Synopsis: In this AI Leadership Insights video interview, Amanda Razani speaks with the CEO of Cognigy, Philipp Heltewig, about conversational AI and its impact on the customer experience.

Amanda Razani: Hello, I’m Amanda Razani with Techstrong AI and I’m excited to be here today with Philipp Heltewig. He is the CEO of Cognigy. How are you doing today?

Philipp Heltewig: Very well, thank you. Hope yourself too.

Amanda Razani: Yes, glad to have you on the show. I think we first met at the Voice and AI event.

Philipp Heltewig: That’s right, yes.

Amanda Razani: Can you share a little bit about Cognigy before we go into speaking about AI and some of the discussions at the Voice and AI event?

Philipp Heltewig: Yeah, sure. So Cognigy is a provider of contact center AI solutions. So what that means is we provide software to create advanced artificial intelligence based agents, so contact center agents that work alongside human agents, either in the shape of chat bots or voice bots, phone bots, or as bots that help the agents, we call that agent assist or AI assist during the conversation. We’ve been doing that now for a number of years. So we were founded in 2016 in Germany, but we have operations all around the world and we were named as one of the leaders in the Gartner Magic Quadrant fall sector, and mainly work with large enterprises around the world.

Amanda Razani: Awesome. And so what were you doing at the Voice and AI event? Why were y’all there?

Philipp Heltewig: Well, the Voice and AI event, obviously when you look at what we’re doing as Cognigy, it’s exactly the right event to be at. It’s about voice and about AI, what we’re doing is voice AI, at least in parts, we’re also doing chat AI. So we exhibited, and we held a couple of interviews, sessions, things like that, and overall it was a great event.

Amanda Razani: What were you hearing from some of the other business leaders there? What were some of the key issues surrounding AI?

Philipp Heltewig: I think there’s a lot happening in the industry, there is ChatGPT coming on the scene, everyone talking about the generative AI and things like that. And it is an amazing technology, obviously. We can do things all of a sudden that weren’t possible just 12 months ago, and it really brings the whole promise of AI into the real world. All of a sudden there is really AI and so there’s a lot of talk around that, around the benefits of that, but also around the challenges because now, of course, we’re back in this world where we hear things like, “Well, why don’t you just do that? ChatGPT reads the whole internet.” So things like that where because we each to have a bit of a more differentiated view on what it actually does. So I think it’s a blessing and a curse, more a blessing than a curse, of course. So there’s a lot of chat around that.
There’s a lot of chat around the uptick in the industry, so more and more large enterprises around the world are turning to AI to optimize the operations in the contact center, mainly because it’s not really cost savings driven. It can be cost savings driven, but mainly because there’s a customer service gap, because there’s not enough agents in the contact centers and they’re turning to AI to fill that gap. And that is something that is very positively received around the industry.

Amanda Razani: So it has been a big year for conversational artificial intelligence, or CAI for short, with the emergence of this generative AI. So how is it a game changer? Can you share some use cases?

Philipp Heltewig: Yeah, of course. I mean, there are a gazillion use cases and as the CEO of Cognigy, I’ve firsthand seen how generative AI can revolutionize the way that we interact with machines. I use it on a daily basis from a personal perspective, and from a professional perspective, and we use it all around the company, in engineer, in new marketing everywhere. But it has also, as I said earlier, there was always this promise of bots help you in the contact center, and that’s true, and it was true before the emergence of generative AI, but now they do that on a different level. Generative AI allows for more natural context aware conversations, making the conversations pretty much indistinguishable from human to human conversations. The bots can add empathy into the conversation, the bots can understand things that were traditionally very difficult for machines to understand, such as if, for example, the AI agent says, okay, so let’s say it’s a [inaudible 00:04:34] reservation, so how many people would you be coming with?”
In the past you had to say four or five. Whereas now you could say, “Well, it’s me, my wife, and the two kids, and we’re bringing the dog.” So that’s four, or I guess depending on the size of the dog, it might be five. But it’s this type of human level understanding and then also the human level to respond. So if you want to book a flight and the AI agent says to you, “Okay, so how can I help you today?” And you go like, “Well, I’m going on my honeymoon and I want to book a flight.” The AI agent can reply something like, “Oh, that’s great, I hope you have a great time. Let me help you with that.” Whereas this might be an inappropriate response if the person just relayed that there was an emergency that they need to attend to like, “Oh my god, let me see how I can help you with that.” So that will be a better response. So this empathetic type of reply, I think very important.
And then there’s of course the customization aspect where generative AI can be tailored to specific industries because it has base knowledge around these industries already. So if you, for example, say, Can I bring my elephant on the plane with me?” Then you don’t need to teach generative AI models that that is a bit of a silly question. “Well, you cannot bring your elephant.” It needs to be a special plane from FedEx or whatever it is. So yeah, it’s definitely done a lot. We had a marketing slide at one point, which was bots no longer suck. And I think it’s a bit of an aggressive statement, of course. We didn’t use it externally, but it’s true. Now we’ve entered… Like Bill Gates said it, “The age of AI has begun in 2023.” And it’s true. So now we finally have AI that we’ve previously only seen in science fiction.

