Max Clark: [00:00] No slides.
Michael Cibelli: [00:02] slides.
Max Clark: [00:04] We don’t do slides.
Michael Cibelli: [00:06] okay, I was totally unaware of that because April sent me a slide deck and said she sent it to you.
Max Clark: [00:14] I asked April for a case study to talk about. We don’t do slideshows.
Michael Cibelli: [00:15] We’ll you.
Michael Cibelli: [00:20] Okay, I can do that. That’s not a problem.
I think we lose a ton by anyone not hearing a recording of one of our live calls, but that’s your call.
Max Clark: [00:36] Yeah, it’s from experience, as soon as we introduce like a presentation and slides, it just goes flat. And it makes the conversation not great. So it turned into a thing that just like, you know, it’s one of my non-optionals. We just don’t do slides and presentations anymore because I’m trying to capture recording and a conversation, right? Like the conversation then doesn’t flow and doesn’t get into its The ones I regret and the ones that I think back on and cringe, it’s always got a slide deck involved.
Michael Cibelli: [01:11] All right, well, okay, then can you give me an idea of kind of the flow where we’re gonna go? What kind of questions?
Max Clark: [01:21] It’s pretty much ad hoc. My request to April was give me a case study that talked about a customer problem that you solved. And we’re going to talk about that case study and the customer problem that you solved. then through the course of that, that was the one that she sent. Yeah, unless there’s another one you would like to talk about instead.
Michael Cibelli: [01:33] So we’re going to talk about carvers, is that what we’re going to do?
That one’s great.
Max Clark: [01:42] OK. Yeah, it’s let’s see what else I can, it’s very organic. The point of this is, if we were in a hallway talking, that’s basically the format and what I find works best. You prefer Michael, right? OK, just making sure.
Michael Cibelli: [01:59] No worries. Got it.
Max Clark: [02:08] Some people’s emails sign in with one thing and their names are different things and it makes it really confusing and complicated for me for no good reason. What else should I tell you here? Would you like to do any pushups or burpees before we go? get the, some people, there’s actually a thing, know, was Dr.90210 before all of his QVC things, he was doing pushups before the show. you know, I’m not making that up.
Michael Cibelli: [02:35] did my stretching this morning. I’m all good, man.
Max Clark: [02:38] Okay. Uh, the way this platform works is it does a, um, uh, it does a recording on your local device with a progressive background, uh, uploading the background. So while you, get like a video back and forth with us, the video that you’re seeing is not the video being recorded. Um, so if, if the call quality goes, if it’s a call quality, the video quality goes crazy. If the auto quality goes crazy, don’t worry about it. Everything’s you’re you’re you’ve got your ISO on your side. I’ve got an ISO on this side.
And then I need to figure out a better way of doing this. Let’s see here. think it’s going to be your top right side. So I think if you look here, there’s a little magenta icon with a number in it and a percent and like an uploading. You’re other than talking a lot. And when I say talking a lot, if you ever feel like you’re talking too much, just talk more. Honestly, I’m here to facilitate you talking. So.
Michael Cibelli: [03:32] Okay.
Max Clark: [03:37] Do not for a second think, shoot, I’m talking too much. I need to talk less. This is the moment where all social grace goes out the window, and I’m going to try not to talk. But other than you talking, when we finish and I stop the recording, I just need you to hang out until your percentage says 100. It’ll happen pretty quickly. Fast internet, these things go really quick.
But I just need you to wait until it says 100 % and then we have the recording and we’re good to go.
Michael Cibelli: [04:08] Sounds good. And how long do you anticipate this going?
Max Clark: [04:12] Because we talk about single topic, usually runs somewhere between an hour and an hour and 30. It’s not like a Joe Rogan format where we’re going to sit here and we’re going to chat for four hours. That requires usually we were going to have to hop around with lots of different topics. But I find that with a single topic around the hour, hour-ish mark, we’re going to be running to our end.
You know, if it goes longer, goes longer. That’s why we give ourselves a two-hour block. It gives us plenty of time to not worry about it. Don’t stress about it. We’ll get to a point where it’ll just be apparent. Like, we’ve talked about it enough. I guess it’s the best way of saying it.
Michael Cibelli: [04:57] Okay.
Max Clark: [04:59] last thing is it’s not ambush journalism. we do a very light edit after this goes to the team goes, actually goes into our, our queue and the team will edit it. so if there’s something that was glaringly like, I can’t believe I just said that that was really bad. It’s undisclosed. I can’t talk about that yet. That’s going to give me in trouble. Just let me know, say something. We’ll market. We’ll make sure it gets taken out. so, we do have, we do have, you know, companies that are publicly traded.
So it’s very aware of all the little nuances and got to be careful with some things stuff. I’m going to get another notepad real quick. This thing, let’s see here. I’m going to switch notepads because this thing is not going to be great for this.
Michael Cibelli: [05:37] I don’t think there’s a there.
Michael Cibelli: [05:51] getting over an illness so I’m going try not to cough.
Max Clark: [05:54] Oh, it’s, it’s, uh, if you got a cough, just cough, you know, same thing, you know, just, just take a pause. Um, all right. Anything else I can tell you questions, comments, concerns, let’s get going. Okay. Um, I’m still learning how to do this. I’ve done 200 of them. I have a quasi idea of what I’m doing at this point, but I’m still learning. So if you think it’s some like, you know,
Michael Cibelli: [06:08] No, I’m good. Ready to roll it.
Max Clark: [06:23] We’re just just flying by the seat of our pants here. You know, it’s really it’s really let’s just let’s just, you know, have a conversation, have some fun.
Michael Cibelli: [06:30] Sounds good.
Max Clark: [06:34] Michael, before we get into specifics, I think some background would be really interesting to get into. And when I say background, what I’m looking for is the industry has shifted from the state of machine learning AI to now like LLM-driven AI to this multimodality AI, which of course is driven text and then and then voice. just for a quick start overview, can you give me a quick story arc over the last, let’s call it, five years of what’s actually been going on in industry and what’s the trend change that’s occurred?
Michael Cibelli: [07:18] Yeah, sure. think, you know, over the last five years, what I’ve seen is this heavy lean into the CX world and how that we can solve things relative to CX with various different modalities. As an example, I would say to you, you know, there was the push into chat, which came out pretty quickly.
There was a lot of folks leaning on chat to say, Hey, we can offset a lot of our calls by handling things through chat, move into SMS, texts, et cetera. but what we’re finding is, and I think surprising to many is people still want to talk. And so the ability to create a voice driven AI that truly is on brand and makes people understand where we’re going, easily understands what they’re trying to convey and respond was something that our founders thought was, was very much needed. our three founders came out of the university of Cambridge language lab in the UK, where they were working on various projects, including Siri with their leader there as they were gaining their PhDs in In voice. And so they developed this. This voice led customer led conversation so that. The old IVR is of the past that were driving people down various paths rather than where they wanted to go. Was the idea behind all of this? And I think the results are quite astonishing in terms of what they’ve created. and what we’re able to bring to the CX world into customers.
