Damien and Nikola tackle the age-old tech question: should enterprise contact centers build or buy AI solutions? They cover pros and cons considered in the decision-making process, highlighting the importance of established vendors, internal capabilities, technological advancements, user experience, and the need for accountability and measuring success. Yeah, we sell voice AI, so we're a bit biased. Check out this buyer's guide for practical tips on requirements and vendor evaluations. 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.
NikolaThese days. It's not a case that we often even have to make because people understand that by getting that platform yeah, I'll build something with Dialogflow. it's like a bunch of bricks and planks and. Nails were delivered in front of your house and you're like, Hey, do an extension. It's cool. It's all there. You might need some help, we'll call someone. And then before you know it, you're like building a construction company and learning how to do all that. And you should have maybe just called the contractor to begin with and had them manage the whole thing for you.
DamienHello, friends, and welcome to another episode of Deep Learning with Poly AI, the podcast that brings two humans together to talk about a lot of non human things. I'm Damian. I'm here joined by Nikola, co founder and CEO of Poly AI. Hey, Nikola, how are you?
NikolaPretty good, pretty good. How are you?
DamienExcellent, doing very well, doing very well. I've successfully weaned myself off mushroom coffee, because I know you were probably wondering that. I am now back to regular coffee, and I couldn't be happier about that decision.
NikolaHow are your energy levels?
DamienThey peak and they value as normal, but at least I don't have this fear of some internal spore takeover. But I wanted to
NikolaOne fear off your mind?
DamienYeah. Yeah. It's very, it's a very salient fear, I think. Speaking of salient fears, let's talk about moving into the next generation of bringing AI to the enterprise. It is not an if, but a when. I think it's a conclusion that if you are the person who has AI working for you in business, you are going to be far behind in the competitive race. Obviously we say some serious low hanging fruit include opportunities and CS and CX specifically within the call center, but in general, this is going to make the question of, okay, we're going to do this.
DamienWe don't allow subject matter experts in our existing enterprise, because this is such a new groundbreaking technology. That a lot of the folks who are really learned in the space out there building their own companies, like Nicola Merchick so the question is, do we build versus do we buy, which is the eponymous question that goes back since the dawn of enterprise and corporate governance infrastructure. We've talked a little bit about. At dynamic and that dichotomy in the past, a few episodes and we touched on it, but I thought it would be a good use of time to dedicate an entire episode towards talking about the pros and cons of building versus buying when it comes to conversational AI, generative AI, anything within the enterprise, what sort of, I don't know.
DamienWhat sort of personality or what sort of organizational structures do you think are most prime toward one versus the other and just talk about a little bit of the pros and cons. So just right off the bat, when it comes to AI and the enterprise build versus buy, trying to be as neutral as possible, seeing as how we are a vendor of world class AI solutions, what is your topical take?
NikolaI
Damienmean,
Nikolathat's a tough one. And as you said, it is the oldest question in like software. And you've got companies with like big I. T. Departments, big budgets, enthusiasm, interest in doing new things. And, it's very interesting to build conversational assistance. It's and, we're at a very interesting inflection point where the underlying technology, a part of it, at least has gone way better. And that just allows people to be more effective at building it. So maybe before it would have taken them, two years to produce something, their perception, at least now is that with a newer, more effective tool, they can do a whole lot more, a lot more quickly.
NikolaAnd I think it's like every kind of like technologists It's like instinct to go and look at a bunch of vendors out there and say, ha, now that we have this, I'll do all that they do and I'll do it better. And I think, there's like the fundamental, trying to be not biased. One thing is like, go knock yourself out. Build something. Keep our, keep us on our toes. Make sure that we have a reason to exist. Do we have a reason to exist? Did the now just create like an AGI style thing that can do absolutely everything that we do in the backend help?
NikolaMaybe it did. You should try it. I think rolling out a full fledged voice assistant that does a good chunk of the work that your company does with clients on the phone. It's no small feat, right? Either way, either. You're working with Polly or with someone else or on your own, you're going to have to do, you're going to have to put your house in order. You're going to have to understand what you're doing, how you're doing it, what your internal systems and integrations are. Then you have some choices to make, right? Like which channel first, how do you go about it?
