In this episode of 'Deep Learning with PolyAI,' Damien and Nikola reflect on the significant trends in AI throughout 2024, including the rise of generative AI, the stabilization of models, and increased enterprise adoption. They also discuss the importance of customer experience and data management in AI applications. As they look forward to 2025, they predict further advancements in hyper-personalization, enterprise scaling, and regulatory clarity. Join us for an insightful discussion on the evolving landscape of AI. Follow PolyAI on LinkedIn Watch this and other episodes of the Deep Learning pod on YouTube
Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.
Transcript
From the show. Tap a time to play from there.
Nikolathe world shifted a little and we started getting people who did 10, 20 different gen AI projects in different parts of their business. So they took something that's very reminiscent of. Venture capitalists a portfolio approach, right? And then, most of those pilots failed.
DamienHello, and welcome back to another episode of deep learning with Polly AI. The only podcast that is ringing in 2025 with a full luxurious beard by 50 percent of the hosts. How's it going, Nicola?
NikolaIt's pretty good. She told me I wouldn't have shaved.
DamienI know you can pull one of these off in 24 hours. I'm very, very jealous. How's your new year kicking off?
NikolaIt's good. It's very good. With a bank, like a lot of work, I think we were I think December 31st, both two new deals and stuff. It was it was very exciting. So we were like photo finish and then a bit of time off. How was yours?
DamienAbout the same, very excited to see our growth. And I'm really excited for 2025. It's going to be a very interesting year to say the least. But not to put into the shadow just yet the year of 2024, which was a massive year, I think for us as a business, it was a lot of fun. As you said we grew like mad and we had a lot of. A very busy December as well, but also just in the world, the January or the general environment of AI CX context center transformation, where we play, we saw a lot of like trends galvanize, I think in 2024.
DamienSo I thought a good thing for us to chat about today was a year in review, taking a look at last year and getting your expert opinion in the trenches, academic. Observational, like what were some of the massive trends that you saw in our general space that really emerged and shown in 2024. How's that sound? You want to go through some of those points today? Yeah,
Nikolafor sure. For sure. Not that I have a list, but
DamienI'd love to put you on the spot. What would you say? What would you say was the, what would you say was the key shining beacon from 2024 with regards to this ever changing world of AI?
NikolaNot to be a Debbie downer, but it didn't change all that much, right? If you just like interpolated at the end of like 23, the excitement, and then you looked at everything that happened in 24, like A huge amount happened, right? Like I feel like. If I had to bet, I would have maybe wanted even a bit more, but a lot happened in terms of it actually, like on the model side, right? Like it feels like it's stabilizing a bit, which is a sad thing. I would want it to, I want you to be defined. I want like, I want more news, right?
NikolaI want your changes. I think we were primed to have it accelerate even more. And maybe it's just, cause we're in the thick of it. It doesn't feel as fast. Then I think in terms of bringing it out to the world, it was the year, right? I think, we had like first, second, third, 20th fully generative voice assistant rollout in the year, right. And a couple of their platform self serve everything around it that was really focused on bringing that to life. It was hugely exciting. Yeah, it was a good year.
DamienSo generative. I agree. Like there wasn't any new groundbreaking models. I feel that came out in 2024 that transformed, but there was a lot of evolution. I also think that with regards to enterprise adoption and like openness to bringing in generative, it seemed like 2023 was the year that it came onto the scene for wider adoption, but also the year that there was a lot of hesitation. Are the risks outweighing the potential rewards and enterprise applications. It seems like in 2024, most business leaders were like, no, These risks are overcome.
DamienIt's easy to manage for them. It's impossible to eliminate them completely, but the benefits definitely front way, the reward. So we saw a spike in adoption or at least openness to looking at generative solutions on an enterprise level in 2024. Would you agree?
NikolaYeah. So I think that if I looked at what people in implementing AI virtual assistants before, it was like the early adopters. It was a very evangelical push where, you know, people who are open to experimentation, we're doing it. And then the world shifted a little and we started getting people who did 10, 20 different gen AI projects in different parts of their business. So they took something that's very reminiscent of. Venture capitalists a portfolio approach, right? And then most of those pilots failed.
NikolaAnd I think that, somewhere halfway through the year, maybe a bit later, you had a lot of like commentary from analysts and others like, Hey, most of these things will never make it to production. And like it's not all that it's made out to be. But that was wrong, the winning ones to emerge. And actually pilots are by and large, like they should be failing left and right. That's the purpose of a pilot, when you think about it. Especially with wildly new technologies and many vendors trying to make them work and integrated with enterprise workflows in different ways, it takes time, right?