Max Clark: [09:20] You talk about like IVR is taking down paths. I mean, I’m the worst. I’m the guy that’s like pounding zero as many times as I can just trying to get out of the IVR and like, I don’t think anybody has a really positive IVR experience at this point, right? You know, it’s too much, too complicated.
Michael Cibelli: [09:35] They don’t, it’s funny because here, here in the US, especially where we adopt things more quickly than most places around the world, the IVR experience was absolutely awful. And as soon as it tripped up, the next thing you said was representative. Boom, boom, boom. I’m pressing that, that zero as many times as I can, right? That then developed into…
Max Clark: [09:56] You
Michael Cibelli: [10:02] The early adopt or the early signs of, of virtual agents, which weren’t understanding very well and didn’t sound very good. We’re very robotic cause people instead of pounding zero saying representative representative.
Max Clark: [10:19] Mm-hmm. Human, human, human, human.
Michael Cibelli: [10:21] Some expletives and representative. mean, it could have been a myriad of things that they would scream into the phone. And it was funny because as I said, these guys that developed this came out of the UK. The UK was not as predisposed to bad IVRs as we were in America because it wasn’t adopted there as quickly. So they’re more apt to listen to us from this voice AI perspective because they didn’t have all that.
Max Clark: [10:25] Right.
Michael Cibelli: [10:52] Here, I think now what we’re seeing is an understanding of, wow, this could be really good. These can sound very natural. And by the way, the other side of the equation, which we can talk about more in depth, is even more important, and that’s the understanding. What we understand is incredible. So there are two sides to a conversation, and someone can make something sound really cool.
But in fact, can it really understand what’s coming back the other way? That’s a big deal. When you talk about LLMs and you’ll talk about ASRs and things of that nature, that’s what makes us really unique.
Max Clark: [11:35] Let’s get into IVR a little bit, right? Because I think initially, the idea behind the IVR and really press 1 for this, press 2 for this, there was like, it was team selection. I mean, you were talking about, I need help with this thing. And so I’m going to have a specialized team that I can route traffic to. And at a high level, I feel like that’s a really appropriate thing to do. You start talking about skills and skills-based routing. or the extension to that became like persona-based routing or caller-based routing. I’m a whatever tier Hertz customer or loyalty program. I meet something in your CRM that says, OK, Max is a nightmare. Send him to this team. Or we’d like Max and send him to this other team. And I don’t think.
That wasn’t really a problem with the IVR world. But then all of sudden, these IVRs have gotten to the point where they got overly complex. And you kind of just get trapped into this tree of this IVR. And I’m curious in your perspective of why that happened and what has driven this IVR spider web nightmare that a lot of companies have ended up deploying.
Michael Cibelli: [12:55] In my opinion, it was really driven by, again, the IVR was sending people down specific paths. And when you send down a path, you then needed people that were competent in terms of understanding and responding to that particular question or path that they were being sucked down. What that caused in many instances was compartmentalizing people into different groups that could handle certain tasks. versus a very open ability to understand what was said regardless of what was said and be able to handle that particular response. In doing so, it grew those groups that needed to handle that because to your point earlier where you said, if I was a terrible customer or a pain or whatever you wanted to put it as, you were sent to this group. Well, if you got a lot of those, that group grew or that group sat in queue, which diminished the CX experience for that customer because if they were difficult customer and calling in and now sent to a queue that had them holding for several minutes, they became even more aggravated. So what you were trying to solve, I don’t think was actually being done well, but it was the best we had back in the day.
Max Clark: [14:06] Mm-hmm. Mm-hmm.
Max Clark: [14:17] I think the perception of context or the problem in the context or versus like the actual experience of a company with a context center are very disconnected, right? Like the, and what I mean by that is, A contact center gets stood up to do something, to solve a function. And loosely speaking, that’s either like sales or some sort of support or service function. So a company that has a contact center doing sales wants to facilitate revenue and transactions. And then a company that has support, and they get very intertwined very easily. a support function is a sales function, and a sales function is a support. And that line gets blurred really quickly.
And so this thing of like running and having a contact center that’s become this like, you know, horrible word is really very, it’s very core to a lot of companies operations. Like, you know, you can’t have a business at scale without having some sort of interaction with your customer.
Michael Cibelli: [15:20] Yeah. But when you do what you just said, which is you have sales and you have support and they’re intermixing and we’ve all experienced this. I started here, but then I ended up having to go here. The context of that call didn’t follow and I have to start all over again. And it, it just destroys the entire complexion of that contact center. That is not the way to handle a customer and handle them well.
Max Clark: [15:31] Mm-hmm.
Max Clark: [15:37] Yeah. Yep.
Michael Cibelli: [15:50] The way to handle a customer, in my humble opinion, is by allowing them to be serviced in the way that they want, regardless of what they’re asking to do. So if I’m calling in and I say, well, I’m calling because my order is not here, right? Great, I can help you with that. Or I’m calling because I want to return something.
Max Clark: [16:12] Mm-hmm.
Michael Cibelli: [16:19] Great, I can help you with that as well. I’m not being moved around in our space, regardless of what that customer is calling for, because the virtual agent is purposely built to handle multiple transactions of many different types. So we’re doing business on the grounds in which you want it served and not having all these little different fiefdoms of contact center people that handle specific tasks. It doesn’t work that way.
Max Clark: [16:48] So that’s a good segue. The current state of AI for Contact Center really feels to be, initially with technology, have silos. And then the silos, like specific functionality, and then the silos kind of merge together into platforms. You look at anything, and it’s like we do this feature, and then we expand our capabilities. And Contact Center really started that way as well, where it was you know, these verticalized approaches of like, we’re going to do call recording analysis, right, your quality control functions. That way you get to, you know, from 2 % coverage of calls to 100 % coverage and from having to have a person, a compliance officer listening to all your calls to being able to actually flag things that need to be reviewed, right? So you’ve got to, there was like a compliance and then there became analysis and then you see agent assist. you know, something running the background with the agent on the call to surface knowledge-based information or prompt the agent with what they need to make the agent more efficient. You know, and then there’s like this, like the holy grail for some people becomes the deflection, you know, this idea of like how, what is your actual staffing requirement, you know, within, and of course becomes the hardest problem to solve, right? Because now your stakes are much higher, you know, like the, quality control is relatively low stakes to actually interacting with a customer where you can damage a relationship becomes very high stakes. So how is that maturing across? mean, am I completely off base, first off, I should ask? And do you look at this differently? And how is this maturing and evolution taking place?
Michael Cibelli: [18:34] You know, you’re not off base at all. What you’re saying is very true. And I’ll tell you, it’s evident because I’m seeing a lot of the companies that came out to work on QAQM as an example, are now trying to cross over and come back into our space, which is virtual assistants. The answer is why. I mean, the question is why. Well, a virtual assistant doesn’t need QAQM.
