Regal Brings Generative AI Agents to Customer Calls

AI Leadership Insight Series. I'm your host, Mike Vizard. Today, we're with Alex Levin, who's the CEO for Regal, and we're having a little chat about how AI is being used to answer customer calls of all kinds of shapes and issues.

And I think part of the thing is that we hear a lot of people complaining about AI, but yet it's being used a lot. So something is definitely amiss here, and there's, shall we say, a mismatch in opinion and reality. Alex, welcome to the show.

Thank you for having me. So explain what's going on here, because it does seem that businesses are making great use of this stuff, and some customers are happy and maybe the people who are complaining are more of the exception than the rule. But what is the maturity of all this stuff at the moment?

Yeah, it's a good question. So when you run a survey with a general population and say, "Rate how you like human agents in interactions, rate how you like AI agents interactions," what you get is a few sort of interesting pieces of information. First of all, they're not that happy with human agents, so that's not the best.

They don't like if somebody is on the phone, they can't understand or can't help them, doesn't know the policies. And then on AI agents, what's interesting is very few people have actually experienced an AI agent. Mostly what they're talking about when you interview them is they're talking about IVRs or what we would call in the industry IVRs or IVAs, which are largely phone trees, press one for this, press two for that, or perhaps a little bit of intent detection and natural language sort of interpretation.

How can I help you? And you sort of answer a general thing. And I think I'm frustrated with those systems too, is what I would say.

That is the sort of standard of care in the industry today. When I talk about AI agents, I'm not talking about human agents, obviously, and I'm not talking about what people have experienced at their main retailer of choice or travel company of choice. I'm talking about true generative agents that are having a full conversation with you.

And in the same way I think the first time people have an experience with ChatGPT and they go, "Wow, that was not what I expected," the first time people talk with a true generative agent, they are surprised at what it can do. The question that annoys me the most is, oh, after I experience that, it's incredible. Why isn't every big company now using these things?

That's a sticky one, because the technology is incredible, and it is light years ahead of IVR and IVAs. I think the reality is 500 million calls is a drop in the bucket. So we've just hit 500 million calls, and that's nothing compared to what's going on in the industry.

And we have to challenge ourselves and our customers and potential customers to move faster, because the promise of customer service was always that you would always answer, you would know what the customer wanted to talk about, that you would treat them like an individual, and that they wouldn't have to explain themselves again and again. And we've been failing at that for years. So I think finally, there's a technology that is ready to actually bring that into reality, but it's moving more slowly than I would like.

So describe, if you would, what exactly is that customer experience like? Because I think everybody, they nod their heads when they hear the phrase AI and customer calls, but I'm not sure they are fully understanding- Yeah ... what can be done or could be seen.

So let's talk about it. So the AI agents are not yet indistinguishable from humans. We're probably 18 months from truly you not having no idea whether you're talking to AI or humans.

So we're not far. On some things today, the AI agents are better, on some things worse. And when I say better, these are agents that have full access to every piece of data within that company, every SOP, everything you've done with the company, everything you've told them in past conversations.

They're omniscient, so that's nice. They're available 24/7. They can be designed to have a specific personality that's on brand.

So they're never annoyed, they're never out of sorts, they're never going to give you back talk or whatever it might be. They're never going to want to go on a bathroom break. Every time you call, you can have the same agent if you want.

It could be whoever you spoke with last time could be the same agent next time, right? Because there's not a limitation on staffing or hours or they're not going to go and leave the job and go somewhere else. On all those axes, they do much better.

And they can actually resolve the thing you want, which is nice. They know the SOPs. The thing that people sort of say is not quite there yet is on latency.

Human conversation is about 500 milliseconds between turns. If I stop and you start, AI agents are closer to 750 milliseconds, so they're a little slower. So I notice it certainly being in the industry.

And then I'd say on mirroring, so if I say, "I'm really excited," you'll get excited. If I say, "No, I'm sad," you'll get sad. There are certain parts of human speech that AI agents are not yet good at mirroring.

But other than that stuff, it's much like talking to a human in that they are fully versed in whatever topics you want to discuss, they're having a general conversation. You can change topics. You're not stuck in a phone tree.

You're not getting the run around on some silly thing that makes no sense. They'll give you the straight talk, so to speak, and they'll get to the bottom of the issue. And still, even with these AI agents, they are able to escalate, right?

So we have tier one agents, tier two agents, tier three agents, much the same way there are structures today. And sometimes we still transfer to humans when there are issues that go outside of the normal SOP and there's not an answer for. So we have to give the conversation to a human agent who perhaps can do something that isn't in the standard SOP.

I'm happy to talk about the underlying technology if you want. That may bore people, but that's another whole side of what is actually happening under the hood. And there's a lot of stress in those gigs.

So, is their life ultimately going to be better? Because a lot of the lower-level stuff that over time conspires to make you tired and cranky will be dealt with automatically. Yeah, if you'd asked me even a year ago, I would've said mostly what we're seeing is truly people are taking what were offshore agents and moving that to AI and replacing human agents.

