns like what is an action item. So that’s why we were really serious about building the taxonomy.

Sam Zegas: Yeah. That’s so interesting. You know, one of the macro trends that we are constantly watching at Deepgram is the fact that speech technology is really moving from a world where a lot of the research is being done turning speech into text and is now really at the beginning of doing deep speech understanding. So understanding intent, being able to summarize what it is that a person wanted when they were speaking

And we have decades of research ahead of us to really make those things mature, but it’s really cool and exciting to see you taking those steps as it relates to, you know, action items and intents and calls.

Yeah. So I’m curious, right now, is this entirely text-based or is there some sort of audio component to the way that you evaluate the statements that people are making? 

Josh Schachter: You know, I’ll hand it over to Sergii in a second. it’s a great question. And I get that question a lot. And when you get a question a lot, maybe that’s an indicator for you to put something on your roadmap.

So that’s, so actually, it’s funny because I just recently, you know, poked Sergey about that. And I said, okay let’s put that on the roadmap. Right now, we’re based solely on the text.

I think tone is interesting to us, but foundationally, I think the text was more important. That’s where you’re gonna get more of the more precise signals.

But in the future, yes, it’s something that we’re looking into. 

Sergii Kondratiuk: Yeah, so as Josh already told the text is the basis for us.

We’re looking into I guess we’ll do this research on how important is the tone and emotion in the voice. And I assume that this could be a good additional feature, but still the core, the base is the text that’s what we are working on right now.

Sam Zegas: Yeah. It seems to me, like, your point is well taken that when you’re just dealing with trying to find a concrete action item that someone has stated when they’re speaking that the text itself is the most reliable input for that. It seems like down the road. And I say this not just in your context. 

This is also something that we at Deepgram think about a lot is how can we, in an ever more sophisticated way, incorporate tone of voice into our product so that we can catch things like sarcasm and really understand the difference between a sentence that is said and happy wavers something that’s said in a sarcastic and disgruntled way, and whether that changes the meaning ultimately. And that’s just a really interesting problem space that I think is gonna come to fruition more and more over the next few years. 

Josh Schachter: I think so. I mean, I totally agree. My last role was a company called BCG Digital Ventures, Boston Consulting Group. And– 

Sam Zegas: Mhmm. 

Josh Schachter: –we had an internal project where we’re working on detecting emotion for customer or service centers, you know, and for healthcare. And I think there’s a lot of people working on that problem. 

I think it’s gonna be really tough. But you know, like I mentioned, if all of the questions that I get from folks about, like, it is it based on the tone, if that’s any indication of desire then, yeah, I think it’s a huge opportunity.

Sam Zegas: Yeah. 

It is a huge opportunity because fundamentally the mission that you’re talking about is using speech technology as a way to build more natural feeling relationships with customers where the CSM can be more focused and attentive because they’re not focused on writing notes scribbling down things that are next steps. 

They’re really just listening and engaging and– 

Josh Schachter: Yeah. 

Sam Zegas: –there’s a ton of value in that, especially as we see how important CS already is and is becoming more and more important to the way their business is run. 

Josh Schachter: Yeah. So I think a huge opportunity is to help whether it’s customer success managers or sales representatives, anybody that’s on the call, back0to-back calls all day with the customer, to stay more present and more focused. 

We pride ourselves in having what we consider to be the best action item detection in the planet. 

We’ve actually done studies against others that have and its demonstrated that. 

Sam Zegas: Mhmm. 

Josh Schacter: But we know that even the best AI-based detection, NLP-based is not gonna be, you know, as good as human grade, oftentimes.

So our core focus is on helping to give CSMs and others a backstop just that that safety net of, you know, they can stay present in the conversation. They can chill out a little bit with you know, the note-taking because we are there in the background to make sure that nothing slips through the cracks.

And that subtle value prop right there is actually not so subtle. Right? Because that type of stress that you endure little by little throughout the day, throughout meetings, adds up. And so, you know, I totally agree with your point. Just keeping folks present, relaxed taking a deep breath in their conversations is so important.

Sam Zegas: Mhmm. Totally. Yeah. So you said that you focus on CS today and you’re moving initially into a space in which sales has all the toys. You’re trying to level the playing field there a little bit. I’m curious where you think things are headed, that could either be you at your own company or also it could be things that you’re interested in in the market over the next 5 years that you’re interested in watching.

Josh Schachter: Yeah. For UpdateAI, I think we’ll find our place. We’ve really been content with the community that we’ve immersed ourselves in within customer success. And that was also a reason to start with customer success is that, I think it’s just defacto of the nature of customer success, it’s all about building relationships. So go figure that the customer success community is so welcoming. So they’ve really welcomed us with open arms, which has been fantastic.

And we’re here to stay. We’re not just gonna, you know, build an abandon. 

We really want to we are really on a meaningful mission for CS. And these longer tail customer relationships. 

And you can imagine that branches then out tangentially to you know, more of the traditional account management and agencies and that sort of thing. 