It’s handling the calls in the method at which it needs to handle them each and every time and doing the same thing over and over and doing it very well. If I have varying degrees of agents that have different skill sets, they may need some assistance. They may need that prompting to ask a particular question. On the virtual agent side, it handles the same thing every time, the same way, with the same demeanor, the same empathy. with the same excitement, it changes according to what the question is. And it does so in a way that is moving away from, I need to do this, this, and this. No, I can do whatever you need to do because we can continue to build that pace of questions and experiences that that virtual agent has to handle whatever flows their way. That’s a big, big change that we’re seeing. You mentioned the word deflection. I don’t want to deflect.
I want to contain. I want that call handled by that virtual agent in as many possible scenarios as it can. And we’re doing that with some companies up to 90 % of the time. That is a big deal. And we’re doing it, not just handling the call, but handling it well to a satisfactory conclusion that the customer expects. And when you can do that,
Max Clark: [20:00] Mm-hmm.
Michael Cibelli: [20:27] That changes the game. And that’s exactly what we’re doing.
Max Clark: [20:32] Doesn’t that lead into then the AI is coming for your jobs boogie man conversation, right? Because then this becomes the big fear. And there’s a lot of fear and hype, like negative hype cycle around AI in general, which is like, AI is going to take everybody’s jobs. And what’s the reality of that?
Michael Cibelli: [20:50] Yeah, that’s a great point. And we hear that all the time. The fact of the matter is there are many things that AI can handle that can, excuse me, that can be incredibly mundane for a contact center agent. For example, FAQs, identification and verification, some very simple things that we can do over and over and over again. That leaves the live agent there. to handle some very specific tasks that need to be done that we may not be able to handle or that the customer wants to talk to a live agent about. In those instances, we will pass that off to a live agent and send the context with it so that they can handle that need. Now, as I said, we contain a large number of calls.
Give you an example. In a contact center where you may have 200 agents in the contact center, if I’m containing 40 % of those calls, maybe you need 140 agents in that contact center. Right? And is that going to take some jobs away? Yes. The key though is will it repurpose those employees to do other tasks within that company that are far greater than answering an FAQ? That’s where I see it going and that’s what I think is happening. I don’t think we completely replace the contact center agent. I think we augment the contact center agent to allow those in the contact center to handle other complex tasks that we may not otherwise do or they don’t want us to do, or they repurpose them to other roles within the company.
Max Clark: [22:43] So the elephant in the room or the dirty secret of this that we can talk about a second, let’s just say domestic US companies, right? A domestic US company, especially at scale, has already shifted a significant percentage of their contacts and their labor globally into BPO’s. And when we talk about organizationally within that within that context in our environment. And what I see very commonly is a skills-based hierarchy of the agent that becomes, we’ll have an entry point into a BPO-driven environment. And then it’s a certain skill or a certain routing tier goes into our US team, which becomes our advanced team and our sophisticated team. And maybe that’s tier two. And it might be a tier three or whatever these different things are.
When we have this conversation, we start talking about reallocation or upscaling of labor. Don’t spend time on FAQs that are repetitive and burn out people. What percentage of this really impacts the domestic contact center at this point versus the BPO? So when people are like, AI is coming for your jobs, my reaction to that becomes, OK, plausibly, but which jobs are really coming for and who’s really at the highest degree of risk here?
Michael Cibelli: [24:16] Interesting question. I think from my perspective, I can tell you I’m getting calls weekly from BPOs that are saying, Hey, I need some help. You guys are replacing my agents and I need to augment that with the ability to sell your service.
So in fact, NGOs are looking for us to do that front end work of those mundane repetitive calls so that their agents don’t have to handle them. I think that’s where more of it’s being hit than a US based agent that may be that second tier that’s taking other pieces, right? So I know for a fact, these guys are calling me on a weekly basis saying, hey, can you help me out? Like, can I sell you a service? And frankly, what they’re seeing is,
Max Clark: [24:39] Mm-hmm.
Michael Cibelli: [25:09] that the margins in the service that they could earn by having a virtual agent handle many of that is greater than them putting butts in seats. It works for them too. It’s just a matter of how they augment that, just like anyone would in their own contact center, right?
Max Clark: [25:29] Because then you have the underlying issues with most contact centers become hiring, training, retention, and churn of your labor force. So if you’re running a large contact center and you’re dealing with churn statistics of double digit, like 20 to 30 % churn, mean, that’s a significant problem that you have to deal with across that contact center.
When you say like BPO is talking to you and asking about reselling or selling your service, it’s interesting because it shifts the conversation a little bit in my head to outcome-driven service delivery. What’s the actual outcome or goal that? And a lot of times, especially in technology, I feel like we get disconnected and distracted from what the actual business outcome or the business driver is. you know, talk to our customers and help our customer do business with us either before the sale, during the sale, sell them something, support them after the sale, retain them, offer them new things. And when you get up to the higher levels, like you don’t really think about it in terms of like, I need to have chat on my website. It’s like, well, that’s how the customer wants to interact with me to actually do what I want them to do, which is spend money with me, right? So.
Is this like just the beginning of a bigger shift in context? And do you think of like, you know, how do we drive the outcome that we’re actually trying to get to?
Michael Cibelli: [27:02] I think it is. You know, you bring up a lot of good points there relative to retention, turnover. You said, you know, 20 % or so. I mean, in many of these companies that we talked to, the agent turnover is upwards of 80 % at times. And sometimes even higher. It’s, let’s just think about that for a second. it’s, it’s, it’s not uncommon. So it’s,
Max Clark: [27:21] Ugh, my goodness.
On an annualized basis, that’s ridiculous. That’s an absolute nightmare.
Michael Cibelli: [27:32] rehiring, retraining, how long does it take them to really get up to speed to handle things in a way that you would really want your customer experience to be handled? It takes months for somebody to get sharp enough to do that, in which time many of them are gone and you’re redoing it. It’s this constant turn. We take every single one of those situations that you mentioned and we put it into an ROI analysis for the customer, for them to understand turnover.
First call resolution, average hold time, all the things that you would think in a contact center. We build that up with their actual numbers to bring them to say, here’s what we believe we can return to you in terms of an ROI. And we stick by that ROI. So is there a change? Yeah, it’s a great change because let me ask you a question. How many times have you talked to an offshore contact center agent and said, wow, was awesome.
Max Clark: [28:35] Um, not a lot. Very few, probably. Yeah. You know, very few. And it’s really obvious, No, I mean, it’s also very obvious. It’s like, it’s not even, you know, I had, many years ago, I had a business based in El Segundo, California. And if you’re native to Southern California, you pronounce it El Segundo.
Michael Cibelli: [28:40] So you’re putting you’re putting Band-Aids on an open roof. Yeah.