NikolaAnd then like, how do you report on how the whole thing is working? Is it working? How are you going to measure it? How are you going to roll it out gradually and demonstrate success? Who are you demonstrating success to? Was the arbiter of truth, right? Is it a science project or are you doing something that your business really needs, right? In our space, at least like you can do a quick take around. How's your customer, how's your context? I'm doing right. Do you have high attrition in the contact center? Are you struggling with those service levels?
NikolaHow fast do you need to fix? And then, I think like the real question to ask is, the same way you would ask yourself, like why vendor a not vendor B it's like, why do you internally have the right kind of talent to reproduce all this? And then the boring fiscal responsible question to ask is is that the best use of resources? If you've got such a great tech team in Europe, I don't know, an insurance company, are those guys better off maintaining and building better risk models? Or should they be building a conversational assistant?
NikolaWhat's it going to cost you and how long do you have to fix the problem that you have?
DamienBefore we dive into the cost a little bit, let's go back to that internal talent piece what are we talking about when it comes to the profile of we've I think often Proselytized, is that the right word the necessity for an AI competency center within an organization, especially if you're going to do a pretty significant implementation. But these people don't all need to have their doctorates in machine learning or, have a really firm grasp of of deep learning. And in, in our case, things like natural language processing and understanding its recognition dynamics, They don't have to be experts in the field, but what do they need in order to be able to approach an implementation like this if people were going to build internal?
NikolaOkay, so you're asking basically if we went for the full takedown of like, why not build you're asking what team do we need to assemble to build?
DamienYeah, if you're going to do this, you're taking a look, obviously a failed approach is to say, I'm sure we could do this and do this with who we have what sort of characteristics is an organization going to need within their org chart to be able to have a functional competency center, if they even want to approach the concept of building.
NikolaYeah, so look the main thing we got from LLMs and like the way that people use them and how they can use them is that it's really added to the democratization by allowing more people to do good work, right? And, hacking around with prompts, changing them, seeing how behavior changes. It's a lot more accessible. Too many more people, right? Cause all you're doing is using a one giant text file and then you you see it off the races and it's never been easier to build a demo. What do you start to learn as you push this thing out into the market is you need telephony integrations.
NikolaYou need to understand the workflows. You need to understand Just for architecture, right? What can you do programmatically? What can you not? And how are you going to do it? But then as you start exposing people who are not just your internal hackers playing with it, then you have to start thinking about user experience. Does it sound right? And I think what a lot of people working with LLMs right now are really starting to feel is that there's a certain GPT voice, right? I was on a plane back from New York last week, and like always, I downloaded a bunch of different YouTube videos, right?
NikolaAnd one was on the history of Somaliland. And about 10 minutes in, the voice of the narrator It just started grating on me. I knew it was text to speech, but I thought it was a text that someone had written and then just like pull through a text to speech engine. But really then the form of the sentences, I was like, this is so generated. It had that unmistakable, lofty, unnecessary amount of words that comes from the fact that LLMs are trained on a lot of, fiction and similar kind of literature, rather than the more factual tone that you would have.
Nikolain a documentary and definitely in an actual one to one conversation, right? Because no one loves a guy who goes off on a monologue and talks for half an hour, especially not if they're using complex sentence structures just to come across as more intelligent than they really are. And that's like more
DamienHawthorne than Hemingway out there, huh?
NikolaAbsolutely. Absolutely. And if, or like, Dostoevsky or something like that, sure. Just probably trim the 80 percent and not lose much meaning other than yet another description of the step. But There's something about it, right? Like where that user experience, it just, it doesn't sound great. That's not how your agent speak. And what people are finding is that it's really hard to control for this. Cause the safety mechanisms that say open AI has built into GPT 3.