NikolaAnd it takes trying out brave things. And some will pay out massively and others won't. And you'll just limit funding and double down on the winners. Much like a VC firm, but there's in his portfolio, right? And you need one company to deliver like 30 X. And if you allocated it right, you have a 10 X fund and everything's great. So I think J I has followed that person. There was a lot of willingness. Like if you think of it as like seed rounds. It was quite easy to start a lot of things. I think at the end of 23, it may be a bit harder now, right?
NikolaJust the flurry of different things. You hear a lot of talk of just they're really like three players when it comes to foundational models now, right? Sure. Open AI anthropic. And I think X is XAI is has emerged as like a. A late addition that managed to like through Elon's, fast reaction, buying GPUs and stuff. They managed to like maybe secure that third spot. And then it feels like it's crystallizing there in the kind of like model layer, right?
NikolaThen when it comes to applications, much like, enterprise sales cycles, it takes years for winners to emerge. And. Companies like Poly AI are just like working and growing fast and like really building on like the initial basis we had by building more complicated, more valuable use cases. And I think that, wherever you look there are companies starting to really capitalize on the new things that are now possible with with LLMs. So it was a big year, but I think, Much like that portfolio now needs to grow. Yeah,
DamienI hear you. And that makes a lot of sense, double down on the established leaders because so many popped up in, in 2023, what's interesting is I feel like some of the hype that could have been discovered with new foundational models. Coming out and really transformative foundational models was translated to enterprise application of said evolutions and new functionalities that were coming out. So while we're not getting a five or we haven't gotten a five yet, still how GPT four and four turbo grew going into omni channel going into multi lingual going into, different functionalities The way that those were adapted to enterprise use cases, I think was the fire of 2024 when it came to generative AI, it was really, how is this getting used to, increase operational efficiency to whatever the case may be to transform the enterprise landscape.
DamienSo maybe enterprises actually were able to stand in the spotlight while the foundational model providers, LLMs and whatnot were just like, okay, they, I guess last year was our 20, 23 was our big year for significant model releases. I
Nikolathink, when you look at just like the multimodality, which became like your main topic of time it was bit right. And like, yeah, many advances happened and cumulatively they're probably. As important as what happened the year before. I think it's just like, emotionally, as I look back, I expected that we would further accelerate and I don't really feel like it, it happens. So I'm hoping that this year ends up being more exciting. The end of the year had very like, weird undertones of like people saying, Hey, scaling laws, we've hit a limit, et cetera. And like, we have not, right.
NikolaIt's just going to take a bit of time and more work. I'm like Data and just like maybe some architectural changes, innovations on that front as well. Yeah, we'll see. But I think when it comes to the enterprise, I think what I found interesting is that. Towards the end of the year, we started making huge progress with a few, very large clients that have traditionally been builders, right? And when it comes to gen AI, they were the ones that thought that they would do everything in house, right? So I think that it's interesting that those outsized investments initially, as they often didn't bear fruit, then they went back to buying from the application layers, companies like us.
NikolaSo that's, so the builders became buyers in
Damien2024, huh?
NikolaThe most sophisticated builders. Have swung into buyers and also I think decisive influence influencers on our roadmap in a way that I didn't see coming. If you ask me,
Damiendo you think that was largely centered around the perceived complexity of managing LLMs within an enterprise application?
NikolaYeah. I think it, it came down to the ability to make progress and justify the investments. So I think, you basically. You can backpedal and say, Hey, I won't give me all that much, or I'm going to use a different approach to get more out of AI, right? So if they're like, halfway through their night in Vegas and in terms of the money, you go like, okay, rollout is not working. Maybe it's time for blackjack, right? So I think that's really like more of what happened. And yeah, it's been really interesting because those clients are the ones that actually have a strong opinion of like how they want to partner and collaborate.
NikolaSure.
DamienWe're a platform. They have been really valuable. Incredible. Oh, one of the things that you mentioned was obviously access to data. I would safe to say, and I'm so much so that I'll put it in the Chiron here. I think that this, I think 2024 was the year of big data. I think it was a year of revisiting the essential nature of data, both in knowledge management, data management, where you're putting your data stores, because. As like LLM and AI adoption practices go on within organizations, the essential fuel for efficiency, there is data. And so it put the big spotlight on what are we going to do with it?