Max Clark: [29:04] And if you’re not native to California, mean, if you have any Spanish background, you still say El Segundo. But if you’re not, even within the United States, it comes out immediately. Where are you talking and where are you going? And you get the accent, the inflections come across really quickly. so yeah, it’s really obvious. It really stands out. think I’ve been in the industries long enough that you know, understand what’s going on in the background, but I know a lot of people don’t really appreciate that. And you see like a lot of collar rage that comes out, you know, it’s like, who are you and where are you and why am I talking to you? And, know, I saw an AI demo recently for a company that their entire focus was doing, accent masking for human agents. So that way they could mimic, you know, local accents, regardless of what the agent actually spoke. And I found that very interesting in terms of like, know, problem to be solved for. But yeah, it’s a good point. It’s interesting.
Michael Cibelli: [30:07] You know, you bring up several things that I’d like to touch on here. So you can accent mask and it helps. It’s, it’s a good thing, but regardless of how you say it, you’re still handling the call the same way in terms of experience and knowledge and all of that. What we do. And I think this hits to exactly what you’re talking about and how you say El Segundo, right? Which is on. brand virtual agent. So we work with the company to ensure that the accent, depending on where they are in the country, how they want it to sound. If I have a company based in Boston, I don’t want them to sound like I’m in Texas, right? So we work very, very diligently to provide
Max Clark: [30:56] Yeah.
Michael Cibelli: [31:03] the exact representation that every brand wants. Whether do you want a female or a male to be answering this call? Do you want an accent from a particular part of the country? Do you want to hear keystrokes in the background? Do you want… I’m hating. Some go, I want to hear the keystroke. I want it to sound like they’re putting in the order. Whatever. You can have anything. But most importantly, it’s on brand, right?
Max Clark: [31:17] I hate that. Stop.
Max Clark: [31:32] Mm-hmm.
Michael Cibelli: [31:32] And so when you have a marketing organization that says, this is how we need our brand represented. How many of those actual reps are going to represent your brand exactly as you want it to the virtual agent does it on brand every time in the same way and lets it, I mean, it, it doesn’t change.
Max Clark: [31:52] Scaling? Scaling, a consistent experience, is an incredibly difficult challenge for any business to undertake. I want to ask you about training and LLM-specific stuff. Pre-LLMs, when you were talking about, again, AI and air quotes, AI for chat, or AI of whatever these things were, pre-LLM, Really, you were going through and you were programming this giant flowchart decision tree nightmare of what are we looking for, what can be asked, how do we want to respond. And that was shocking for people going through it the first time because they had this sales presentation that became like, yeah, we can just turn on this chat agent. Everything’s wonderful. And then you got into actually programming the chat agent. You’re like, what just happened here?
How do we get this thing live? What do we have to do? How much my team has to program this thing? Now, LLMs change that game a little bit because you can ingest. But how do you take and ingest and evaluate knowledge for a company? and decide what knowledge is important and what knowledge is relevant and what knowledge is out of date and what knowledge needs to be updated. Like what is that life cycle like?
Michael Cibelli: [33:21] That’s great question. So one of the things that we do is we deploy multiple LLMs in our environment. So first off, we have our own LLM. That was created over nine years ago. Nine years ago in AI is a long time. Our LLM was specifically built for CX.
So there is a ton of CX knowledge and information in our own LLM. We then use others, that many of the names that you know. We then import all of the information from various databases that the prospective customer has into the knowledge base that allows us to pull things for our LLM, et cetera. So it’s a mix of many things that we do. that dates back again, nine plus years. So it’s something we’ve been doing for a really long time in this space. And I think our dedication to CX and how we handle that with these multiple LLMs makes a huge difference in the information and knowledge, the conversation that we can enhance in any customer service experience.
Max Clark: [34:43] We’re seeing this a lot right now in development pipelines. And I think enough people interact with LLMs to experience hallucinations, where it’s like you ask an LLM the same question multiple times, and you get different output from the LLM. Or you get a completely made up answer, because you can push the LLM into like, I want an answer from you. And the LLM just goes like, OK, I’ll give you an answer, even though the answer is completely made up.
And this is a big issue for development teams doing coding, AI-assisted coding development, because it’s getting at different output. And when you start talking and looking at this in terms of actual, again, brand-impacting customer experience interactions, how do you protect against that and maintain consistency of response and also protect against hallucinations with some of these LLMs?
Michael Cibelli: [35:40] Yeah, there’s multiple layers in which we deconstruct a particular response from a customer. We strip out a lot of noise. We make sure everything is as crisp as can possibly be and that the questions that are being asked are relevant to a CX experience and what we’re doing.
If in fact we find somebody’s trying to play a game, if you will, or challenge the virtual agent, we can pick that up very, very well. In fact, when we do these deployments, we put a CSM on the back end of each customer. That CSM does a tremendous amount of monitoring to see what’s going on.
That monitoring can catch many of those things, but can also do things as simple as, let’s say it’s a healthcare deployment and a particular physician has a very difficult name and you’re having trouble picking up that name. We’ll notice that and we will then reprogram things to make sure that that name is clear and understood regardless of how it’s said.
So I would tell you that there’s continuous monitoring that goes on so that we understand the spoken word. What is truly aligned with what we’re trying to solve here? What isn’t? What could be just an interpretation that’s very different in terms of naming nomenclature and different things of that? So it’s a constant evolution. These things are not a set and forget experience. They are constantly moving.
Max Clark: [37:32] Is verticalization important when you start talking about deploying AI into a contact center? We’re focused on banking. We’re focused on health care. We’re focused on transportation, airline. Because the more data you get across a variety of customers feeding into an engine, the more refinement you can make in that engine. And there’s commonalities between all these things. But know,
Michael Cibelli: [37:34] It’s living breathing.
Max Clark: [38:01] banking and financial services needs different things than, I mean, you mentioned healthcare earlier, like a hospital is just going to have different stuff that they have to apply. How important, I mean, how important is that verticalization or specialization, or is that already a solved problem at this point?
Michael Cibelli: [38:11] Yeah.
Michael Cibelli: [38:17] Now, I think you can get better and better with some verticalization. Let’s take health care as an example. There are words and phrases that are used in health care that you’re not going to use in retail, right? So having a better understanding of those is going to make your virtual assistant, I think, more aligned with what’s trying to be solved within a particular vertical.
Max Clark: [38:34] you
Michael Cibelli: [38:47] There are several verticals that we really focus on. Healthcare is actually one of them. It is our fastest growing vertical. Our largest vertical is in hospitality. But there are a variety of things that I think you can gain experience and superiority with the more you work with them. Just as an LLM gets more experienced and sharper as time goes on,
Max Clark: [39:07] Mm-hmm.
Michael Cibelli: [39:15] So too do the tweaks that you make within a vertical. And we’re very focused on what those are to make sure that we have the actual nomenclature that’s common to that particular vertical in our vernacular and how we respond to things. because you know, if you’re in that and somebody’s talking your language, you as a customer feel more at ease. So if you ask a particular question in a vertical and I respond with words, that are only common to that particular vertical. makes you more, yeah, absolutely. Yeah. We’re seeing that more.