I think what I hear and this isn’t my own belief system. I don’t think I have enough education, quite frankly, in the world of this to know. But I hear from a lot of folks in sales that believe that there’s gonna be this marriage of customer success and sales and that you’re gonna see more of these, like, full stack sales managers– 

Sam Zegas: Mhmm. 

Josh Schachter: –That are not just there for the upfront, but they manage their relationship as you go. And so we’ll see how that evolves.

Certainly, we would love that expansion or that marriage of those markets, it would allow us to expand into it as well.

Sam Zegas: That resonates really strongly with me, and it’s something that we see at Deepgram too is that that the tighter we build that connection between what happens presale and post-sale, the happier teams are, the more effectively we’re able to listen to what customers need over the course of their whole lifecycle with us and really empowering the CS team to be a productive part of that ecosystem is really critical.

Josh Schachter: Yeah. And you hear all the time from customer success leaders and managers about the importance of the sales to CS handover. 

Sam Zegas: Mhmm. 

Josh Schachter: And it’s just it’s so critical. There, there’s no room to drop the baton. And so I do think that is probably a direction that you’ll start to see things go more in.

Sam Zegas: Yeah. Two themes really jump out at me listening to you, talk about what you’re working on.

The first one is just that customer service, even provided by humans, becomes more humanized when they’re supported by better technology. And I think that is something that is very true in the speech market, and I think that trend will only continue as speech understanding products get better and better.

And the second one ties right into that, which is that speech understanding, this is so exciting for me, for everyone at Deepgram.

That the advent of speech understanding, as this exciting new area that really has human-like powers. 

And over the next 5 to 10 years, we’re really gonna see that take off. 

And as that gets more sophisticated, an ecosystem is forming where you have some platform players that are providing inputs, and that’s often the role that Deepgram plays. And then there are companies like UpdateAI who are taking some inputs from Deepgram, but then building these highly specialized, sophisticated use case models that solve really human-like problems on top of what we provide. 

You know, understanding the analysis and the follow-up and something that seems so fundamentally human-like. And I think that partnership is really interesting, and it’s impressive what you guys are aiming to do. 

Josh Schachter: Yeah. I think so too. I mean, in the world that we live in, there’s kind of three layers. Right? 

There’s the infrastructural ecos the infrastructure platform technology that you guys are providing us, of course, we have our own data science infrastructure, but then we exist at more of that application layer and that user experience layer. 

And then we fall into in our case, into Zoom, which is building an ecosystem of enablement tools with all their third party partners, which I think is absolutely amazing. I’m very bullish on that to for, I suppose, for distribution. Right? And for ease of access.

So from what I see in the conversations in the Zoom apps community amongst the Zoom team and founders of other applications. It’s really exciting because there’s lots of these enablement productivity types of solutions that are predicated on foundational technology like what you guys are providing.

Sam Zegas: Mhmm. 

Yep. Well, great. 

We’re almost at the end here, but before we go, we always take a minute to remember how far technology has come even within our lifetimes because life it’s gonna change a lot in the next the next 10 years. 

What the technological environment that we live in today is gonna be a completely different thing just to take it from now. 

So I’m gonna ask you to explain one piece of outdated technology, could be two pieces of outdated technology that makes you feel nostalgic for your own childhood, and explain it like you would to someone who’s 10 years old today.

So go ahead.

Josh Schachter: So I pray that our children don’t lose the art and skill of handwriting. I don’t know even know if his cursive still, you know, does it still exist. 

But imagine you know, the young kid today, try and explain to them the concept of a pencil.

And this idea that you take, I believe it’s graphite and you whittle it that you lodge it into wood, and then you whittle that down into some cylindrical shape and have a sharpener. Right? 

Sam Zegas: Like, a sharpener on the wall. 

Josh Schachter: Yeah. But it has shreds that you put the trash can underneath. Right? And then you have some rubber that, you know, undoes that and leaves markings on the paper, you know, that compared to just tapping on a screen night and day. So I think that’s a pretty stark difference.

Sam Zegas: Yeah. Really is a big change. I know that the culture of writing in cursive has really fallen away quite a bit at least in the United States.

I think there’s still a different culture of it in different parts of the world. I think people who write Cyrillic, for example, languages that are written with Cyrillic tends to write in cursive a little bit more still.

Sergii Kondratiuk: Yeah. So, actually, I would think, like, It’s crazy how many applications that we have right now in our smartphones, were devices like 10 or 15 years ago. And for instance, like some of them, they continue to leave like a camera. 

But some of them like a calculator, were very popular before, but no one uses them right now until everyone use applications in their smartphones. So yeah.

Sam Zegas: Awesome. Thank you. Appreciate you playing along. It’s been great talking to you guys. Thanks for tuning in.

Come check us out more at Deepgram and for information about UpdateAI. You can find them at update.AI. And, of course, you can find us at and at @DeepgramAI across all of our socials. 

So with that, we’re out, and we’ll catch you next time.