DamienHow do we manage it? Where are we storing it? How do we access it? How are we organizing it? Et cetera. And then of course with all those ducks in a row, you get the benefits of having a more organized both view of the customer view of the business. What do you feel? Do you feel like, do you agree? Was this the year of big data? 2024.
NikolaI don't know. It's just an overloaded term. I feel like I've heard it once, yeah, like it's been around for a decade. I don't know. I think that, just to disagree for the sake of it. Like, I think a lot of times we run out of data, like we don't have more data. I think the multimodality and all of that stuff. It's going exactly in the direction of like, we got to use a different form of data, richer form of data where we're not exhausted the signal, right? There were a lot of lawsuits and stuff, copyright things, et cetera, being worked through. I think on our end, we've definitely put in like a serious amount of work into contracting and figuring out, The right to use different voices, the right to use different kinds of data for training, different kinds of models, people opting in to make their product better or others.
NikolaSo I think like the level of literacy around like the legislative side of data is definitely growing in buyers and users and everyone. And I think that like that question of like monetization, I don't know if we really have an answer. You have that New York times lawsuit where People who have a lot of data are hopeful that they can they might not have figured out how to monetize it, but they want to preserve the optionality of potentially monetizing it in the future. I don't know. We have any answers yet there. I
Damienmean, those lawsuits gave birth to like a new business model for, traditional media companies, which was, you got to pick your horse and go with it. And license access for training and purposes, because it's going to happen anyway. So you could either sit here and battle exactly what level of access is entitled to these models or these model providers, or you just say, okay, I'm going to give you sanctioned access, open AI, I'm going to give you sanctioned access, Anthropic, and then let you have exclusivity agreement there. And that seems to be the course that those big media companies took.
DamienYeah.
NikolaYeah. Yeah. I don't think it played out in a particularly interesting way, but I think that it has played out. Right. So like there, there is a landscape of how you do these things. And I think, yeah, I don't know. If you feel that way like, um, like a lot of it was definitely done with data. Right. So like, I think about accumulation and stuff definitely turned into More applicable models and things last year. So I guess we'll see how it goes. Like, as we turn through all the video and stuff, and that gives us like that next generation of models, but
Damienmaybe also one of the things that I think feeds into another trend curious on your thoughts on this, but I do think there was our new. reinvigorated sense and view of the customer experience in 2024. And a lot of that was data driven, but many of the AI and customer service applications were with the customer in mind and not necessarily focused on more traditional, like operational excellence, cut handling time, make it as streamlined as possible. Like bare bones customer experience knowing that customers are getting more fickle.
DamienWe've had a few episodes in the past where the consumer experience is not as forgiving now as it was in the past. And they'll forego patronizing a consumer brand on one bad experience, like 60 percent of people. It's something insane. Just, you took me off once and I'm out. So there has been this new reinvigorated sense of the customer experience. And so I think that was a key trend within our space is a lot of both operational metrics that were within contact centers shifted to how, what's the customer like, not just CSAT scores either, but reducing things like average handling times or focus on that in favor of, are we solving the problem?
DamienAre we containing and solving the issue? A lot of that is data driven, right? Because some of the other elements of that. Personalization, predictive support, et cetera, are all based on consumer data. So I, again, another stake in the ground. I think that this was the year of the customer, the year of the customer's importance to a lot of businesses. What are your thoughts on that?
NikolaYeah. I think that. It's definitely the year where you have the technology to put the customer first and to create that experience. Cause if you think about like how we used to buy things and do business, right. It was like often a very personal relationship with individual seller. Right. I was just speaking to one of our colleagues and she was saying used to buy insurance for like the same guy. Girl shows up every year and goes like, Hey, how's your family? They don't partner, children, et cetera. And like, they would just take care of it for you. And you have that level of trust.
NikolaRight. And then we went into this like online thing where like you click, it's very non personalized, right? It's just, you're one of many, right? That became like the website as a non personalized channels, like log in and do whatever. And if you log in, maybe we have your like payment details saved, but that personal touch, whether it mattered or not, or how much it mattered, Debatable, but it was gone, right? And I think now with if we go to your kind of like, data claims previously, like that first party data and like it's organization in a way that allows an AI model to make use of it.
NikolaTo become that seller at scale that knows things about you, your preferences, et cetera. Like, like it's starting to happen, right? We have many in a system launched where when you call in, like, we know who you are, we know what you pulled about the day before and we're like, Hey, like, continuing on your, router troubleshooting thing. Did that work? Right. That's very different. Like, even with a very well provisioned context center, unless you're able to You know, get the same human that spoke with you yesterday, if they remember, it'd be hard to emulate that.