Max Clark: [39:48] inside language.
Max Clark: [39:51] Well, this goes back to what you said earlier about understanding, Like this is not only having under, you know, not only having inside our language, but actually having context and understanding of the actual, you know, what’s the like, I don’t want to say how many, how to phrase this, right? It’s like, what’s the actual root of the interaction that we’re probably going to get to, right? Like there’s, you have to have context and you have to have awareness and understanding. And so that becomes really important of the… know, creation and then deployment when you start talking about using AI and any sort of interaction as well, right?
Michael Cibelli: [40:25] Yeah, let me give you a really good example of that. Let’s take something very simple when it comes to any conversation. You call in and I ask you for your account number and you say, my account number is ABC one, two, three. And the next person calls in and says, my account number is alpha beta Charlie 123. You both said the exact same thing. You said it completely differently, right? So what we.
Max Clark: [40:41] Mm-hmm.
Max Clark: [40:52] boy, yep.
Michael Cibelli: [40:54] What you have to think of is, OK, what did I hear? So this is how we solve for that. And it’s pretty darn unique to us. We have what are called ASRs, Automatic Speech Recognition. We have our own ASR. And most companies will put an ASR against something like alphanumerics, which we just experienced in ABC 123. What we do is we say,
Max Clark: [41:20] Mm-hmm.
Michael Cibelli: [41:23] All right, we’re going to take multiple ASRs we’re going to put them in the background. The multiple ASRs are going to hear what was said, ABC123, alpha beta Charlie123. It goes to those ASRs. They listen for what was said and they come back to us and say, this is what I heard. We take the aggregate of what was said, and then we look at that and say, what was said was ABC123. That’s a big deal because as soon as I say, I’m sorry, I didn’t get that.
The next thing the customer says is representative.
Max Clark: [41:58] Yep. Yep.
Michael Cibelli: [42:00] Give me a human. That helps us contain that call. Again, we go back to that containment thing. I want to contain that call. So off the bat, I now get the ABC123 account number. Great, I have your account number. How can I help you? I’m not sending you down a tree. At that point, you’re gonna say, I’m calling for this. awesome, yeah, I can help you with that.
That order that you’re calling in about that’s associated with this phone number you’re calling from or from that account number you’re talking about is due to be delivered between today between 10 and two. Would you like me to send you a text with that information so you can track it? Yeah, I’d love that. Sends the text, off it goes. I’ve contained that call. I’ve handled it from start to finish. That’s the experience people are starting to see on a regular basis. And thanking… the virtual assistant at the end of the call because they either didn’t know they were talking to a virtual assistant or while we’re not trying to fool you, we’re just trying to make you comfortable in talking to a virtual assistant and let it handle your issue. In which case it did. You said thank you. And somebody did the virtual assistant. You’re welcome.
Max Clark: [43:14] I mean, I’ll just use myself. The frustration doesn’t come from talking to a person or talking to a computer. It just comes from how much time do I have to waste or do I perceive I’m wasting in the interaction. And if you get transferred to four different person to person to person to person, and you have to repeat your authentication, your context, every step of the transfer, by the time I get to the fourth person, I will tell them this is not about you. You’re my fourth person I’m talking to, and I’m really not in a good place now. You know, it’s like, but I’ve learned also with myself to like preface that a little bit, just be like, I am sorry, this is going to go really poorly because I am sorry. My, my blood is boiling. You know, like it’s, it’s, it’s just like, why are we doing this? You know?
Michael Cibelli: [43:54] I think we all do at that point. It’s like, feel so bad. I feel so bad. Excuse me. Ripping that person on the phone when it’s not their fault. They’re just trying to help me or do their job. but you’re so heated at that point that you’ve just had enough. How about this one? How about you’re in a chat session with a live agent that’s doing the chat behind the scenes or it’s handling multiple chats and you type, type in what you want.
Max Clark: [44:05] Yeah.
Max Clark: [44:10] Yeah.
Michael Cibelli: [44:23] and you sit there and you’re like, okay, well, you’re gonna answer me sooner or later. I’ll just sit here and wait. Today’s experience is I wanna be handled and I wanna be handled.
Max Clark: [44:31] Yeah, other dirty secret.
Michael Cibelli: [44:40] And the way to do that is by having an experienced agent that can answer your questions and or need and solve it. Our virtual agent do that and do it quickly.
Max Clark: [44:55] We haven’t talked about this at all because we’ve been focused on voice. the example you just gave is really irrelevant, right? Because in a human-driven voice interaction, you have one-to-one. You have one agent, one caller. And as soon as you switch modalities into some sort of textual modality, the default is to increase the density, right? How many chats are being handled simultaneously? What’s your ratio? Two, three, four to one? Like four chats with one agent? And.
And then you’ve got context switching on that agent because that agent has to keep track of four different conversations at the same time of what’s going on. Right. And, you know, like you just highlighted the, the end use, you know, the, the, the customer’s interaction with that becomes a lot of time between I’ve, press the send button and how much time does it take to respond and are they paying attention to me or are they paying attention to something else? And why is it taking so long? And then if I don’t keep that bloody window open, then You know, the chat times me out because you know, I went to do something else or I got distracted with my email or some YouTube video, right? And they responded. I didn’t hear the beep. And then they’re like, you’re not here anymore. Goodbye. You know, and they, they give you like 10 seconds before they turn them out the chat. And then you just look at that, you know, and just like, okay, you know, here we go. No temperature, temperature.
Michael Cibelli: [46:10] Yeah, know, let me talk to you a little bit, if you don’t mind, about a use case that I think can demonstrate what really can be done with a virtual agent in this instance. We had a particular customer come to us, Carter Zaskos, which I’m sure you’ve heard of, the children’s brand.
Max Clark: [46:22] Yeah, let’s get into it.
Max Clark: [46:33] Mm-hmm.
Michael Cibelli: [46:36] Couple of years ago on Thanksgiving and Black Friday, had 400 people plus in queue, inordinate amount of wait time, frustrated customers, frustrated employees because they were getting that next customer that was ready to blow up because they were in queue for so long, right? And so between Thanksgiving and the new year during that,
Max Clark: [46:54] you
Max Clark: [46:58] That’s bad for everybody,
Michael Cibelli: [47:06] that hiring time, they had to hire an additional several hundred associates. As you can imagine the quality of the first 10 you hire versus the last 10 you hire. How many of those also turned over before they even got into the seat because they went through the training and then, but you understand all that. after the matter is they had multiple people on hold and inordinate wait times. had three contact centers. One is in Canada.
One is in the US, one is in Mexico. By May of that following year, we deployed the virtual assistant. The virtual assistant was containing so many calls and handling them so well that by the following Thanksgiving and Black Friday, the US team had off to spend Thanksgiving and Black Friday home with their families, and they handled any overflow with zero wait times through Canada and Mexico.
