This one starts off strong! Episode 3 in a weekly series touching on AI, voice and the customer experience. Today, Michelle and Nikola of PolyAI discuss the specific applications of genAI in the contact center, and what risks/benefits exist on both sides of the call. 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.
Introlive agent. I want to speak to a person, a live human, a man, a woman, a person, a dog. I don't care. I don't want to speak to your robotic.
MichelleHello everyone. And welcome back to polly AI's podcast. This is our third episode. And I think we're getting pretty good at this. What do you think, Nicola?
NikolaYeah, it's fun.
MichelleCool. So I'm one of your hosts, Michelle, and I'm here with Nicola, Polly AI's founder and CEO, a Forbes 30 under 30 recipient who also happens to have a PhD from Cambridge and machine learning and natural language understanding. And today we're talking about generative AI in the call center. The last time we got together, we gave a primer for generative AI for LLMs, and we talked a little bit about voice and a little bit about measuring the intelligence of artificial intelligence, but not in human ways. And Nicola, you said something really great that I want to call out again here, which is really about measuring the intelligence of these sort of tools that we're bringing into the, into business in terms of their business impact and not in these sort of you take an IQ test or Complete the LSAT and get a good score.
MichelleAnd so I think we're picking up in a really good spot with focusing specifically on the call center and what generative can do do for them. So I guess to get into it what does bringing generative AI into the call center look like? Do you see it as something that it can immediately start to drive value? Or is it something that's going to need a phased approach? Like how would you think about it if you were leading a call center?
NikolaYeah. I think the first thing to say is AI is already in the context center, right? We have voice aviar applications. If you go to very large enterprises like Fortune 100 in the U. S., you are actually almost expecting that you will be speaking to something automated before you get to a human. In fact, I think people are usually just trying to find the best way to get to that human as fast as they possibly can. And if you think about it that way, then really what we've been doing so far, We, the industry as a whole is automating the first 15 seconds, 30 seconds, and maybe more right of every customer service interaction and contact centers, especially enterprise ones, they tend to do a lot of things.
NikolaThey cover a lot of different products, a lot of different issues different support streams, different kinds of customers. You're often first, just trying to figure out who should you push. These calls too, right? Like the the routing, the like early triaging, answering some questions that's been there for a while now, right? Typically and even before the First Voice. Have you had, one for that tool, for that three? For that, we're just looking for ways to organize, to work, to get people. The agent that can help them the best, because with many contact centers, you have agents with different skills who are specialized to do different things.
NikolaThe first thing was getting the right agent and instead of having them listen to the menu and make a judgment around, the best group to help them, you will just have them express their problem in actual language and get to the right place, or maybe answer a question. And that's really been the bulk of what's being done so far that, and some heavyweight very often repeated transactions were done again in very robotic ways. So that's like world before generative. People call in and they don't really want to have that experience. So they just try to get out of it by screaming, shouting.
NikolaAnd when you think about the upgrade of that AI that we as an industry are providing to the contact centers industry we will do a lot better if instead of that average interaction lasting 15 or 30 seconds, we get to a minute, we get to two and, longer term you want to cover as much as possible and many contact centers are about edge cases where companies that continuously evolve had different issues and problems. You'll never really be able to handle everything and contact centers themselves. You have tier one agents that are typically newer on the job and then you have their managers, people who walk the floor they're tier two agents, escalation points who will come in and help when maybe a newly ramped agent can't help, or it's something a bit more specialized.
NikolaThis is already very familiar to contact center leaders, and it's just really a matter of building technology that will do a lot more of that. Intake work and the more kind of like high volume use cases.
MichelleYeah. So it sounds like, the phase is how long or how much of a call, how much of a sort of support interaction you can automate. And I hear you say, this is not something where you're going to be able to automate absolutely everything, every single edge case, but where is it possible today that generative AI can already replace? People in the call center.
NikolaYeah. Like the other thing to note maybe about the industry as a whole is that the goals, like if you look at it from like a PNL perspective and you're like, sitting in a dark tower and planning how to optimize, maybe you're as you do. I'm not sure that's your typical classical question of a marketing person, but Hey It's really more about figuring out how to get really good support levels, how to make sure you have enough staff, you're picking up calls in the right time. Those leaders typically think in terms of their budget and the service levels that they're trying to meet, and then what past service levels at the NPS and CSET scores.
NikolaSo when you think about. Adding automated labor in addition to human labor. Really there, there are places where you can do a lot, right? In the case of poly AI, the places where we do the most in terms of very high automation rates, we're talking 90 percent and above would be things like hotel and casino receptions, right? If you're going into order, an extra pair of towels or to ask for late checkout to check how to get to the airport, the classical things that you pick up the phone and call a reception for, that's where you can do a lot.