Nikola5 and four mean that it's really hard. To mess with it's like actual literary style. And if you can't even do that, then you're stuck, right? Instead of asking yes or no question, you get like a disclaimer about, Oh, as a voice assistant, I can't really advise on that, but I could do it and you're like, Oh my God, like this is not working. This is out. And that's where that's not the world between fun prototyping and Oh my God, look, it works. Isn't it a miracle that it works into the real person interacting with your business is like, Whoa, what is this?
NikolaThis is I am not your experiment. I'm your client. I'm paying you real money for something. And I want to speak to you. I want to get that.
DamienMaybe then maybe the perspective is if you're looking at an internal communication platform. Where that tone of voice or that warm, fuzzy feeling that you're getting from this touch point with a brand less important than maybe build is a good vehicle, but you don't expect to carry that to your customer base and have them not perceive it just as you did. When listening to these YouTube videos where they're like, this is supposed to be something that makes me feel like my voice is being heard, that makes me feel like I'm a valued customer.
DamienAnd actually, it's getting on my nerves a bit because I'm detecting that uncanniness to it. So in application, it's what is
Nikolayour starting point, right? If at the moment you actually have a contact center, that's good. You can't replace it with something that's way worse. Maybe if you don't have a voice channel and then you add it, and this is a new medium that people will be grateful to use because they have nothing else, then it may be a step up, right? It's all relative, but if you have a good contact center, you will have antibodies in your company fighting your science experiment and then other things will happen, right? So the UX is only one piece that I think many are now starting to experience when using LLM mostly to build.
NikolaYeah, and we recently I came across another large multinational company that had built a chatbot and then, as they invested in brought up a team that was larger and larger, they started working on speech recognition. And they were told classical chat first vendor thing of let's build chat first. And then we'll plug in speech recognition. You'll have a voice system. And then these guys, it's actually not working. It's not working that well. And the answer was, Oh, of course we need to tune speech recognition. So they hired a few people for that.
NikolaAnd before you know it, a large company that is no way that company has a team of 60 people tuning a speech recognition. Now that's a large number of people. We have probably, I have about 150 people in the whole company. Now I can't really possibly say how the best resources for that company was to go on that crusade because while fun, no doubt, especially for the person spending that money, I think, the investment officer, they're looking at do we have to do this?
NikolaShould we be doing this? It would be akin to you and I deciding that this thing we're using for this podcast. It's not really the best software and, let's take three of our best engineers. Let's
Damienmake a studio. Let's build a better one.
NikolaAbsolutely. Let's build something that Nicole and Damien will enjoy using a hell of a lot more. I definitely have the top three feature things. Like I really want you to be centered in my screen rather than having to zoom you around so that your eyes are close to the camera. So that I don't look like I'm, losing myself in space, but is that really the best for a time? I don't think it is.
DamienFor organizations that have unlimited resources or near unlimited resources, so these, these megaliths it's a little bit of a different story. They can afford to experiment. They can afford to dedicate, they can afford to build their own studio and app, right? By all means,
NikolaIf you're like a fortune 50 company. And there's a powerful internal stakeholder whose only way to get that spark back into their, daily life is to rebuild poly inside, a large insurer or a bank.
DamienThat's an odd spark. I'll just say it right now. If you're that person, that's a very odd spark to have.
NikolaNo, kudos to you. You're actually a very kindred spirit to us. We should talk, we should go to conferences together. But it's not really the other stakeholders appetite or maybe, tolerance of that initiative will wear thin over time. And you will lose credibility because you're embarking upon, mastering, doing a masterclass in something that's, It's pretty niche still, right? And you're going to have to do a triathlon of machine learning that involves everything from like understanding and speech recognition to like reasoning.
NikolaAnd the other obvious thing in this space at least is like the both observability and analytics. Knowing what you're actually doing, how it's performing. That's a whole different set of analytics tools that are built for voice assistants to understand how humans interact, like what are they doing? What's the next best thing to automate? Where are we falling short? Just, the way we've run analytics internally, the number of fears, debates, and disagreements. We still don't have a properly canonical view of, what a good dashboard looks like for a voice assistant.