That’s an experience and a change for a company from both a CX and an EX employee experience situation. Cause I now allowed my employees to be home and spend Thanksgiving and Black Friday with their families of which if you’re in that space, you never get that off. And the customer experience was increased tremendously because there was nobody in, in, on hold and everyone was handled within.
Max Clark: [48:19] Mm-hmm.
Michael Cibelli: [48:29] time frame that they want because regardless of the number of calls that come in, can handle them. I’m handling them with the virtual assistant. It doesn’t matter. The scalability.
Max Clark: [48:40] Let’s talk about this deployment because there’s a lot now that I want to ask and talk about. So you’ve got an event. So you’ve got to underline pressure on the business, is seasonality, which causes staffing issues. And so the old way was like, have to WFO, WFM. We have to try to figure out our statistics and how much are we going to hire. And then you’ve got to front load it, hire a bunch of people, train them, to your point, people that are hiring into seasonal seasonal jobs, right? Like on both sides, it’s just, you you’ve already like, you’ve already shot yourself in the foot before you’ve even started, you know, in this situation a lot of ways. And, and then you’ve got to train. now if you’re getting ready for, for, you know, black Friday, right. And then the Christmas rush, you know, you’re probably starting your hiring cycle 120 days before that, because you’ve got a identify hire, train, bring everybody in, right? So then you go backwards in time. I think about this in terms of like months, right? And calendars, quarters. So your hiring cycle starting, you know, probably like in August, I would imagine, to like try to prepare for this thing, which is gonna be a two month long, you know, probably span of time that you actually need to have access capacity. And you said they came to you in May, right? So they, you know, got,
Michael Cibelli: [49:58] nothing came to us, you know, what we deployed in May, they came to us right after it happened.
Max Clark: [50:04] So OK, so right, like wheels fell off the bus and they said, got to make a change. They come to you December, January. A four month rollout, mean demo, evaluation, contracting cycle, proof of concept, all those different things. I’m immediately like, OK, the fastest this is probably being contracted is end of January, probably February. And so you’ve got like a three month go live from that point?
Even condensing the contracting cycle, that seems incredibly quick to me.
Michael Cibelli: [50:43] Yeah, I can understand why you would say that. I started at Poly AI two years ago. And the sales cycle has compressed tremendously in that timeframe. Initially, it was a nine to 12 month sales cycle. And then on top of that, a deployment and all the things that we were seeing. Now, I think customers are so open. to this change and realizing the value of the change. And frankly, the ROI that we’re producing is so crazy that we’ve saw sale cycles last quarter that were as short as 30 days and 45 days. Again, not the norm, not every one, but we are seeing compressed sale cycles as a result. In terms of The deployment, which I think that you’re, saying on top of that seems to be short. I think it’s because many of the deployments that we’ve done, and we have so many use cases now in a variety of, of verticals and, and use cases, use case specific that are very different. We know what the heck we’re doing and we’re able to do it pretty darn well. So we’ve compressed not only sales cycles, but deployment phases, but not compressed them to the point where.
It is causing too early of a deployment and a lot of fumbling. We will not do that. We make sure that that deployment is done correctly and it is done to the customer satisfaction because they’re listening to the calls. They’re hearing the plays. They’re hearing the responses. They know exactly what we’re doing through the whole piece. And then as I said, it’s not a set and forget. We stay with it. Are there any words and phrases that we’re not catching? Are there things that additionally have to be added? at the back end of it, is there information that we’re seeing come through that is allowing us to take all of that knowledge that we’re getting now and look back and say, hey, here’s something else we should do. All of that’s happening simultaneously as that’s all going on. And it’s allowing us to do these quickly and efficiently and very,
Max Clark: [53:02] I mean, what’s a typical milestone? mean, you know, a lot of times, I mean, I’ll use the expression, you know, crawl, walk around. didn’t invent that expression, right. But like, you know, like, how do you, how do you dip your toes into it and roll this thing out and get into, like what’s like the logical do this, then this happens in this, I mean, I can’t imagine, right. We’re just, everything’s turning on day one and we’re just going into like, you know, that would, that would be really impressive.
Also from a customer mentality of just like risk, right. And understanding and measuring risk and trying to moderate risk. But you know, what, what are like a typical deployment milestone and how do people approach this thing usually.
Michael Cibelli: [53:41] Yeah, that’s great question. And you’re right. You can’t turn on the faucet full blast on day one and expect that everything’s going to be perfect. So what we do is when we deploy, we do it in various percentages. As an example, once we get the OK, we will push about 10 % of the calls into it to allow us to then monitor those 10 % of the calls. How are we handling them?
Are we containing the percentage that we said we would? If not, why? What has to be tweaked? And so on and so forth. So you start with a smaller percentage. Once you feel like you have that buttoned up, which I mean, we had one last quarter that we did where on day one, we turned up 10 % and 84 % of the calls were being handled. The customer almost fell over. They’re like, I can’t believe that you guys are containing 84 % of the calls. We did that in that small portion. We then turned it up to 25%. You then turn it up to 50 % and then you finally let it flow. In that instance, those calls never varied from 84 % to 86 % of calls being contained. So you do it in small increments until you get it to where you really want. Now, once you do get that deployed, in that fashion and it’s humming along at those percentages. It’s not over. Like you’re on that back end looking at analytics for other things that you want to do. You asked how you do it. those are from percentages. The other thing you do is you take on a use case that is not the most difficult thing in the world, but many customers start with things that I talked about earlier, FAQs, ID and V, identification and verification. You do some
Max Clark: [55:38] Mm-hmm.
Michael Cibelli: [55:38] some more simple things in the beginning, and then you start to deploy things later that could be more complex. As an example, in the Carter’s instance, when we turned up and we did some order confirmations and things, or cancellation, whatever, we then moved into more difficult use cases. Carter’s has a particular policy that allows someone to return an item or an or to, I’m sorry, not return, but to get a rebate on an item that may have lowered in pricing over either 10 or 14 days. I forget exactly what it is. So let’s you ordered a particular item for a child and it had reduced and you found that out and you said, okay, I’m going to go in and get that price matching. So I’m going to go back and get that. They were processing those with live agents.
Max Clark: [56:19] Okay, yeah, yeah, sure price matching, yeah.
Michael Cibelli: [56:37] and giving people back money. You’re talking a double whammy in terms of expense to that business. It was a seven step process that attached to multiple APIs in the background that enabled that particular agent to look at the dates, the price difference, do the calculation, refund it back to a credit card, all of the above seven steps. And it took them quite a bit of time. We were able to solve that for them. with building a second cart. And then we would look at all those things and made sure those matches where we did the calculation, we funded back to the credit card without ever touching a live agent. That’s more difficult, but you start here and then you move into those more difficult scenarios, which is what we did with them. And it, it’s significantly to return more ROI to them that they were already realizing it was paying for itself on the front end.
And on the first use case, and now you’re doing more. Here’s another good example. This one blew me away because you don’t think that you’re going to do things like this for a customer and change the way that they’re thinking about doing their business. We have a particular customer who we do this restaurant chain and
Max Clark: [57:39] Mm-hmm.