NikolaAnd there were pretty close to almost never needing to hand off. You go into things like delivery, appointment making, et cetera, and you're pretty high, right? We're talking like. 70, 80, even close to 90 percent success rates as well. So there, you can say that you've pretty much successfully done that heavyweight use case. Now, if they're edge cases, someone's calling a particular hotel and they called it because they're actually in front and they're a supplier and they have. The carpet's back from dry cleaning and someone's come out to get them.
NikolaThat's where you will still be going through the agent. And you'll probably have humans who are trying to circumvent that that automated thing by habit. This is especially true in the U S where I think the first generation of voice IVR did a lot more to basically alienate people and to get them to be pretty rude to these things. So I think, we can at least make sure that technology is replaced with something stronger, something that can do a lot more and offer a better experience while doing that. I can play an example of things. I think since we're now podcasters, I think we have to be we're going to have to play the sanitized version, but I think people will be uh, they'll be able to tell what what these guys are saying.
NikolaThese are calls mostly in Las Vegas, mostly past midnight.
MichelleAre you there? What can I do
Nikolafor you today? Give me the pool.
MichelleSure. Let me transfer you over to the pool. Please hold the line for one moment. I'm looking for the
Nikolabuffet.
MichelleWho at
Nikolathe buffet can take the call? Please hold the line one moment. I'd like some food in my place. Absolutely. Let me see if I can transfer you down to someone to help take a room service order. Please hold one
Michellemoment.
NikolaIf you could please hold for a moment, I'll transfer you to the reservations team. Remember to ask them about our 24
Michellekarat club. Name of the person you want to talk to
Introlive agent, live agent. I want to speak to a person, a live human, a man, a woman, a person, a dog. I don't care. I don't want to speak to your robotic.
NikolaI've played this to a lot of people. Our sales teams were. Often very worried when when I just shared this, I'd often play it even without without showing the beeped out version. And I think that at first there was a lot of trepidation. Then they just said that I'm an eccentric European and that's okay. And it seems to work quite well, but, this is a typical reaction, right? And people are, if you spend time in the contact centers, people call their upset and sadly, what they end up doing is a lot of cursing and a lot of being extremely unpleasant to other humans as well.
NikolaSo in that case, you're actually replacing some of that pain and just absorbing it because frankly, I think most of those tier two Asians, managers, et cetera. Knowing that the caller is going to sound like be a version of this. I don't think they'd really have a problem with, downgrading the McHugh to to speak to an AI agent. And I've always been impressed with just how resilient people in the contact center are dealing with this stuff. Cause you know, the stuff I'd be saying back to a lot of these, a lot of these callers would would be equally colorful. But, yeah, it's this kind of stuff that can be done extremely efficiently.
NikolaAnd then we go further and generative is going to be a big part of just giving us the capabilities to further change the attitude that people have to this technology. And one thing that I always point out and that I think is interesting is in America, the VR. Once in two weeks, right? You call large telcos banks, other industries. And you speak to something automated in the UK. It's about once in two months and country share language and much of the similar business practices.
NikolaAnd the UK is hardly a small market. You go to France and Germany and I'll bet you that half of the people at least have never interacted with a voice IDR, right? In the last three months, we launched the first automated thing in Sweden, in Croatia, in a few other places. And to me, growing up looking at these things It's, there is ROI, right? And the U. S. had higher ROI for these kind of 10, 000 plus contact centers where, people may have even been looking to just like replace labor, cut out pegs, from that dark tower.
NikolaThese days, post COVID, the churn in the contact centers really means that people are just looking for ways to stay above water. Yeah. That just means that they'll use automation. They'll do whatever they can to get those service levels back up. And that's really the name of the game these days. It's not so much about, a menacing new technology that's taking away jobs that people want.
MichelleYeah. I don't think anyone wants to be on the other end of the calls that you just showed earlier. And I think, this about me, but I used to run a contact center. About a decade ago for web hosting conglomerate It was just like a random sort of thing that happened in my career, but the biggest thing I was managing for aside from optimization and just cost really was burnout was actually employee burnout and people who were just. Exhausted at the end of the day, taking on like the emotional work of dealing with other people's unhappiness.