NikolaAnd we've been breaking our, head over it for seven years now. So I dare say that there probably isn't a canonical answer. And just embarking upon all that with, A few hours of your time. Hell, maybe you are better than we are. And if you are, you absolutely should be doing this, but you're probably not. So if you're just going to invest surely, then you should be buying, and you should be experimenting with vendors and you should be making them prove their worth to you. Especially if you're a large company, then, you've got a big price to dangle, right?
NikolaSo you should be. Working these companies, gradually increasing your investment and things that work and, getting educated in the process and. Help. Maybe you rebuild it later, right? Or maybe you don't, right? That then becomes like a pricing decision rather than a technical one. But for now it's very much a technical decision, right?
DamienThat is an interesting point. The concept of buying in an experimental phase, but then if you feel like you've got a good sense after a limited deployment to go back and build it yourself. We're talking about, like you said, the fortune 50 out there. We're talking about a very limited subset of enterprises out there that have ability of resource to do that sort of approach. And chances are, if they've done that with other technical systems, then they're more apt to do that with this, because at the end of the day, I is just another tool, right? So that's a very specific characteristic of a very small subset of businesses out there, but not largely what we see the sort of buy to experiment and build later on.
NikolaThe real question, to be honest with you, especially for companies that have a strong, like technical identity and a propensity to build a lot of things, people who hope, are your companies that have, five, 10, 000 developers and, really have a tech first DNA that wants to do this. And like, it's a positive thing. It's by no means am I saying don't do anything. But if you're that kind of company, then it's really more about, are you working with startups and smaller companies that have deep differentiation in one thing, or are you using wider platforms, mostly by like large cloud providers who are trying to just.
NikolaProvide, a level of technology that's slow enough to give you control while giving you like, I don't know, LLMs or packages of things that you can then use to build your own thing. And, take a telco telco is a classic example of a company that will almost always built. Rather than buy, right? And it makes sense, right? A company like BT has, their own contact centers, but it will also have like their own like a size like BT global that will go on sell to others, right? So for them, they're building technical differentiation knowledge so that they themselves can have a go to market of their product.
NikolaTeams taking a product and running it maybe as a managed solution for someone else in great scale. So to them, it's a two fold investment. It's not at all irrational for them to dabble in the build. But I think in this space, at least we're not yet at a point where the fight is between three startups with platforms and, the large cloud providers and platforms. It's still in the process of building the platforms, all of us, right? It doesn't really matter what your allegiance is and who you work with, right? It's just who is carrying the development?
NikolaWhat are the Manhattan projects around the world building a platform that can create like a good voice assistant? And it's really there aren't that many, and there aren't that many that care about it specifically. Many just think that, AGI will take care of it. At home, maybe it does. That was another
Damienepisode. We already touched on that. Another episode, regardless of the fact that telecoms have like the tech savviness and the teams embedded in their organizations that probably have an accelerated amount of prowess for this type of platform that they would do that, we're still talking about the issue of, a CX application to where that is it's commoditized across various industries, right? Just because you have the technical infrastructure and the teams in house to be able to build this sort of thing, doesn't mean it's necessarily going to pull off without a hitch when it comes to a customer facing application and especially for an industry like that, utilities, these other largely monopolized spaces where they never cared about customer service in the past, now they have to, because decentralization around.
DamienAround those sort of utilities that are now as a consumer, I have options, right? I don't have to necessarily go with this telecom depending on where you are in the world. I think that's a great
Nikolathing allowing people in that industry to make further investments and to make the case that they have to. I've spoken to probably thousands of CX leaders over the past, decade they all care. It's not that they don't care. Utilities care about customer service. Every industry does. I think that, it's more about the level of investment decision made by someone else that allows those leaders to do it or not, but I think that they're able to invest a lot more now just because it's a competitive edge. And, I, this has been like a narrative for about 10 years now. It's not the customer service is actually like a key part of the product differentiation, but yeah.
DamienAnd we just talked to Kristen punter recently she, we had her on the show and she discussed about how, validating what you just said about in the utility industry, it's often ill perceived that, People that the companies just don't care because it's like, you're going to be our customer, whether you want to or not, they do strive towards having like good customer experiences. And so that's refreshing to hear, but at the end of the day, outside of those business units and people who are tasked with that, I think there is perception that, it's not as important, which is why you're often faced with an hour long wait or something when you're calling into your telecom provider, your mobile phone provider, et cetera.