NikolaAnd now it's starting to happen at scale because you're just able to give that like hyper well trained, but also personalized and continuous state preserving thread. Yeah, I think that at least in our space, like we, we've made great strides. And I think that the customer, like the customer being the enterprise is They're getting a lot more. Ambitious about what it is that they can now build with this kind of technology. I don't know if this was the year of the customer. I think we have a decade of work on,
DamienFair, but I think as AI finds its footing, especially in certain applications, a customer experience and customer service was low hanging fruit. Because I. I feel like to your point, you go back to those old school insurance days, right? It was somebody showed up at your door. How are the kids, et cetera. That was the personal touch. So you bought the person. You didn't really buy the product because the product was a commodity. Then you move into the era of automation, right? And then old school automation efforts. If I walked into a store and I was, my biometrics were scanned and I heard some really robotic voice say, Damien, if you want more cereal, go to aisle seven.
DamienI'd be like, this is creepy. I hate this. But you blend the two right now we're in an era where you have more personalized experience, even with automation solutions. You have LLMs that are providing really good like voices and a sense of a personal experience. It sounds more like a conversation in text, in voice, et cetera. It takes that person who's showing up at your door and selling you insurance, but mixing it with the capabilities of, a broader sense of customer data and. Service at scale. So it's making it, it doesn't feel as creepy now if I call in and I'm like, Hey, this sweater is, it doesn't work.
DamienAnd they're like I've seen that you bought three other sweaters at this size. Do you want to try and up a size or down a size, et cetera? Like that feels more natural and organic. And I think blends the best of both worlds. I really think that experience has only come up in the last year or so.
NikolaYeah. Yeah. Look, I think, especially in retail, right, where there is a level of what products where, I exist on John Lewis jumpers, right?
DamienWe got to get them as a sponsor. I think this is the second mess mention of them during the pod.
NikolaYeah. It's if they go, Hey, like there's new. Great jumper. And you worn out your previous, like five, like, I'll probably just be like yeah, send it, charge me. Like, yeah, you upsold me, but you also like serviced my recurring need for John blue jumpers. Right. So I think like, that's not easy to do, right? Like you've seen humans try and look, et cetera, but it's clumsy. It's a bit, as you said, like, whether it's a human or a robot, like Damien cereals on aisle six, like, Whoa. But if I'm like looking at sauces, weirdly, and they go like, Hey, we're out of Carolina Reaper, but you can buy it on Amazon.
NikolaOh, you're a hot
Damienfan, huh?
NikolaOh,
Damienthey're
Nikoladropping some Carolina Reaper
Damienmentions. Nice.
NikolaThere is a, there's an incredible British company. That makes like a Carolina Reaper ketchup. And it is, I will put the link in the description. I'm kidding. Oh,
Damienplease. Yeah. All right.
NikolaIt's a second
Damienunsponsored product. I have a
Nikolastash of 10 of them at all times. They don't go out of business so that I want to keep it going as long as humanly possible. So these are all things that you could pick up from like my purchase history. Right. And you really took the time to think about it and then you made a recommendation, I'm almost definitely buying, right? Whereas like, if you try and I see you doing a clumsy job at it while I'm calling about a refund or return for something else then I'm going to get pissed off in a major way. Like, you're actually like decreasing my experience by trying to make more money, whereas really all I want is more money.
NikolaThere's what I want. And there's like a really good recommendation on that. Like space in between is a dangerous space where I think, yeah, because when you non tactfully
Damienlike mentioned that you have so much information on me and it hits me the wrong way, then I do get extremely turned off by that experience.
NikolaYeah. Like only if I can get you a recommendation for something incredible, then you're going to go like, Oh, Like, big brother, everything about me, but damn it, I loved it. And you saw people opt in to share their data. If the product becomes better, but you have to explain the necessity of that.
DamienYeah.
NikolaI know some people that, like don't share their data with Google maps because,
DamienFear of the drone showing up, following around. It's like,
NikolaI know some people who are like that, but very few are. I think most people are like buying to like, Hey, if I share my data, like I'll be able to see traffic and that's like being good citizen of that application. Right. So I think like, people are starting to see value. And I think we're only really, like, you, I use chat GPT. You've noticed the, like memory, save memory. Yeah. Don't know about you, but like, for me, as I've like, had conversations with it about like different products and things, like when it brings in like, Oh, let's talk about cars to kids.
