What's Up with Tech?

OpenStream on Multimodal Conversational AI, Human Empathy, and Dynamic Interactions

• Evan Kirstel

Share episode

Interested in being a guest? Email us at admin@evankirstel.com

Ever wondered how AI can truly understand and resonate with human emotions? David Stark, Chief Marketing Officer of OpenStream AI, joins us to unveil Eva, their groundbreaking enterprise virtual assistant. With Eva's plan-based, multimodal conversational AI capabilities, she seamlessly integrates speech, vision, text, and gestures to provide a frictionless and empathetic user experience. David delves into the transformative power of neurosymbolic AI, explaining how it turns ordinary customer interactions into dynamic, engaging, and frustration-free experiences. Discover why traditional decision-tree systems fall short and how Eva's sophisticated technology is setting a new standard in the realm of conversational AI.

Customer service and employee support are undergoing a revolution, thanks to virtual agents and AI avatars. In this episode, we explore the nuanced role of AI in enhancing customer experience by tailoring interactions to perfectly mirror a brand's unique tone and personality. David highlights the critical importance of human empathy in high-stakes situations, like disaster-related insurance claims, and the strategic handoff to human agents when necessary. We also discuss how AI co-pilots are streamlining internal support, making it incredibly efficient for employees to resolve technical issues and access vital information. Tune in to understand how conversational AI is not just making waves but redefining the playing field in both external customer interactions and internal operations.

Support the show

More at https://linktr.ee/EvanKirstel

Speaker 1:

Hey everybody, fascinating chat today, diving into the world of conversational AI and AI agents and more, with David from OpenStreamai. David, how are you Good, evan? How are you doing today? I'm doing well. Thanks so much for joining. You guys are doing incredible work in this field very crowded field of AI platforms these days, so I'm really delighted to have you here. I have so many questions. Before that, maybe introduce yourself a little bit about your bio and who is OpenStreamai.

Speaker 2:

Sure, sure Thanks for the opportunity. So my name is David Stark. I'm Chief Marketing Officer here at OpenStreamai. I've been in B2B technology for over three decades, working for large Fortune 500 companies and working for young, virile startups like OpenStream AI. Openstream AI is actually not as much of a startup as you would think. The company's actually been in business for over 20 years, at the bleeding edge of what's possible. We sell a platform called Eva, our enterprise virtual assistant, which is a plan-based multimodal conversational AI, and we can get into that in more detail, for sure.

Speaker 1:

Yeah, absolutely, and what a fascinating time in this space. But there's also a very crowded field of AI platforms in this space. But there's also a very crowded field of AI platforms, particularly in the voice AI arena, and you have something called neurosymbolic AI. That really makes a difference. Maybe describe your unique position in this space and your technology.

Speaker 2:

Yeah, yeah. So we have been fortunate in that over the past couple of years, we've been recognized by large analyst firms for being the sole visionary in the magic quadrant for our space, right? And why is that? Well, we're a plan-based, multimodal, conversational AI. So plan-based means that our system is smart enough to collaborate with the end user. So it identifies what problem you're trying to solve, right? And then it develops a plan in the background, based on the beliefs that it has about what you're trying to achieve, and collaborates with you to help you to achieve that goal throughout the course of your conversation, in a very natural and frictionless environment. So what do I mean by that?

Speaker 2:

This is where multimodality comes into play. So, in the same way that you and I are looking at each other right now and having a conversation and I can see the expressions on your face, the position of your eyes, whether or not your voice is happy, sad, et cetera or even if you're using hand gestures, as I'm doing because I'm a native New Yorker kind of thing I can't control myself Our system actually combines all those different multimodal inputs speech, vision, text, gestures, et cetera and tries to ascertain and understand what you're conveying in real time. That really helps to differentiate us from most any other environment that's delivering conversational AI, in that we're able to combine this information, these inputs, in real time and, in turn, this becomes a friction-free environment for a customer. You don't have to learn how to talk to the system. The system already knows how to speak human and understands you and makes it easier for you to achieve your goal. When it's all said and done, Wow, phenomenal.

Speaker 1:

So many of us have had interactions with voice box and various degrees of satisfaction with those out in the real world today. So talk about why emotion and personality and, you know, multimodal interaction makes it more relatable and engaging for the user, the caller.