MichelleBut to that exact point, I remember the brand stuff, like all of the things about loyalty and just getting customers to be really excited about you and trying to turn a bad call into a good one and all the things that are super aspirational about, support leadership. I think that was secondary, like we almost didn't have time or energy really to put toward that. That said you saw a lot of, in the digital transformations that happened during COVID and actually started a little bit before then, you saw a lot of people starting to consolidate data for customer experience.
MichelleOn the support side, on the selling side, obviously, on the post sale side, trying to just make sure that people do product reviews, that you're encouraging loyalty, that you're encouraging more value, more lifetime value from each customer. And I'm wondering if that data consolidation, if that effort that all these big businesses put into making sure that they have great data lakes, they have sound ways to make decisions about what their customers should see and do, if that helps in this generative AI landscape, if that can help customer experience in the call center.
NikolaIt absolutely can. When you think about creating a really human powerful voice assistant that can have a free from conversation, right? They can take on both the instructions like, you had a dot context center where it's like hey in this case try to upsell in this case try to ask this question In this case try to prevent customer churn by offering, two three three months, right? You can play with these things if you have A really strong voice assistant and like just do a b testing over these things And just see hey this tends to work a lot more Then this other thing does.
NikolaAnd when you're running with a human workforce where, they're already stretched, there's no time to do any of this. It's really hard to programmatically flip a switch and say, hey, from this point on, do this. And I'll give you a mundane example, right? A lot of our hospitality clients run special promotions for events coming up. Think like Mother's Day, Father's Day, Christmas, Easter holidays of all kinds, right? Special events and at the end of the call, like the agent is encouraged, Hey, check if I don't know, let's say if the customer has booked something for mother's day.
NikolaAnd typically they haven't. You go, would you like to book? Would you like to book between two and 6. 00 PM? We'll give you a free round of drinks if you do. And then, the client has a bargain. They book, you actually managed to get the occupancy up because no one books between two and 6. 00 PM. And places are overstaffed for those events. And, someone might end up booking there because that's the only time available, but this way they managed to pack the venues full throughout the day. And, we've seen revenue jumps of 13, 15% on those days from doing that. Now, if you're.
NikolaYou'd love to do that. Everyone tries to do the best job they can, but you're sitting there, you're evaluated on the number of calls you took, you're evaluated on average handling time, how fast you are at resolving queries, right? You're probably not even properly comped for doing some of these like revenue generating things. Even if you are, it's clear, you don't really know. And the job is so taxing that you're just one call to another. Trying to go as fast as you can in that case You might just forget to do it in many of these calls whereas a voice assistant it's got an infinite supply of itself so it can just go and basically Ask in every single call and make sure that it's upselling As much as it can so they're big perks and that's, I'll put the data lake and stuff.
NikolaThe things you see when people actually start engaging with the agent and expressing themselves fully, not just with a yes or no, but they hear a human voice, they go out and they say they speak for a while, they express themselves with I think the average number of words that people say to our assistants in the first turn in the deployments where it's not like heavily transactional. It's 42 words, right? So someone's coming in. Hey, I'm calling because I was looking to come in. I'm coming in next week. And, I was looking for a place to park my RV van, right? This is I'm narrating a call that I recently heard.
NikolaAnd, I was looking at whether you have that kind of parking and you're like, okay that's a pretty information packed nugget that you have there. And if you surface that kind of data from the context center you're doing a lot of good potentially, because then. Yeah. Especially if you have, and we can probably show this in some of the next episodes, but we have a product in beta called smart analysts where you can just ask questions, right? And like you have a super intelligence that's looking at all of your data, right? And you say, Hey, do people ask about this?
NikolaWhat did they say? How did they say it? Does it happen often? When does it tend to happen? Do callers seem to be like frustrated when they ask about this or are they happy? What tends to happen after they ask that. And, that would be like talking to your most experienced agent, except they've seen absolutely everything they can quantify over it. They can tell you that, people are frustrated when they ask about this and get this answer. And maybe, that's a policy that needs to be changed. Maybe you're getting a lot of questions about something that, should be poking their eyes out on the website or, always, and these are all operational improvements where you basically then get that kind of like.
NikolaPositive reinforcement loop between the contact center and the business at large. And, there's generations of companies that have promised to use analytics and all of that to like mine that data and give you something back, but with generative AI we're able to harvest that data in a much more natural way. Our our colleague, again he likes to say that information companies like flows upwards, not really cards, but in sentences, writing clear, actionable insights. And that's what you can get when you have basically a conversational tool that runs over all your conversations and it's absolutely the data
Michelleback. Yeah. And it replaces the need for a call center manager, even an agent to need to know SQL to ask a question and get an answer back.