Amanda Razani: Absolutely. So being focused on customer service, what do you think are the biggest challenges today for contact centers and how can they use AI today?

Philipp Heltewig: Yeah, so I mean, firstly, we have to acknowledge that there are challenges, because if there weren’t any challenges, then we wouldn’t really need this new type of technology. And if there weren’t any challenges, then none of us would ever wait on the phone lines for 30 minutes to five hours when we call a hotline, I guess we’ve all experienced that. So why is that? I mean, it’s not because the contact centers don’t want to perform, they don’t want to be efficient, it’s really because there are high volumes of queries, they need qualified agents so they can maintain a certain quality of service. Then there’s fluctuation, like there’s agent burnout, there’s huge fluctuations in the contact center. So if we reiterate those three, again, there’s high volume of queries, quality of service and agent burnout, these are the three major problems we always hear about.
And then there’s another one, that’s a bit adjacent to that, and that is the adoption of AI. Everyone knows, well, I have to add AI agents into the mix, like my workforce in the contact center has to include humans and AI agents, otherwise you can’t handle all these volume of queries, et cetera. So while AI is increasingly common, it’s not yet universally adopted. And the level of sophistication really varies around different industries, but also within certain industries.

Amanda Razani: That human element is still very much important at the moment. So let’s talk-

Philipp Heltewig: Absolutely. And you have to imagine this, if you’re a contact center agent, and actually we have a lot of people working for us that used to be contact center agents, so they can really tell those stories from the battlefield. If you are a contact center agent, you’re mostly dealing with people that are distressed because firstly, they wouldn’t call you if they were not distressed, if they didn’t have an issue to handle. And secondly, if they then wait on the phone lines for 45 minutes, and maybe it’s already the second time they called because the first time the call drops, their level of anger is something that the contact center agents really isn’t their fault, but has to deal with that. So we can use AI to not just make the customer’s lives easier, but also make the agent’s lives easier, lowering average handling time, and thus reducing the time in the queue, things like that.
And that’s something that’s often forgotten. We always talk about increasing or enhancing customer experience, but the agent experience is oftentimes forgotten. And there is a huge issue with agent turnover because of agent burnout, and that is something that can be tackled with these AI technologies as well.

Amanda Razani: Absolutely. So let’s talk about efficiency and cost, what does it look like for a contact center looking to add AI into their technology mix?

Philipp Heltewig: So let’s talk about efficiency first. AI can handle routine queries, freeing up human agents for more complex issues. Or, we have many customers that put AI in front of every single call. So you’re calling in, the AI asks why you’re calling, so it greets you in a nice way, and then it asks why you’re calling. And then depending on the reason why you’re calling, it may identify you. So we call that ID and V, identification verification, and then we know why and who. And by knowing that we can then route much better as well. So we can say, oh, okay, this is a premium customer, we’re going to route them into a premium fast track queue or something like that. Or with one of our airline customers, if you’re calling in because your flight got canceled and the flight was within the next 24 hours, then you’re also put on a fast track.
So we can provide better service by prioritizing correctly. So that’s efficiency of course. But going above and beyond this routing and ID and V piece, we can of course handle routine queries. So if a customer calls an airline and says, “Well, what’s the maximum pounds I can bring on the plane with my carry on?” Or can I bring an elephant? Just like I said, these are queries that can be handled by these AI tools nowadays without any problems, and these don’t need to be handled by human agents. And then it can go further. I mean, it depends on how far these companies want to drive it. Sticking with the airline example, we could, for example, do a flight status check, which is one of the main reasons why people are calling in. Or it can go as far as going through a full re-booking process or a full refund process, 100% automated where there’s no agent involvement [inaudible 00:11:13] require.
Now, the interesting thing is if the bot at one point fails, because maybe it’s a special case, there’s a re-booking and there’s some special requests during the re-booking that the bot can’t handle, you can then hand the case over to a human agent, but it’s already half solved. So we know who the customer is because they’ve identified, we know their ticket number, and so we know which flight they want to rebook, we know where they want to rebook to, et cetera. So the human agent can handle the issue much quicker. So that from a efficiency perspective.
Now, from a cost perspective, cost is interesting because you can realize cost savings in the contact center by letting agents go. Now, we are not seeing that, that much and the reason is because of this customer service gap that I spoke about earlier. If you don’t have a fully staffed contact center that has a top-notch service where they pick up every call within five seconds right now, but if you have a contact center where it usually takes the customer an hour to go through the queue, and now you automate 50% of the contact using AI, this contracts the queue, and I know the math isn’t 100% correct, but this contracts the queue to 30 minutes. Now if you relay that into cost savings, but letting half of your people go, then the queue is an hour again. So it doesn’t really make sense. So it can lead to cost savings, but the efficiency and customer satisfaction increases are more what our customers are after.

Amanda Razani: Absolutely. So y’all just recently had a big announcement over at Cognigy. Can you share the announcement about Knowledge AI and its benefits to the enterprise?