Today, what we're seeing is companies not replacing human agents, but doing what you're talking about, which is let's take the 80% of things that are rote or very time-consuming but not difficult to answer, and give it to AI. So an example is we worked for a big auto insurer. They had actually removed the ability to add a driver from phone support.

So if you called and said, "I want to add a driver to my auto insurance," they would've said, "Nope, we can't do it. " Of course, p*****g people off because, "I have you on the phone. " But they had made this decision.

So with AI agents, they can add that back. And so now when you call in, you can do it. Or if you go to self-support, you can do it.

There's no differentiation in the quality of support, right? You meet customers where they are, which makes customers happier. So I think to your point, the company's able to, from a cost perspective, use AI to serve a customer better in channels like voice that were historically quite expensive and support their team by not making them do the things that are rote and unnecessary for them to focus on.

Instead, they can deploy their human team on escalations and more complicated issues, and overall, their goal is to improve service quality, right? Reduce wait times, improve CSAT, resolve issues faster for customers without sending them to another channel, right, without having to deflect somebody, so to speak. A lot of customers have different levels of comfort with technology, and a lot of customers have different issues, and some issues are more complex than others.

So is it a good idea to try to force everybody down the same path, or can you let the customer more or less self-select? Because there are some customers who are doing everything they can to avoid ever having to talk to a human, and then there are others who, regardless of how cool the tech is, it's still their default to prefer to talk to a human. Yeah.

Again, the human versus AI distinction is a false distinction because within a year or two, you're not going to be able to tell the difference. So ignore that. But within the what channel do I want to talk on, do I want to do it myself or not, yeah, let the customer choose.

When I used to run a contact center, the sort of advice du jour was have self-serve, have digital channels, get rid of the phone. Nobody wants the phone anymore. That's just not true.

60% or 70% of interactions today are still phone interactions, and if given the choice, people rather start by self-serving, but when they can't do the thing they want, you know what channel they want? They want the phone, right? When it's something important, something they can't do other ways, they want to talk to somebody synchronously and just get it done.

So I think what you're going to see is, yeah, a lot more companies being willing to meet customers there on the phone. If the result is that the company implements AI badly and it comes off like an IVA or an IVR, well, that's a fail, because then you haven't actually given the person a real phone channel. You're just still in the world of IVA and self-serve deflection.

The customer has to really be able to interact with you in the same way they would with a human. A lot of organizations have also been trying to, for lack of a better phrase, upsell people on various goods and services when they have a customer support issue, and the assumption is that the customer's issue is resolved, and this is a good time to tell them about some other new capability or whatever it is. Yeah.

A lot of companies have had mixed success doing that. So will they get better at doing that in the age of AI, or is that still just not a great idea? Yeah.

I think two different ways to take the conversation. One is, I'm a strong believer that you have to stop thinking of the contact center as a cost center, and you've got to think about it as a revenue driver for the organization, right? Just in the most simple, basic sense, let's do a thought experiment and say let's remove the contact center.

Would your revenue per customer go up or go down? Well, obviously go down. Customers would just leave in droves.

They'd be p****d. So it is a revenue driver. So the question, I think, before even cross-selling is, what are the points throughout the customer's life where we can really change the trajectory of that customer's experience with this brand?

Cross-selling is a pretty rudimentary way of doing that. There are more sophisticated ways, but I definitely believe in that story. I think the other piece is there's a lot of moments in the customer's life cycle where it historically has been hard at any cost to support the customer live.

" That's something you can do self-serve. " But customers do in droves. And so companies historically have not wanted to handle that on chat, on phone, on anything, because they don't want to take the cost of an interaction that is worth nothing to them and can be done self-serve.

So that's a good example where, one, with AI, it lowers the cost to serve, so lowers the barrier for brands to do it, and two, to your point, perhaps with AI, we have seen success where cross-selling just a little bit, not heavily, but just a little bit, results in better outcomes. And that doesn't have to be selling a specific product. " That's it.

And there's no particular products, but those become 20-minute conversations, and it turns out a lot of people don't know how the banking system works and are leaving too much money in checking, not enough in savings, not enough in investments. Are not thinking through really how they should be flowing their funds and how they should be paying down credit cards. And soIt's very helpful to them to make sure they understand how the system works, and that builds a deeper relationship with that bank so that ultimately the customer stays longer.

So it's not explicitly cross-selling, but it is revenue-driving in the sense I started with. There's no shortage of AI projects that companies can launch. So what makes customer service and support one of the first areas they should be looking at versus the other 500 things they might consider?

Yeah. I'd say factually in my experience, coding is probably the first place most companies have gone. So the coding assistance for engineering.

After that, most companies, I'd say CX is the second, and then perhaps, marketing is looking at things and some other groups. But just for what factually seems to be high on people's lists. As to why, I'd say on one side it is a big cost center, and so people are constantly thinking about how can new technology impact what we're doing because it's a big line item on the P&L.

