075: Human agent & AI cohesion in the contact center. What will it take?

Nikola and Damien are back to discuss the ongoing evolution of AI in the contact center industry. We'll explore whether AI is truly replacing human workers or simply augmenting their capabilities, covering Klarna's automation efforts and recent Gartner predictions, and analyzing the challenges and opportunities in implementing AI solutions. We touch on essential aspects of project management, customer service strategies, and the future of AI integration in contact centers. Join us, or be assimilated!

• Follow PolyAI on LinkedIn
• Watch this and other episodes of the Deep Learning pod on YouTube

Transcript

Nikola 0:00

So in that sense, is it replacing humans? It is doing some of the works that humans did once upon a time. But we need to get there, right? And I think that getting there is quite an endeavor, right? 2% of the population in America and the UK works in a contact center, right? That's a lot of people, right? And we don't have capabilities that are like, plug in an ethernet cable and here's your AI doing the work

Damien 0:30

Hello, and welcome to another episode of the Deep Learning with Poly AI Podcast, the only podcast recording on this, the first day of the Northern Hemisphere Summer. Happy summer to you, Nicola. How are you?

Nikola 0:41

Happy Summer. It is boiling in London, so I miss

Damien 0:44

As the native Floridian, and I always know when the first day of summer comes around,'cause this is when everybody in my home state would basically hve into a air conditioned room and not leave for three and a half months. But this could be a podcast that you're listening to with your windows down on your morning commute. Happy summer to you. There's been a lot of stuff going on in the news, Nicola, and I just wanted to dive into a topic that seems to be bubbling up in quite a few, like little publications, press releases and whatnot. We know what happened with Klarna very obviously going to market and saying that they were going to automate as much as possible and reduce a lot of their human workforce, especially in customer service. That backfired a little bit and recently they've walked that back. They're starting to rehire to bring in a more human ai, human machine hybrid work environment. Recently, Gartner just published a press release. Your co-founder, our CTO, Sean w. Posted on LinkedIn about how Gartner is saying now. By 2027, 50% of organizations that replaced human capital in customer suppor support functions specifically in contact center are probably gonna be hiring back this year because they went a little too whole hog on AI implementations thinking it was gonna be a magic bullet and therefore reduced staff count. Too much. So let's talk about that elephant in the room. And I know that we've had a longstanding opinion on the idea of AI is replacing people, but it isn't. It's actually evolving into a partnership. It's enhancing aging capabilities, improving overall customer service. Again, we've been saying this for years. Gartner's saying it's a key trend for 2025. Let's just talk about this. How do you see the current state of capabilities in conversational AI being an augmenting factor as opposed to strictly automation, specifically in customer service and what sort of problem is this overall?

Nikola 2:33

So look, I think there are a few things to talk about there. One is we're always trying to kinda like really rush this thing to completion and say, okay, AI jumped in capabilities is gonna automate everyone this year. How it needs to find its way into the market it needs to get implemented And a lot of processes of big companies that do a lot of work with a lot of people needs to change in a meaningful way, right? So as AI improves and we implement better systems for enterprises. It will do some of the work that humans do now, no doubt. So in that sense, is it replacing humans? It is doing some of the works that humans did once upon a time. But we need to get there, right? And I think that getting there is quite an endeavor, right? 2% of the population in America and the UK works in a contact center, right? That's a lot of people, right? And we don't have capabilities that are like, plug in an ethernet cable and here's your AI doing the work on behalf of your company. If AI is everything that people make it out to be today, and I just wanna make it clear that I'm absolutely in that camp, and I think that's of course. It's much more powerful than people really mostly appreciate, understand, and it will only play out in the next few years. But it takes kinda like wiring it up and connecting it all up in a meaningful way. To turn the workflow, let's say the work of 5%, 10%, 20%. Of your contact center into an AI agent doing that work on their behalf, whether it's by talking to people on the phone or chatting or doing backend tasks, like moving file files around, doing data entry into different systems. That's all something that you have to plan and implement. So I think those that. Would like you to think that they are priority, remove people'cause they believe the AI would make up for it. I think that's a bit wishful. I think that for the most part, what I've seen is that contact centers either struggle to hire, I. They struggle to hire even more, right? Like it's never easy, right? So it might just be that they hope that they realize some efficiencies. For the most part, that AI that they're referring to is a warning message. Telling people we have long waiting times. Please go and you can do this online, and you could do that online, that really is a question of just how are you willing or how far are you willing to go in trading off your CSAT scores for OPEX reductions? Because if you really can do, say, half of the things that people are calling about online, but they persevere in calling, you have an option, which is to never service those categories of calls. You might turn some customers from your service, say you're a broadband provider or. A utility company that depends on whether it's a monopoly or not in a given state. But if you do that, you might, you will frustrate a lot of your callers. And over time there will be a new study state where some will move online and some will move to other providers. That's,

Damien 5:17

that's giving whispers of the advent of like offshore business process outsourcing and whatnot. That was the entire catalyst for that. What are we willing to trade off? To reduce opex and we, by no means should AI earn the reputation that early stage offshore business process outsourcing came in because it was a completely off-brand experience, largely, and it was seriously frustrating to customers.