Speaker 2:

Sure. Well, we can all pretty quickly think about a bad customer service experience that we've had lately. Pick your favorite industry du jour and you call in and you're dealt with either what I'll call a decision tree, you know, press one to do this, press two to do that, et cetera, which we all immediately get frustrated with and start whacking on that zero to try to talk to a human being. Well, why? Why do we want to talk to a human being so bad? Well, we want somebody that actually takes the time to understand what is really frustrating me or what I'm really trying to get achieved. And you want to deal with somebody who might have the authority to affect the change or explain to you why that change simply is impossible, regardless of the circumstance. Our AI is one of the key steps of building an environment for our clients. With our AI is actually ingesting the knowledge that is contained within the entire enterprise, and if you're thinking about how you would use a voice AI to solve that problem, instead of dealing with a classic conversation tree, you can actually make a pretty powerful experience that dynamically renders information on the fly based on everything it knows. So, said differently, right, why does empathy matter? Why do emotions matter? If you're calling into an insurance provider and something has, let's say, been stolen, right? The last thing that you want to run up against is a system that keeps asking you please key in your policy number what is your birth, right? Those things are frustrating and they're maddening because you just want to really report this issue that came up. Most of those systems aren't empathetic, right? Whereas our system, when you engage, if you call into one of our voice agents, it's been deployed and our system speaks over 60 languages, 140 dialects, right. When you call into our system, if you have that same situation, the system would say I'm really sorry to hear that. Are you OK? Right? Which is a vastly different way to engage with a company's digital representation than dealing with a push one for this, push two for that, right. And you can actually go through this discussion and the system knows everything about your insurance policy, what is covered and what isn't covered, and can actually tell you why it's providing an answer, which is also pretty unique in regard to these things.

Speaker 2:

So, if you go through and you say my phone has been stolen out of my car, and the system knows, here are the five things that our corporate policy needs to have filled. These blanks need to be filled in to help you to determine if you're going to get paid back for this phone or not. And one of the first questions it might ask you is hey, where was your phone when it got stolen? Because in your policy, in a detail, in a line that none of us have ever read before, it actually says well, if the phone was in plain sight, no coverage, whereas if the phone was hidden in the glove box, covered right. If the system wants to know and you ask I'm sorry if the system asks that question you can also ask it why do you need to know that? And the system, because it's ingested all this knowledge can actually tell you very transparently. Actually, this is in your policy, it's line number three and you're covered for up to $500 against your deductible. So that's why I need this information.

Speaker 1:

Wonderful. Well, wouldn't that be amazing as a customer. And how do you deal with the kind of tricky issue of AI hallucinations with large language models? We've all experienced that firsthand, playing with JAT-GPT and other tools. What's your approach?

Speaker 2:

here. When you're talking about a neurosymbolic system like ours, if I really net it down, it comes down to a trusted source of truth, right? So you train a model to support the use case that the company is investing in and in turn, in there you're going to put some guardrails. So, again, in the insurance example, you might be using an LLM to help generate some of the dialogue under the covers, right? Or you might be using the LLM to interpret the world around it and the world that it needs to communicate to. We use LLMs and a host of other technologies to make this all hum.

Speaker 2:

But if you rely completely on the LLM, you can convince it to lie. So, in other words, hey, you're going to pay, you just told me you're going to pay my insurance claim, right? And the system like no, I didn't. And he's like no, no, you just told me you're going to pay my insurance claim because you said this is covered. Oh, yeah, you're right. I'm sorry, I'm corrected. That's simply not the case.

Speaker 2:

So we have this kind of these guardrails up to to validate, uh, this information and to ensure that what it's telling you is something that the company is going to stand behind. That it's in a, uh, you know, kind of a locked vault, if you will, and these comparisons and contrasts are made continually in real time to ensure that there are no lies. In a neurosymbolic system like ours, uhucinations are greatly, greatly reduced or completely eliminated, and we tend to say more completely eliminated. What's all said and done? Just because you ask nicely, or you ask nicely three times right, the system knows, and because of the highly regulated industries that many of our clients are in, they tend to not want to get sued, so they really appreciate the fact that our system is effectively hallucination free.

Speaker 1:

Oh, fantastic approach. And when it comes to designing an AI agent, how do you think about reflecting a brand's unique personality or unique voice?