NikolaHence, like, you probably don't want to look at that because that there, like whatever, if you have more than two suitcases, that's going to be a problem. If you try to put a problem in the back, like. That's good. That's starting to be like, because you can always like add to the context window, but if it's like curated from a lot of conversations, that's starting to become like the ultimate personal shopper. Right? And there's going to be a lot of value from all of that. And I think a lot of value for that, like, in neural network terms, that hidden representation of free editing, You are, I have a vector information, whether it's like discretized or not of like who you are and how I can like make a product better for you.
NikolaThat's going to be really valuable. Cause that's like the super cookie. Right. Of like who you are, the super
Damiencookie. That is the title of the episode. The super no, it makes a ton of sense. And I agree. All right. We're running up on time here really quickly. Three words, 2025 looking forward, the space that we just discussed, give me three words that are going to basically describe 2025 in the year. This fast evolving market.
NikolaI think like the hyper personalization is one. All
Damienright. So continue hyper personalization.
NikolaYeah. I think like we we got to show more, right. We got to keep people motivated. Right. I think like, the enterprise scale is like the second thing. We're like,
Damienall right, scale,
Nikolalike winning investments and like, proof of concepts that have become like. So many figure deals are going to like get repeated at scale. And that will define, I think the winners of the enterprise giant space to play. Right. So what did we say? We said scale personalization. Then yeah, I think it's just like, I'm waiting for the foundational layer. What
Damiendo you think regulation has gone for 2025?
NikolaLook, like, I think we, It was a good year last year in that, like, like the EU act is over, I think slightly overreaching before we see it play out. That is what it is. Thankfully both the U S and the UK have been a lot more reserved and that I think will help define the tone in the Western world of like how we,
DamienI think that's going to continue with the incoming administration as well. There's very much a, let's not talk increased regulation.
NikolaFeels that way. Yeah.
DamienYeah.
NikolaAnd I think with that, I think we have the clarity we need, I think it's also just like from the initial like panic of a world that realized that AI finally moved faster and delivered more at the end of 23. I think that there's always that like, shadow of all the sci fi. Films you've seen where, like AI runs out of control and becomes our overlord and like, it didn't happen. It's not happening. That's not, and I think people, that's settling in slowly. And that I think makes.
NikolaPeople like including regular regulators, a lot less likely to do anything rash. So life is good. Like, I'm very optimistic about everything for this year, right? Not just because we're doing well, but because I think like, there's like, some kind of like the elections behind us. All of that kind of like pent up stress is has gone. And I think that. Head, we're just going to like work right now. I think I'm, I was in a, on Serbian TV with a bunch of people and it was
Damiena great spot by the way. Can I link that in the description as well? Along with the link to sweaters and ghost Reaper ketchup
Nikolaor Carolina Reaper, rather. Yeah, totally. And, I think that there are people who are afraid of it and like, there's a lot of just like general, Population is reticent about like what this will do to their lives. Will their jobs be automated? But the truth is like, I couldn't be more optimistic because like there's an overabundance of so many things that are not like really good. Technology about like the availability of information, specialized services that are not available to scale that might have been available to, like a few people. Right? So I think, like, future's ours to lose. There's so much good coming.
NikolaRight? And I think we're just gonna make sure that, society stay stable and distribute some of those against everyone.
DamienYeah, no pressure. Just keep society stable. That's it. Not a huge ask. Great. I appreciate your time, Nicola. And let's see what 2025 holds. But I can guarantee this. It'll be more of deep learning with Polly and that's for sure. All right, you take care.
Three years ago, Simplyhealth's PolyAI deployment was only resolving simple FAQs over the phone. Today, they run voice, in-app, email, chat, and WhatsApp conversations on the platform. Nikola Mrkšić talks with Paul...
An agent that gets better every day compounds into a very different agent within months. That's the idea behind Wren, PolyAI's new release: an agent whose job is to make your other agents better. Some of what it finds...
Companies are pouring money into AI, but not everyone is getting the returns. Ian Jacobs, VP and Lead Analyst at Opus Research, thinks he knows why. In this episode, your host Nikola Mrkšić sits down with Ian to work...
When PolyAI built its Builder MCP, one client helped shape it: Vixxo Facility Solutions, one of the largest facilities providers in the US. In this episode, guest host Damian Sasso talks with Vixxo's CTO Derek Neighbors...
PolyAI just launched Dialog-RSN-1, its first audio-native model, and your host Nikola Mrkšić sat down with its builder, Matt Henderson, to unpack why it’s a game-changer for building voice agents. Most voice AI either...