Nikola 5:38

No, but listen, much like there are incredible offshore contact centers that are really good, right? And those that are not. There are really good instances of AI being implemented to do incredible work and instances of places where it's been completely Exactly like where it's a failed attempt, right? So that's okay. That's noncontroversial. That will happen and it will give AI a bad name and a good name. And over time we gotta make it better and technology will improve. So over time we'll arrive at a better steady state. Right now, I think like more locally. Prescient is this question like, what do you automate in a contact center? And I've long thought of like how to express it to different audiences. And in a very businessy framing, it's really just a matter of trading off CapEx with opex. And what I mean by that is let's say that you have that you are a bank, right? Will you build out a system that allows. Me like an AI vendor to integrate with you so that I can recognize Damien Smith who lives in this city, that this state, et cetera. Am I happy if you use your mother's maiden name and then two factor authentication? First the CapEx is I gotta pay people to scope this whole thing out. Have a project proposal that I need to get it approved by InfoSec. Then my IT teams or partners IT teams working with me, have to build all that plumbing for the AI system to connect to so that it can be a first class citizen inside the contact center. Right now, the second gate there is either. Customers are willing to engage or not, and it's never binary. They're willing to some extent, some are very willing, some are less willing, and they're just kinda like a continue, right? So it's up to you around like how far you're gonna do what we call harsh containment around pushing people to use it at all costs versus say the second time they ask for a human, you might just cave in and give them a human, so you'll navigate that adoption curve somehow, but really at the end of it, you will have. Put in resources into the implement implementation of the AI system. There will be continuous topics from using the AI models, right? And then there will be just the continuous the upfront investment and the continuous investment into the plumbing around the AI agent to do all that. Right now, if you're building an AI agent that's going to take a few million calls of a particular type of year, and you. Reasonably confident that it will do a great proportion of them, then that might be a very good investment. But if you set out to automate an infrequent call type, and then you start building into your old esoteric APIs that you don't really know that well, and then you encounter one issue and another there, and then the whole project gets delayed, the total investment into human labor, another. Like resources into the whole thing, expands then maybe. Maybe it ends up being something with a very long payback period, where you shouldn't have automated. And that's a question that we've been navigating for years with our enterprise clients, right? We have quite a few systems that now do the work of a thousand people. And let me tell you, some are easier than others. None were particularly. Very star when people didn't believe this was possible. But at this point, some of them have scaled up to that level of usage in a matter of six to nine months. They start, and in month one or two, three, we end up doing the equivalent work to a few hundred people. And I. 6, 9, 12 months later, we're doing the work of a thousand. Some took longer. And it really depends on just the organization's standpoint, ability to like align this giant GaN chart of their work and our work and everyone's work to drive it to that successful completion. And it didn't really used to think much about project management as a discipline, but as we started doing this with some of the world's largest enterprises, I've realized that. It's like one of the most important things in ai, right? Yeah. Because your models could be great. They could be the best in the world. They could be getting better, and they probably are right, but they're not being

Damien 9:27

utilized correctly. They're not being in the right application then.

Nikola 9:31

It's also just what do you do with them? It's no one doubt. Ethernet cable. But whether you build a website that people want to use or not is not the Internet's fault. It is your fault as the designer of the website and as a company behind it with a product to offer. So along, you mentioned,

Damien 9:45

and you mentioned a little bit about how the companies we're working with are abandoning those esoteric KPIs and looking at new ways to establish value and understand where we should put our automation efforts. Not just. We need to automate, let's go carte blanche, but where would this be most effectively used? You said sometimes it takes six to nine months. Other people recognize it early in the variety of organizations that we work with who are recognizing this value and doing it quickly. And by taking a new approach to project management and assessing those KPIs, are there any that stand out as being new or emerging? Within the space that are counterintuitive to the historic, all right, measure, csat, measure, average handle time, blah, blah, blah. Are, is there anything that's popping out to really show a better correlative value early on? As far as KPIs are concerned, maybe even qualitative if we don't wanna go quantitative?