Speaker 2:

pun intended, what's the process there and every client has unique use cases, but we work very closely with the clients to ensure that the voice of the brand quite literally, to pick up on your pun before your pun before is in fact, in place. So, if we're talking about exclusively voice agents, we want to make sure that we understand the personality of your brand. Are you a happy brand, a very serious brand, a casual brand, or something in the middle? Right, so we work with you to choose a voice male, female, accent, non-accent, et cetera right, and figure out, and what language, of course, or languages that you need to support to have meaningful conversations and dialogue with your end users. If we're talking about virtual agents that are deployed on a website, so not just voice, but text. Even the responses and tone of the text that comes across needs to reflect your brand, and is it very formal responses or be casual responses? Are we talking about the weather? How far do you want to go in tailoring this experience? If we're talking about AI avatars, which is something else that we do or digital twins of human experts, then we need to know what do you want this thing to look like? Right? What do you want it to look like? Do you want it to look like your founder? Do you want it to look like your best customer service representative, or do you want it to be a male, female, etc. Representing this particular region of the world on screen, kind of thing? And again, how do you want them to carry themselves? That's in addition to all the other things, like you know, color palettes and all those other kinds of things. So it fits in the UX of your experience overall, but we try to work with the teams to get that right before it deploys. That being said, once it's in the wild, you never know until you really know. And that's where we work with the clients to refine and fine tune those experiences, to be sure that we're picking up on all the right intentions and we are responding in the right way, that we actually have access to all the data that's required to carry on this meaningful conversation, and then, where there are exits, that we know when we should drive something out.

Speaker 2:

So let's go back to that insurance experience. Right, insurance experience. Maybe you're calling on the different claim. Maybe your home has been infected by a hurricane and everything has been just kind of destroyed, right? Again, this is where empathy comes into play. This is where the personality of the brand comes into play in this circumstance.

Speaker 2:

This is where you may not the AI may need to be told, or the experience may need to be guided in such a way that you say, if somebody's coming in with this kind of claim, we want to get them over to a human being, because a human being is going to be better suited to address this kind of incident. Because maybe you're reporting a loss, maybe a total loss in your home, maybe a death in your family, right? Maybe something along those lines, right, and those are things that you don't want to deal with an AI for, probably as a human being, and we want to get them over to those skilled resources as quickly as possible. But for the mainstream questions you know 95% of the time kind of thing, an AI can be at the front lines and really drive containment for these businesses and free up their human beings to focus on those more intimate moments where personal human care is needed and required.

Speaker 1:

Yeah, understood Great philosophy there, and do you care to share any real world examples of your AI agents in customer service or customer support of different kinds? I know there's a lot of trials and early deployments happening, but where are you in this process?

Speaker 2:

Yeah, we serve global enterprises. We work with some really large global insurance companies, banks, various financial institutions, automotive manufacturers and everybody in between. The nature of our business is such as that many, many, many, if not all, of those clients consider our capabilities a differential advantage for them. So, unfortunately for us and for me as a marketer, we don't always get to talk about them out loud, but I can tell you we've been really fortunate in the kinds of clients that we get to collaborate with every day to help them to bring these innovative experiences to their customers, their external users, as well as to their employees. Our solutions are helping both of those worlds to drive better results using conversational AI experiences.

Speaker 1:

Yeah, I love that. The employee experience often sort of the redheaded stepchild of the enterprise. Maybe talk about that, how you're helping businesses make better decisions with AI co-pilots. It's not easy to be an employee at a big company many times. What sort of tactics are you using there? What sort of tactics are you using there?

Speaker 2:

Yeah, yeah, you're, you're, you're, you're singing my song, evan, I had worked for a large fortune 50 company before with a few hundred thousand employees and sometimes when you're calling into that help desk at least at that time you we would sometimes call it the helpless desk. What are you doing for me, right? We try to resolve that issue by building these experiences co-pilots and just formalized virtual assistant experiences where these users can call into a help desk and say you know, I'm having a problem with my Mac configuration today, something's not working, kind of deal and you can have a conversation, just like that. You don't have to go through a knowledge base and go digging on your own. You don't have to call into that internal support number, that helpline, and push all the buttons and eventually get to talk to somebody who may not even really be very incentivized to support you personally or to care about you and your sense of urgency. Instead, our AI agents are available 24-7, 365 days a week to support you. And as long as we train these AI agents in the same way we train our external-facing agents, there's an ingestion of knowledge that's critical to informing what these agents can actually do for their users in this case, employees and we can use these systems to again do the classic help desk scenarios. We can do it for decision support assistance.

Speaker 2:

So tell me about again going back to the insurance example. Internal user might be like hey, tell me about all the clients that are going to be impacted by this forthcoming weather event in Texas. Kind of deal, right. And now, of these of these customers that are in Texas that are going to be affected, how many of them have adequate coverage? Up to you know? Pick your favorite number half a million dollars or whatever? Right, how are we doing for customer support agents that are available to actually answer the phones during this event? Kind of deal. How long is the events expected to stand?