Michael Cibelli: [58:03] We did reservations. Not very difficult for us to do. We have the ability to back end into things like OpenTable, do reservations, verify reservations, do all those things. I told you we put a CSM on the back end of many of these deployments, on all these deployments. And what they saw was that the vast majority of customers calling in We’re calling for two top tables. The CSM had the wherewithal to talk to the powers that be and say, look, you guys, are getting the vast majority of tables requested are for two tops. Your restaurant is built for four and six top tables, the majority of the way. How many more tables could you fit in here if you did two top tables? They said a lot. They said, let me tell you why we would suggest that.
Because if 60 or 80 % of the people calling in are requesting two top tables, you’re not filling your restaurant to capacity. They listened, they changed their tables out, and it returned an incredible revenue stream for them in nine months of deployment, upwards of $4 million in additional restaurant reservations. That’s changing the complexion of a business. That’s returning an ROI.
They are blown away and they are a customer for life when you start doing things like that for someone.
Max Clark: [59:35] That’s the part of this that I get the most excited about, right? When you start talking about how much business intelligence is, a lot of advice is talk to your customers. Are you talking to your customers? What do your customer needs? How do you, and it’s not like CSAT scores. Is your customer happy with the interaction I just had with an agent? It’s like, what does your customer actually need from you?
And being able to end and like how much information is actually available in your interactions that you’re having with people that just isn’t being captured because there’s no way of doing this at any sort of scale. And how I mean, that’s a really good example. It’s just who would have the time to have a business analyst and dig through all the reservation details for the preceding six months and figure out you know, like, we’ve got a, a, a table configuration density issue that we need to change versus just being able to say, Hey, look, you know, we happen to know from, you know, six months worth of phone calls that this is, this is occurring in your business. It’s, I, I,
Michael Cibelli: [01:00:49] Yeah. It’s also something you can get in the old IVRs because remember in an old IVR, after you went from press, press one or press two, it went to in a few words, tell me what you’re calling about. Right. We don’t say that. We say, hi, thanks for calling ABC company. How can I help you today? Well, I’m calling because I need to change my appointment. The reason I need to change my appointment is because blah, blah, blah, blah, blah. Our initial
Max Clark: [01:01:02] Hmm.
Max Clark: [01:01:16] Mm-hmm.
Michael Cibelli: [01:01:19] Ask from that customer in an open conversation gathers way more information than somebody giving you a few words. It gives you a lot of context as to what I’m calling about and why I’m calling about it. And that’s not what you get in most of the, well, in any of those that you’re pressing or saying a few words for. It’s just an open conversation. So people tend to go on and talk more than they normally would. And it gives us a lot of good information that allows us to then come back to you with those suggestions on things to change. For example, if you said, let’s start with FAQs. And the next thing I think I want to do is billing inquiries. What we may tell you is, after we do the FAQs, what we’re seeing in our analytics are more for order changes or order cancellations than it is for billing. Therefore, You may want to reprioritize what our next use case is going to be because you’re going to get a better bang for your buck that 40 % of your calls are coming in for this, where only 10 % are coming in for your billing that you thought was the next thing you wanted to do. It’s not my gut tells me this is what I want to do. It’s the analytics are telling me this is what I want to do. That’s a big difference.
Max Clark: [01:02:41] When you go into, as you just said, when you go into an open dialogue prompt, what can I do for you? And you’re not immediately trying to direct a flow. People are going to give you a lot different information. So then, of course, you have to have understanding and, as you put, ASR and context and all these other things. But ultimately, you’re driving towards this number. How many calls are you containing? What?
What causes a call to break containment? What’s the actual, is this like, you cause you know, Carter’s example, they, you you’re saying, you know, they start here and then they slowly kind of build into additional use cases, right? So like, is, is the containment that you just haven’t configured a made a use case for it yet, or haven’t, you know, programmed against the API to say, Oh, this is a, you know, not a rebate, but this is a pricing adjustment because we were running a sale and we’re to give you the money back, right? well, until you have the API built, you can’t execute that with, with your AI, right? So you have to, but what, but then at some point you’ve got these, you know, in theory, you’ve got all the different business expect, you know, cases built. what breaks containment at that point? And then how do you continue to refine or adjust or, or, you know, increase that ratio of calls that are being contained and handled.
Michael Cibelli: [01:03:59] Yeah, I think it’s exactly as you stated there, which is, you know, look, there are certain things you can solve for and there’s certain things that you didn’t prepare to solve for just because it’s something that you’re looking at down the road. So those are going to break contain. And that’s okay because that’s where I, as we said earlier, I have the live agent there for something that’s going to break contain. It’s just a matter of how often does that happen? want to reduce the number of those happening. Now, the more I understand what your particular customers are asking for, the more use cases that can be built that can further enhance and contain. So without the analytics and without the understanding, without the monitoring from a CSM and folks on our teams, you wouldn’t get that. So we continually do that to help force that contain as much as possible, because frankly, the more we contain, the more value we bring to you, the more ROI we bring to you.
And that’s the whole mantra that we stand on is how do we retain return more to our customers each and every day? And what can we do to make this even better? How does it sound better? How does it listen better? All those things that never stops. Now, I mean, there are situations where you’re going to break contain where as an example, as good as it is, if somebody, just can’t understand because of the way that they’re saying things like, there’s just no way. It’s like, I’m sorry, there’s just no way I can get that. But, and then that would break contained. But those are few and far between again, because the ASRs, the multiple ASRs and the things that we do are so darn tight in terms of listening and understanding. But there are those instances and that’s exactly why, as I said earlier, there are times when you’re going to have to have a live agent. You don’t handle a hundred percent. You can make things look like you handle a hundred percent.
Max Clark: [01:05:57] Mm-hmm.
Michael Cibelli: [01:05:57] you know, when you said, about the ability to handle a call and contain, Hi, thanks for calling ABC company. How can I help you? Yeah. You know, I need you to do this. Yeah, I can do that for you. And I hang up. I contain that call according to what the analytics would say, but I didn’t handle it to the specifications of what that customer expected.
Max Clark: [01:06:21] You just, you triggered a memory for me. was at a car dealership and I’m sitting there and, sales guy answers a call and he’s on speaker. So I heard the entire interaction and the entire interaction is this caller and the salesperson trying to like, suss out from the caller, what the caller was calling for. And it was like, it was like this 15 minute conversation where like, I don’t even know what the caller was calling about. Like, I don’t know if the caller knew why the caller was calling.