NikolaLike you need like different skill sets involved to go after it. And in a busy. Environment that just gets lost
Michelleinevitably. Yeah, but it's funny. I think if we were to sit and talk, maybe not dark tower, but ivory tower, but very light tower high up above all of this stuff, we would be talking about the idea of a super intelligent call center. And I can imagine somebody thinking about that and how they describe it. And I wonder how many people would be involved, like, how many humans would be involved. That said a lot of the things that you just said, talking about how much of a call, a generative AI bot or even an AI bot just generally can actually automate.
MichelleYou're still going to need humans in the loop. You're still going to need people involved. And so our super intelligent call center sounds like it's an interaction between people, insights. Data and then this automation like this really smart stitching of all of that stuff And so I just want to maybe get down to the ground off the ivory tower and talk about a really simple example of how Ai and people, come together here and that's in like the handoff So you have a 42 word turn right somebody's saying all this stuff and like you're picking up on all this data and you're realizing as a bot this needs to go to a person Is that in your mind?
MichelleYeah. Is that really like a valuable way to build a superintelligent call center to get these handoffs really right?
NikolaYeah. You also spoke about digital transformation, so it all comes together because often a superintelligent agent may not be able to do something simply because it doesn't have access to the right tooling. It does not have the APIs. Or maybe it's just something that. We're not confident like the leaders of leadership of the company is not confident leaving in the hands of an AI agent, right? Maybe it's let's say reservations for your most valuable guests or just really it depends on a case by case basis. So often, and, I think one of the most frustrating experiences for people is they speak to an AI agent, they do something.
NikolaThey could hand it off. What was your customer number? I just told you what my customer number is. Yeah, exactly. I actually maybe got half of my answers. I just need like this clarification. So the key bit there is and this is done with a screen pop. In our case where you're basically say, Hey, we've had a quote, made a reservation and there was a question about this that I couldn't answer. The agent picks up. Hey, Michelle. Okay. I see you've asked about parking your RV van. Like actually like we don't have that kind of parking, but the venue. Next door has that kind of parking and, you can pay like 30 a day to whatever, right?
NikolaSomething that might not be in the system, right? Or maybe it's taking a payment or just something else that the agent can do. But it's really important to get that kind of transition, right? The contact center is ongoing anywhere, right? If you think about all means of like technology that we've scaled massively. I don't know, the past century or throughout human existence, we built public transport and did that really. Like eliminate jobs. I don't know. I think tfl. Yeah, london is probably still the largest employer in london And I don't think I think based on the frequency of those strikes.
NikolaWe are well reminded here you know how many people are involved and how important they are for that system to run? So I think the contact center is going to be very similar right the dependence of those humans working there on really The operations team, which is what the context center and the greater scheme of the whole thing is will remain there. And the more we make it valuable. For the business at large by making sure that they feed that data lake and the most meaningful way they can because they are the kind of like front lines.
NikolaThey are collecting data from users, both satisfied, unsatisfied. They're collecting signals about what the company should double down on and where it really needs to improve. And, the more they do that, the more valuable they become. So I think that that whole upscaling puts them much higher in the value chain of the company. And like that hybrid workforce will be there. And, I think like when you look at like self checkout and supermarkets and stuff like that. Yeah, I think that maybe the total number of employees working to support that may be lower.
NikolaNewsflash like it's lower by the year now, people are not staying in those jobs. They don't want to do those jobs. And it's enough to look at the demographic pyramid of the US, UK, anything in the West and you see that. The writing's on the wall. We're not going to be able to scale the growth of contact centers with our populations.
MichelleAbsolutely. I like the rail analogy here. There's, I know we're at end of time. And so I want to just leave with this thought that follows on to what you were just saying. There's this great book by this guy, Thomas Malone, who found, he's the founding director of the Center of Collective Intelligence at MIT, and he wrote this book called Superminds which is fantastic. It basically just talks about how every great leap in humanity, every great, advancement has come from collective intelligence. Basically this idea that, we're better together when we're thinking together.
MichelleAnd there's this this evolution of that between machines and people. And I think the call center is going to be one of the... Great frontiers of this, where we start to see the collective intelligence of generative AI, the intelligence of the, these digital transformations and this, massive well of data that we're sitting on actually being used to transform customer experience. And so I, yeah, this was a great discussion. I think next time we get together, we'll get a little bit more into, what this looks like and how generative is going to be transforming this and how you can tell that it's making those transformations and then what how you can actually incorporate it.
MichelleDo you build, do you buy, how do we think about this as a, as a business unit. thanks so much, Nicola, and we will see you next week. Thank you, Rochelle.
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...