Nikola 10:32

Yeah. Look, I think that qualitatively the call type that you should go after that will be the highest value to automate are calls that are high importance to customers. But intrinsically lower value to the company itself. Because, they might be something that doesn't make them money, but it's part of them providing that service, right? And it might be like very high volume, right? If might be people, think like order tracking returns, stuff like that. Sure. Where someone receiv the good, really cares about them, but they've already paid. And it's a cynical take on it, you always follow the money and you find out just what, any issues related to taking a payment, and like booking that initial deal or selling the product that tends to be done really well. There's a reason we have one click buy for almost everything through four different services. Now, as that one click buy, once you start doing order tracking and other things, it tends to be a bit more difficult and maybe just historically a bit less invested in, which is why these are calls that end up being potentially very complicated for people. And to navigate both in the contact center and buy the Kohler the consumer themselves. And for those calls, some you can automate really well. For others you might not be able to. But then really the equation gets modulated by like, how many of them are there? What's the prize of automating? And for, to do that, you just really have to know what tends to work. And you also, depending on the kind of like profile of the contact center you'll know what to go after and just how ambitious to be in terms of directing people into an automated channel.'cause if you've got a really well staffed contact center and a high CSAT. You probably don't have a problem. And in that case, we really shine'cause we do very soft containment where at the first sign of trouble or frustration we'll hand you off to the human contact center and the discrepancy between the resolution. So what percentage of people feel that they spoke to AI that solve their problem? And containment people who didn't get to speak to a human right. The more narrow that gap is. The closer we are to a system that is not that neutral, but actually positive to both customer experience and the operating expenses of the enterprise. Right now, if you've got a contact center that does not have enough people then you're in luck. You're not in luck, you have a problem, right? But a problem, you call the doctor, we come in and we say, let's automate half the calls. Those calls will now be cheaper than those handled by humans. Sure. And people speak to you sooner, and we'll do a really good job in those calls so that the other half that maybe we haven't automated are now not waiting, instead of waiting for, I dunno, 10 minutes. So the experience for both groups of customers improves.'cause no one's waiting. And if you've really ensured that you've. Automated the call types that you can automate, then you just become a better company at a much lower cost and the more AI improves, the more of this you'll be able to do. But it will still take and build the careers of many of the executives, the kind of people that we work with, where, they're the ones turning this contact center into that command center of the future where they're building the capabilities to go into these different fields of work that they do. And to really wrap AI around them.'cause that's what we do. That's what our, that's what our kind of counterparts that buy poly AI do inside their companies.

Damien 13:46

When you talk about that command center vision of the future, and, we've touched on both in events and talking talking opportunities as well as on the podcast before, I think at least partially about the future of the contact center. Highly distributed, not a big central room. Pretty much there, there could be a, a complete disillusion. But the idea of a centralized contact department versus a more nebulous structure, what does that command center look like in the future? And is that really where we might see the true offsetting of necessity of human agent interaction?

Nikola 14:25

No. Look, I don't think that building the command center means that you have a distributed contact center. Okay? You might have a great distributed contact center, or it might be centralized. That's, I think, a orthogonal question, right? I think what matters much more is just the ability. To do much of the work with automated AI systems that feed that information back in the best way possible instantly into that central cortex that is the enterprise. And it's not just customer service where this will happen, this is, this AI ification is happening throughout the enterprise, right? Yeah. And what better place to identify. Then in the department where you have the most agents right now, you have so many and you're doing so much in the contact center with high attrition that you will still have, if you now have thousands, I guarantee you'll probably still, if you have 10,000 people in the contact center, even in five years time, you'll probably have at least a thousand. And that's still like a chunky organization, right? That those 1000 people will be paid a lot more and their jobs will be way more interesting'cause they are the co-pilots and the kind of the officers of that company going to the future because it is the central ops department, customer experience where, the downstream effects of operational decisions made upstream and then the impact to the customers at scale. Makes them, if they have the right kind of AI agent capabilities, they're basically the best outlet for a marketing department that could exist. Also, the best collectors of signal that you could ever have. So I think that aggregate

Damien 16:00

customer data, potential upsell opportunities the concept of personalized marketing beacons in the contact center,

Nikola 16:06

it all converges, right? Yeah. Companies become smaller, but far more effective and far more powerful, nimble, agile, and I think, a lot of people then think well. Doesn't that mean that these companies will increase their profits and be even bigger while being, like smaller nimbler and what are the rest of the people who are no longer there gonna do? The truth is, in the world accelerating as fast as we are, we're all working like 16 hours a day. And it's never been more exciting. And part of the reason for that is that there is so much going on and so many people are doing new things, right? So I think humanity rather than seeing. Disillusionment by this new world where there's intellect that is not just human that we work with. I actually just see humans enticed and challenged and absorbed into an effort that is even more fulfilling. Yeah. Because that same company, no matter how good it is an old company transforming itself, trying to fight. An onslaught of in incoming new companies that have, with AI built up capabilities, ground up with ai. That these guys never even had. So therein you will have a collision between your Walmarts and your Amazons. And we'll see who wins, I think. And at the end of the day, the inherent structural advantages of very large players are usually sufficient for them to, remain very important. But new ones will emerge. And then in the next generation, they're, they will both be old companies and they'll continue to reinvent themselves or to get eaten by. Capitalist endeavor. So well, according to,

Damien 17:36

according to Idiocracy, the winner is Costco. According to demolition Man, the winner of the restaurant wars is Taco Bell. So either way I bet, I think a bleak future in that future maybe, but we'll see what happens with the rest of the organization.