On the other side, more proactively, I think a lot of companies are worried about the world that they've created. What I mean by that is over the last 20 years, companies have backed away from customers. And it was incremental, but it was sort of now we're at a point where customers are going, "Does this brand even like me?

Does this brand want my business? " Even like an airline where I'm spending, as a business traveler, I'm spending $100,000 a year with an airline. They still make me feel like I'm not something important to them.

And so I think brands are realizing that they've gone too far in that direction. They have to come back. And so in that effort to come back, AI plays a big role in that story around how do we do things in a very different way because we don't want to go back to the fully manual people only, no technology way of serving.

So I think those are the two reasons. Then there's a reason that customers don't understand yet, which I think they will come to, which is part of why coding assistant works so well is it's a closed system. What I mean is that you can say to the AI agent, "Make me a code that does X," and then it can go test against the set of predetermined tests whether that succeeds, and then you can push to production without ever checking it.

Very few industries have that sort of capability. CX is one that does. So we have automated evaluation, synthetic evaluation of the AI agent.

So you can say, "Edit the agent to do X," and then you can have it run against our automated synthetic evaluation to see whether it does all the things you want it to do, and then immediately put it in production. So that closed loop system is quite powerful. On the other hand, art is subjective, so there's not a closed loop for art that's possible.

But because the closed loop is possible in CX, the ability to automatically iterate and improve agents is much more powerful. There's also been a lot of concern lately about the cost of AI, and not all AIs are the same either. But how should I think about this kind of initiative in terms of affordability, and what are my costs and what are my issues going in?

Yeah. I'll give you a funny thing that we've seen. So when we first started, mostly we were using the most advanced frontier models and all in, let's say the cost ended up being for an AI, a voice AI agent, call it 15 cents a minute.

That's in comparison to a Philippines agent at about 30 cents a minute, and the US-based agent about a dollar a minute, just for relative scale. Our hypothesis at the beginning was that as the cost of those models came down, people would continue riding that cost down, and just take the cheapest models that did the thing they want. What's happened is something slightly different.

0, which was like the OpenAI model that was a couple of years ago. As the cost of that model has come down, people have just jumped up to the next best model, which is still about the same 15 cents all in, let's say, cost. And then as that comes up, for whatever reason, that cost structure around 15 cents per minute seems to make the ROI on all these projects work, and people are happy to stay in that cost structure.

Not very many customers want to use the newest frontier models, to your point, that are maybe all in, let's say, 30 or 40 cents per minute. Because now you're more expensive than the Philippines agent, and it's not such a good trade-off. But for whatever reason, about half the cost of a Philippines agent seems to be a place where people are very happy with the ROI.

So to your second point on the overall cost, I think a lot of organizations are thinking about what is the ROI of this, I guess. And in some cases, very simple. I have a team in the Philippines that does X work at 30 cents a minute.

The AI will do it at 15 cents a minute. I've saved 50%. That is a very easy case.

No muss, no fuss. The cases that are slightly more complicated are where they go, "Okay. Currently, I spend $40 million a year on CX humans.

" So it's not pure cost savings, it's the organization investing in what is a better CSAT. But more and more, we're seeing the latter, not the former. So you've been at this a while.

What do you see folks doing that just makes you shake your head a little bit and go, "Maybe we want to be a little bit smarter than that"? So the one I always explain to every new customer is every new customer has the feeling that they should test small to start. m.

" Like, yes, let's test small, but I'll explain how to do it. m. , nobody cares about it.

No resources are going to help you. m. calls.

So instead, let's take whatever is your biggest call type, whatever's the main one, and let's do 1% of those calls because then everyone's going to care about it. And once it works, we can easily scale it up and have more impact. And people eventually understand that, but it'sIt's human nature to be risk adverse, and so slightly reframing how you are risk adverse is important.

" And it's great that people identify that as an area of continued investment. In our testing, that's actually not what drives better CSAT necessarily. So when we improve latency by rate, sometimes it costs more, or we improve voices, sometimes it costs more.

It's not actually resulting in higher CSAT. The things customers care about the most are, did they answer me right away? Did they resolve my issue?

Were they nice when they were doing it? Whether it's 500 milliseconds or 750 is not what drives the difference in CSAT. So I think that's tough when people's initial feeling about something is one way, and they have to wait until they see real data from this in production.

And now we've done this enough times that we can just show them the case studies, and they don't have to just take our word for it. But especially at the beginning, people over-indexed on what was the feeling of latency in the agents. And the example I try to give people is I say, "Okay, call a human agent and what happens?

" Another two minutes. Now, so I go, "Guys, so what are we talking about latency? Like yes, the AI agent is 750 milliseconds instead of 500, but they never put you on hold for a minute to look up anything because they know everything.

" So it's a little bit of reframing what matters is important. All right, folks, you heard it here. When it comes to AI, the outcome is what matters most, and if the customer's happy, then maybe that's the way to go.

Hey, Alex, thanks for being on the show. Yeah. Thank you for having me.

All right. ai Leadership Insight Series. You can find this episode and others on our website.

We invite you to check them all out. Until then, we'll see you next time.