Philipp Heltewig: Yeah, so Knowledge AI is our latest offering that’s based on generative AI, what you can do with that… Let’s take a step back and let’s take a look at how did you build knowledge into bots, previously, in the past, before generative AI, you would build up so-called intents, which are like question answer pairs. And then if the customer asks a question, then it would be compared against all the questions that are in the system and the one that matches closest, it would output the answer, the hot COVID answer that you have provided for that question. Now, if it couldn’t find an answer, it would say, “Well, sorry, I don’t have an answer for this.” And it’s very tedious putting all these question answers pairs in, obviously, and maintaining them. Sometimes it’s 1000s, sometimes it’s tens of 1000s. Now with Knowledge AI, we’re entering a completely new age in how we can apply artificial intelligence to fix that issue.
Because what you can do with knowledge AI, you can provide PDFs or Word documents or text files or websites or anything like that, anything that contains information, and the AI absorbs all of that. And then when the customer asks a question, we’re not just spitting out exactly what’s in the document. Well, you can configure it to do that, but in most cases, you would use generative AI to get exactly the answer that the customer was asking for. So to give an example, maybe we stick with this elephant example with, can I bring an elephant? So an article is absorbed that talks about bringing pets on board. Now you search for can I bring an elephant on board? This article is found, but outputting this article, which obviously doesn’t talk about elephants, but talks about dogs and cats, doesn’t really make sense. So it wouldn’t be the right answer to the question. What our system then does using large language models, it extracts or it takes this information about the cats and dogs, it takes the question the user has asked, and it creates a fitting answer to that question.
So it might say, “We’re very sorry, I don’t have any information about bringing elephants on board. I know that you can bring cats and dogs if they weigh less than eight kilos into the cabin.” Something like that. So it’s a really good answer as compared to either saying, “Well, I’m sorry, I don’t know the answer to this.” Or just outputting a wrong standard answer. And that is what Knowledge AI does, it makes the creation of these standard bots much faster, it uses this cutting edge type of technology to really extract and formulate a perfect, very human-like answer.
And you can also ask follow up questions. So I could for example, say, “Well, okay, what about dogs then?” And that is something that may in the past not have found anything because I’m referring to something… Or maybe I don’t say, “What about dogs?” Maybe I say, “What about if my pet weighs 16 kilos?” Now it knows that it refers to the article we just spoke about, and it creates a fitting answer like a human would give like, oh, we’re very sorry. You can only bring your pet it weighs less than eight kilos, if it weighs more than eight kilos then it needs to travel in the cargo hold and then go like, well, is there water there, for example? And again, we’ll know that you are referring to an ongoing discussion.
So all of these things were really not possible before. So it’s much more easier to build. And because it’s much more human-like, and because it’s much more flexible, of course the resolution rates are much higher than it was in the past. So it’s a milestone in the history of our product, I have to say, it’s the first true generative AI based product we’ve brought out, and it’s getting a lot of interest in marketplace.

Amanda Razani: Well, congratulations; it sounds great. A way to take all that data and produce the best answer; that’s wonderful.

Philipp Heltewig: That’s right.

Amanda Razani: So what advice would you give to anyone listening that might be interested in incorporating AI for better customer experience?

Philipp Heltewig: Well, you’ve probably heard that oftentimes, but I’m going to repeat it anyways, the first one is really start small. We keep sticking with this airline example, say you’re an airline, you are the Techstrong airline, and you want to automate your customer service with conversational and generative AI, don’t try building an AI agent that can answer all the questions, rebook the flights, change your seats, request a special meal, all these kinds of things, because you’ll never be done. Start small by maybe using something like knowledge AI. So you start by automating FAQ type of interactions, and maybe start also by then collecting analytics and measuring why are people contacting us? And then after a while, so maybe after a month or after two months, you run some analysis over that, and then you see, oh, most people are actually contacting us to rebook a flight. So 80% of people are one that want to rebook the flight, let’s just say.
Then that is a process you can look at. Is it relatively easy to automate? Because you need to, of course, put some effort into then automating this process, and then you can get the best ROI. So start small, focus on the quality, I think that’s also really important. And very important, always have the ability to hand the conversation to a human if the automation fails. I think we’ve all experienced frustration with bots in the past, and the frustration was when the bots got stuck in a loop and you couldn’t get out, you couldn’t do anything anymore. But I think everyone would agree, well, if I can get my issue resolved quicker by maybe talking to a bot first, and as soon as it gets stuck, it hands me over to a human and I can still get it resolved quicker because I worked with a bot than anyone would do it.
So maybe let’s summarize those again, start small. So begin with a pilot program, understand what you want to do, focus on quality and make sure that there is a way to still speak with a human if possible. And of course, in the end, also choose the right partner. And the right partner is not just a tech company like Cognigy, but you also need someone that really goes with you on that journey that understands conversational design, et cetera, that can really help you set that up. Because that, I think is also super important to know how to talk to the customers because these AIs will, in the end, be a part of your brand, so you want this to be at high quality.

Amanda Razani: Absolutely. Well, thank you so much for coming on our show and sharing your insights today.

Philipp Heltewig: Thank you, it was great.