Michael Cibelli: [01:06:23] Thank
Max Clark: [01:06:49] Right? There was some plan in this person’s head that wasn’t being verbalized or communicated anyway. And this, and this poor sales guy, you know, of course you have this like as a car dealership, so there’s already all these like emotional responses and like bad raps that come out, but this poor sales guy was just trying to get the guy on the call to tell them what he wanted so he could help them. You know, like it was just like, what can I do for you? and you listen, and I, and I think about that, you know, you’re talking about containment. It’s like, many of these calls like you can’t understand people are just like the person has no idea why why they’re calling in the first place like what are they trying to accomplish you know
Michael Cibelli: [01:07:26] Some just call the ram, right? If you’re calling a ram, there’s nothing you’re gonna do except for be empathetic and listen, in which case the virtual agent will do that. Look, I’m sorry, that’s how you feel, but I’m here to help you, how can I help you now? And try to redirect and push that call back or pull that call back into a conversation that might make sense. yeah.
Max Clark: [01:07:28] Yeah, that’s a good point too.
Max Clark: [01:07:44] Mm-hmm.
Max Clark: [01:07:48] interesting. So you have defined empathy, like responses, you know, built into this.
Michael Cibelli: [01:07:56] Absolutely. We have an example of a bank where the person that answers the phone is super cheery. Hi, thanks for calling. ABC Bank, how can I help you today? Well, I’m calling because my mom passed away and I have to figure out how to close out the accounts.
Oh, I’m so sorry to hear that. I’m here to help you in any way possible and make this as easy as I possibly can. That’s pretty incredible that understanding that response. But I mean, what are going to do? Come back to your ego? Oh, well, I’m here to help you. You know, it’s no changes the empathy. It understands that that’s how you feel. And it’s trying to get back into that space to where I can answer you in a way that you feel, which is, I’m really sorry to hear that.
Max Clark: [01:08:25] Yeah.
Max Clark: [01:08:33] boy.
Michael Cibelli: [01:08:45] but I’m here to help in any way possible. I that’s pretty cool.
Max Clark: [01:08:50] And one of those phone calls ends up on the internet. The other phone does it. I mean, that’s, you we, we, touched on this briefly, right? Like with Carter’s and being in processing credit card refunds, and we talked about healthcare a little bit, you know, one of the challenges in a lot of contact centers becomes these like, you know, confidential data input, output things that take place, right? So with credit cards, you start talking about PCI compliance and are you processing credit cards and taking payment data? And then what does that mean for your system? Are you doing call? quality control and recording, do you have to do call masking and how you trigger call masking and field input? With healthcare, you’re dealing with PHI and what does that look like? how does deploying AI and virtual agents change the game for companies that are already dealing with this problem or inside of these regulatory regimes where they have a lot of very strict compliance requirements that they have to pay attention to?
Michael Cibelli: [01:09:50] Yeah, it’s a good point. I think one of our advantages is the fact that we were founded in the UK. So GDPR was built into our platform. The standards on GDPR compliance are pretty strict. It’s pretty tough compliance. So we have that, but you have to have all those different guardrails up that allow you to maintain all of the…
Max Clark: [01:10:05] Mm-hmm.
Michael Cibelli: [01:10:15] the things that you need to ensure that something is safe and secure. And we do that with various compliances, whether it again, it be GDPR, SOC 1, SOC 2, PCI, all those things. You have to mold all those into your ability to take care of customer data or where does the data reside? Can you keep it local? Can you?
Can you compartmentalize it? Can you only allow certain things to be accessed? Those are all the things that we take into account and ensure that we do comply on that is something that we heavily, heavily lean on again, because of where we grew from.
Max Clark: [01:10:55] So you mentioned earlier, the world of AI is evolving very quickly. And the pace of iteration is just unbelievable to witness. What’s next over the next 6, 9, 12 months? What should we be getting prepared for? And what becomes normalized?
Michael Cibelli: [01:11:19] That’s the toughest question that you’ve asked today. And it’s because we’re scratching the surface in this right now when it comes to AI. As you probably hear all the time, oh, AI is in everything. And you probably don’t even know many of the things that it’s in. And while that’s true, it’s only going to grow in usage and in how it’s deployed.
When you say, what are we going to see in the next nine months? Are you saying specifically around CX or do you just mean in.
Max Clark: [01:11:53] Yeah, now let’s talk about CX specifically.
Michael Cibelli: [01:11:59] I see greater complexity of use cases because that’s always the way it typically goes. So like I said, simple things like FAQs and ID and V and checking my order, WISMO are very simple. I think it’s going to get more complex in the things that it handles without question. as our LLMs get more sophisticated and stronger and more knowledgeable, it’s going to allow us to do many of those things. I also think that you’ll see what typically happens in areas is you’ll see it pushed down into from enterprises into mid market into SMB and you’ll start to see some of these things being able to be utilized into smaller companies. Because obviously once you start to deploy and you start to realize the return on your investments as a company, you get to push that down into other people. We’re already seeing that in our space as an example in hospitality or restaurant reservations that I mentioned to you. Because the connectors are all built and it’s what we have is pretty darn exclusive into OpenTable, we’re able to take what is a very simple use case like a restaurant reservation and now have all that pre-built to where it’s not a large PS fee to rebuild that for each and every customer. because I can simply change what I’ve already built to be a particular name of a restaurant. I change the voice. I change how they want that call specifically handled, which for a restaurant reservation, there’s not much to change. And I can start to push that down into others. I think you’ll see some of that. So higher sophistication, would say, in terms of use cases. And then in some of the easier use cases, the difficulty in deploying that
Max Clark: [01:14:02] Right.
Michael Cibelli: [01:14:18] and pushing that down into smaller enterprises is there.
Max Clark: [01:14:24] Last question for you. What is one thing that you wish more people understood about AI as it relates to CX?
Michael Cibelli: [01:14:25] Mm-hmm.
Michael Cibelli: [01:14:40] You mentioned earlier that there’s the fear of AI just replacing all the people. And I think that I would love for people to understand that everything is not an all or nothing. It’s an and. So in other words, you’re going to have deployments in AI that yes, it’s going to take some jobs because we’re able to handle things efficiently. The key is Can we take those associates and put them into other things that are going to be more sophisticated, more useful than handling an FAQ as an example? I really wish people would understand we’re not coming for your job. We’re coming to augment your job to enable you to do other things. At least a percentage of those people. And I think that’s kind of where my head is at.
The other thing is there’s a lot of fear that I hear from some folks around AI, not so much in CX, but in other areas. AI can be very dangerous in the hands of certain people. And I do believe that that’s true, depending on the guardrails and how it’s handled. I think from our perspective, what we’re doing in AI is to make things better for the world, not to make things more difficult or dangerous for the world. And so, you know, there’s people that do things for good, there’s people that do things for bad. I don’t like to see everybody grouped into one piece relative to that.
Max Clark: [01:16:20] Right. Technology is a tool. There’s a lot more that goes into it. Michael, thank you very much for the time. This was fascinating for me. I appreciate it.
Michael Cibelli: [01:16:27] No, no doubt.
Michael Cibelli: [01:16:33] Yeah, I appreciate the time as well. Your questions were targeted and well done. And I really appreciate you giving us the opportunity to express our beliefs and thoughts around this with you. So thank you.