Nikola 17:48

So you smell that. I don't know how you feel about this, I feel like global food has just continued to improve massively. When I. I don't think I had sushi until I moved to the uk. And when I go back to Belgrade right now, wanna sleep like sushi on every corner. I, it's not every corner, but it's serious. Yeah. So I think that's the proliferation of, everyone's oh, we don't like chains. If really good sushi chains brings up around Europe, that means that Europeans get to eat sushi. And that's

Damien 18:14

a logistics problem. That's the logistics problem that they're solving is how to get sushi to Belgrade. Yeah,

Nikola 18:18

That's always, yeah, that's always interesting. I think you look at just like the proliferation of all of it. And then you look at your New York and London and, there's someone banging their head around how to innovate further and create the new wave of pokeballs or whatever. So I think that will happen with all of these large enterprises as well. There will just be sustained innovation and core productivity that. We'll create a world that we won't be able to recognize five years from now. Ten five. I dunno anyone's guess.

Damien 18:42

When you said the 10 years you also said look, if you had a, if you had a contact center of 10,000, it's probably still gonna be 1,010 years from now, or five years from now. Those thou, those thousand. We talked about this a little bit historically, the re-skilling of those thousand, are those gonna be people who still are fundamentally contact center operatives, agents at heart, or are they coming from other parts of the business to now run that function? If the former, what sort of skills are they adopting to keep up with the Joneses for that organization in the future?

Nikola 19:14

So there's always movement inside companies and I'm these functions that are like, kinda like growing closer together as you integrate everything. We'll definitely have people who like, mixed in, but for the most part, it is the business. It is the contact center agents, their managers, coaches QA people who will get to take fundamental of now managing this AI behemoth that interacts with most of their consumers. So they will be the ones that. Form the vast majority of that 1000 people. I think that, it has a really important role to play, what we have seen previously and continue to see is that it's not just. Of the it, it's not like it comes in and builds something and now, like the contact center is no more. It is the contact center that has to reinvent itself into the command center. And it can only really do that when companies like Poly come in and create the technology for them to do that. Yeah. And we do it hand in hand with them because, frankly, it's the only way, like we've tried just giving technology to people and we think it's great, but then you know, those people don't live the daily struggles of a contact center, and hence, they can't affect meaningful change in the contact center either. That's true.

Damien 20:23

Some may say for those who oversee those day-to-day operations within the contact center, that these assertions by Gartner, these moves by Klarna are indicative, at least in the short term. On a boon to things like agent assist and using a AI tools to help out human agents. Do you think that's the case or are we still looking at a potential track in the short term of operating in tandem and not necessarily cand in hand? I.

Nikola 20:51

I listen, agent assist, that's like Henry Ford's famous. If you ask people what they wanted, they would've told you they want faster horses. Faster horses. And try to optimize this. And, a lot of agent assist is built to, it's really not solving the root cause, which is like the need to service that thing at scale. And it's really about ramping an agent that might have joined yesterday to be at least. Okay.'cause you know they're gonna quit six months later anyway. Sure. That's a cynical way to fix. And of course it can help. And there are parts of the agent assist stack that are very useful, like close summarization and all that. And the agent will still be there. So where you can build like robust assistive tools and make their lives better. You absolutely should. But I think, and we always kinda invite our prospects to think about it. If in a world where you have 10,000 agents, you imagine a world in which you have a thousand agents, right? Making those 1000 agents 10, 20% more effective becomes a lot less important. Then getting to a world where 90% of your work is done by an AI hive mind that is. Far more powerful and capable of both servicing your customers and helping the organization figure out what's actually happening with your customers at that scale. One carries with it an infinitely bigger price than the other, and you should prioritize it. Most companies do not. All of them do. I think that a lot of CCA vendors really Agent assist as a way for. Companies to tick the AI box. Like we've done ai, we've AI from our CCAS vendor and I think you have to follow the money, right? The CCAS vendor gets paid on the number of seats you have, right? Correct. So they will defend that seat count at all costs. And if you. It's like turkey's voting for Christmas, right? But if you believe them that they will automate half of their revenue away for the benefit of your organization you're a better person than I am.

Damien 22:42

Okay. On Turkeys voting for Christmas. I think that's a good outro. Hey, I really appreciate your time today, Nicola. This was an enlightening conversation as usual. Any last thoughts on this, or are we gonna end with turkeys?

Nikola 22:56

Oh, it's, let's end with turkeys. It was good to see. Sounds good to

Damien 22:58

be. Sounds good. All right. Talk to you soon. Thank you so much.