NikolaYeah.
Damienit is interesting to push back to the fact that this is largely something that's going to benefit the customer experience that more CX leaders are involved with that. And they are obviously vying for a better end user experience as such. The technical savviness of an enterprise or of an organization or the progressiveness of their tech adoption isn't necessarily a key factor as much as it maybe
Nikolacorrelates, you you mentioned a lot of our clients just now, but most of our largest successes in our like multimillion dollar deployments come in industries that are not actually, build first, thereby first, because those leaders then, write down the requirements. They look at vendors, they make some bets, they gradually increase them over time, and then they end up achieving something great. In the meantime, the builders are mobilizing, a team of everyone from machine learning scientists to engineers to UX people. They're like figuring out how to build poly inside a telco.
NikolaAnd that takes time as we know. And will there, will they waiver? Can they attract the right kind of talent? This is not a singular mission and pursuit of their company. And people know that the best people know that as well. Will they want to work on it or are they going to look at it and be like, someone's going to pull the plug because we not deliver something in six months, in 12 months, in 18 months, or, deliver the first thing and they'll learn what we learned like three or four years ago. And then, is someone going to reinvest? Are they going to get another funding round effectively internally to do startup has to do?
NikolaIt's a lot of this. And I think with those industries it's really hard for them to do this. So I think we'll see what utilities is
Damienand business function in the industry, right? It's, this has been a battle, an uphill battle for contact center leaders since the dawn of time is to sit here and fight for more resources. So the last thing that they want to do is be tasked with, conveying the value of a built program or project internally and then just to try and like sell that off when things don't go right Absolutely, and you know
Nikolaif it's an it led initiative at least in what i've seen Like IT around the contact center is not really typically the right kind of an IT team with the kind of resource and time to support something like this. So it's really much more, where I've seen people do this is recently. Would be like startups that are like, this is so cool. I want to work on it. I'll just do it quickly. It's so easy to do it now. I'll just use an LLM, right? And then they build it and it's like speech recognition is failing.
NikolaThe system is hallucinating left and right. Your latency is seven seconds. So like team of a dozen for each problem, let's go, let's work on it. Yeah. It'll take a bit of time.
DamienAll of a sudden you got that podcast studio and you're very small startup office and who needs that?
NikolaI think we should do it.
DamienI guess we need that. That's the answer. You mentioned a little bit about this, but recap if you are looking to buy a buying makes the most sense and for what, from what we've been discussing, it does make a lot of sense for most organizations. Just go with established vendors who know the pitfalls, who know the ways to measure success, who can continue that value proposition internally when assessing What do you suggest for end user organizations to look for and vendor partners? You talked a little bit about it topically, but let's drill down to integrations.
NikolaLook, I think the other thing that just you have to really look at is, if you work with us, if you work with someone like us, you'll spend a few hundred thousand dollars, right? And you'll have it up and running quickly. You'll see if it makes sense for your org and if it makes a big difference, and then you can scale that spend up to have really massive ROI later, if it makes sense, if you were to spin up a team of, 10 engineers, you spend that money and you're seeing the outcome later. And I think you just have to accept that it's going to be slower. Is it going to be more cost effective if you're looking maybe at some imagined long term unit price, perhaps, but that's quite a lot to stake such a big, important decision on, so I think, companies, the other, of course thing we see a lot of is people, like, no one ever got fired from buying IBM, IBM in this case was more really realistically applied to Google.
NikolaWhere I said, yeah, I'll build something with Dialogflow. And like, it's like, it can't be. A very, very obviously wrong mistake. And I think, this was a hard thing to present maybe a year, two, three years ago, these days. I think, look, it's not a case that we often even have to make because people understand that by getting that platform, it's like a bunch of bricks and planks and. Nails were delivered in front of your house and you're like, Hey, do an extension. It's cool. It's all there. You might need some help, we'll call someone.