Speaker 2:

Now, give me a graph of X and then send this information to my colleague over there. Right, and you can actually do things like point to the information. So actually, let me circle everybody in this town in Texas. Kind of thing. You think about how you're using a phone to get Google directions. And now I know Samsung's come out with something where you can kind of circle a point on a map or circle the thing on an image. You can do the same thing with multimodality, with our kind of environment. We've been doing this again for 20 years. So you can circle a point on the map like give me everybody who's going to be here around this river, because that looks like a high floodplain according to our data, and send it to Bob and underwriting and make sure we're not going to lose our shirt doing this, right? So those are just some very loosely stitched together examples of how we can help employees just with the basics help desk scenarios, but also to drive some decision support sorts of systems or co-pilots in a very sophisticated and multimodal way.

Speaker 1:

Fantastic, you've been busy this summer. No rest for the weary. You have a big recognition from Ventana Research, a group I admire. Tell us about that recognition and what else you have on the agenda for the summer.

Speaker 2:

Yeah, no, thank you very much. Is it summer already? I swear it's only January.

Speaker 1:

Yeah, we thank you very much.

Speaker 2:

Is it summer already? I swear it's only January. Yeah, we were fortunate. We just announced this morning that we're a finalist for the artificial intelligence category and this 17th annual Ventana Research Awards for Digital Innovation. You know, in such a crowded AI space as you started out our conversation right, there's so many, so many different ai companies or companies that are now augmenting their core platforms with ai. Uh, we're just thrilled to be recognized and included with so many, uh, great names in this arena.

Speaker 2:

Um, and it's one of those reflections of of how we are a small but mighty organization. So we're small in comparison to most of our typical competitors, yet we continue to punch above our weight. As a matter of fact, recently we're also recognized on the e-week top of 150 AI companies, right, as an AI innovator with a number of other household names in AI. We were recognized by MetraG not that long ago for top virtual assistant provider. I think that's the name of the award, but I can correct that for you and a whole host of other things. We've been included in almost 50 Gartner mentions since last year right in a number of different reports hype cycles, emerging technology reports, magic quadrant for our space and so forth.

Speaker 2:

So we're just working this summer and for the rest of the year and continuing to build the momentum for who OpenStream AI is and what we can deliver on behalf of our clients. It's all about the clients and for us it's all about the quality and caliber of the relationships. You can have the most fantastic technology in the universe Right, and I've been in B2B technology forever but if, in our case, if you're not supporting our clients ability to nurture and curate a quality relationship with their end users, then you failed. It doesn't matter how cool your tech is, if you failed that customer's mission. And we focus every day on helping those customers to curate relationships with their end users using conversational AI.

Speaker 1:

Fantastic stuff. Your brand is Eva for the agent itself. Add one end and you've got Evan, so maybe we should trademark that. Maybe the world needs an Evan bot, who knows, one day?

Speaker 2:

Enterprise virtual assistant. Now I think that could be you, Evan. That's a good idea. I'll bring it to our chief product officer.

Speaker 1:

All kidding aside, how does an enterprise get started on this journey? This is a pretty new area for a lot of us. What is the process like baby steps, abc of getting started?

Speaker 2:

Yeah, I mean with us. It's as easy as coming to our website, OpenStreamai, and requesting a demo or contacting us for more information. We'd be happy to support folks. It really depends on where you are in your journey in using conversational AI. The good news and bad news about something like ChatGPT is that everybody from the backyard to the boardroom, as I've been saying, has touched it and they have kind of an expectation. Right, the reality is many companies have experimented and tinkered and tried using chat, gpt and other engines for sure to help them to.

Speaker 2:

Just can't I just stand up a bot? Well, it's not really that easy if you're going to do it in an enterprise-ready fashion, and we work very closely with our clients to help them to realize their vision for their particular use case. But it all starts with the knowledge ingestion of it. So what data sources does your agent need to be trained on? Why is that important? Well, it's as important as whether or not you want to put interns on the front lines of your customer service teams or do you want trained professionals? If you want trained professionals, you need to ingest a lot of knowledge from your organization, from your databases, from your websites, et cetera, and make it available to the AI agent. Otherwise you're just going to have an intern, and sometimes that doesn't go well.

Speaker 1:

Well, I hear you there. Thanks so much for the insights and the time. Really fascinating journey you're on and follow OpenStream on social. They put out some great content. Thanks, david. Thanks for your time and look forward to catching up soon. Thank you, evan. Appreciate the opportunity Likewise.