NikolaAnd then before you know it, you're like building a construction company and learning how to do all that. And you should have maybe just called the contractor to begin with and had them manage the whole thing for you. But
Damienso essentially It's, if you find someone who's been around for a few years, who has some successful implementations of the belt, who have some, massive logos who are obviously shown some success. So not just a poly AI, even though we have all those things, but someone like poly AI, it's very rational and reasonable to expect that they have some of those basics covered the security protocols, the, personal or the private information controls, the security layers on top of the bots. And just, something that can scale out very well and not just, as you mentioned earlier, the commoditized LLMs that are out there are available that make it very tempting to want to build your own.
DamienThere's a lot of things that are beyond the surface layer that you need to consider.
NikolaThe real challenge you have, if you are this kind of decision maker is there's someone in your team. Who built something and they showed it. It's probably Chad. Maybe they've piped it into voice. They're showing it to you on a laptop, typically, right? Not even like on the phone and they're like, look, it works. And they'll tell you, say this and you'll say it and it'll work. And like. I think at that point you're going to say, okay, like either you get smarter about like prompt injections and all of that, and you take a really good run at it, but you need to like red team, your own team. And definitely need to red team, all the vendors that are going to, they're going to come and try to sell you stuff.
NikolaSo I think like you should just treat that, that internal thing as another piece vying for your business, right? Cause that's what you have as a CX leader, right? You have. A lot of spend that you spend on both human labor and software around human labor, and you're going to work pretty hard to create an overall better customer experience for your consumers. You're working on behalf of those guys and on the, on behalf of your business, right? So if you want to make the right decision, I think you have to scrutinize The internal builder, as much as you're going to screw the outside vendors.
NikolaAnd the outside of others, be they poly or someone else they'll probably have some reference goals you can make, and you'll probably be able to dial a few of the systems they've built. You're going to see some examples of outcomes, their dashboards of the kinds of business outcomes they've driven, the integrations they've done, the timelines. Like when hiring people, hell, you should go and do backdoor references. You should ask around, you should ask people work with this company. And then internally, if you're going to make that bet, I don't think you should, but it'd be, if you're really going to do it, then I think you really need like a clear set of gates that say, okay, like more funding will be unlocked when you show this and then that, and I think you need to get people to commit.
NikolaWhat they can get done by when, what level of investment, and then, you pull the plug. If you have to pull the plug the same way you would with a vendor, right? You're not going to go commit 20 million immediately, but you'll go and you'll spend six figures and you might have an incredible thing that makes a huge difference for your business. Two months later, hell that's all right. You can make us work for it. We'll do our best to prove you, to prove that we're, That we can back it up. I think with the build initiative these days, I think for that business decision maker, it's hard to actually test the internal team.
NikolaAnd that's where I think the DNA of the company prevails. Yeah, if they're going to do it, it's the kind of company that's just going to come and say, we're doing this because we are X, right? That's how we do things. And you, if you're CX stakeholder in that situation, what I've seen in the past is people who get very frustrated with it, building something that they don't really need and telling them that it's great. And then disagreeing. And then, we speak to them two years later and they're in a full buying cycle and they're like, we're ready. Like they call us, right?
nobodyYeah.
NikolaAnd it's because a staffing decision, something changed, some funding was pulled, that one builder left the company or moved on to building something else. Sure. And then they have a bit more freedom.
DamienSo key takeaway, no matter what you decide to do, The challenge with doing it internally is you don't even know how to measure success and you basically have to entrust the implementation team, the building team themselves to set their own standards for success, which, as could have vastly different outcomes. Whatever you do. Be diligent and holding holding your partner accountable. Makes a lot of sense. I think we're at time. This was a fun discussion as always. I'm eager to discover any in, in, in office studio builders who want to build us our podcast studio, please let us know, we'll give you a shout out.
NikolaAbsolutely.
DamienBut Damian,
Nikolawe should build some hardware for it as well.
DamienOkay, let's do it. Let's make it happen. Thanks. It was great to talk to you. We'll talk to you next
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...