00:10
Ciaran
Hello and welcome to Customer Friendship Conversations, the show where we bring you the latest trends, tools and insights into delivering customer experience as it’s meant to be. I’m Ciaran Nolan and I lead relationship management. Predictor, today’s episode is an exciting change from our usual format. One thing that comes up time and time again in Customer Friendship Conversations is the role that AI is going to have in shaping the future of customer experience. That’s why we decided to bring you the audio from a webinar Dixa held on this very subject, featuring Rob Krassowski, Tue Søttrup and Mirza Beširović, you’ve probably heard all about Chat GPT, the new AI powered chatbot that’s taken the world by storm. It might be the most widely adopted tool in recent memory. And amid market downturn, do we all need to be worried? Listen to hear discussion of how we should navigate emerging technologies, whether or not Chat GPT can impact customer loyalty, and how we’re integrating Chat GPT into the platform here at Dixa.
Enjoy.
01:16
Rob
Welcome, everybody. My name is Rob. I’m CPO here at Dixa and I’d like to welcome to you to Dixa’s Webinar on Chat GPT. Is it coming for your customer service job? It sounds scary. Hopefully you leave today feeling informed. Maybe scared, maybe not. Who knows? We’ll have to see where things go. Really excited for all of you that could join us from across the world. So let’s go ahead and jump on into it. So what can you expect from the webinar today? We’ve got a few big things that we’re going to hit on first, chat GPT. It’s a big topic, it’s a complicated topic. There’s a lot of technology involved. So we’re going to talk a little bit about the rise of AI, why we’re here today, why Chat GPT is such a big deal, and a little bit more about just the space that Chat GPT occupies.
02:04
Rob
AI, large language models. What is all this stuff? After that, we’re going to check out a little more specifically, what actually can Chat GPT do for your service organization? How might it affect the way you work? How might it change the way you work and how can you use it? How can you leverage it in your daily work? And then finally, we’re going to check out what to do. How do we approach working with generative AI in a way that makes customer service teams better? Of course, this isn’t all about us talking here. We want to hear from you in the audience as well. And so we will hit a little Q and A at the end. Excited about that? So who are you going to be hearing from today? Let me introduce our panelists. So, first off, my name is Rob Krassowsk. I’m the Chief Product Officer here at Dixa.
02:51
Rob
I’ve been at Dixa for almost three years now. Once I moved to Copenhagen. I’m American by birth, but I’m in Denmark by the grace of God in a beautiful, sunny day here in a beautiful, sunny Copenhagen. My background is all about consumer technology. It’s building apps, especially in the wellness space. But I’ve made the jump here to working with Dixa in the customer service space and excited to be here directing our product and engineering organization. With me. I also have tua. Tue is our VP of CX Excellence here at Dixa. Tue, do you want to say a few things about yourself?
03:26
Tue
Yeah. Thank you, Rob. Great to be here, and thanks for joining today. I’ve been with Dixa just over six years, and I have a background working as both an agent and as a manager. And I work in the crossfield between sales and product, so I listen to a lot of prospects and customers and hear what they’re talking about. And Chat GPT is, of course, something that comes up a lot, and then I try to take that and feed it back into the product.
03:49
Rob
Amazing. Great to have you, too, and really looking forward to hearing about your vast experience in the CX space and how you see GPT fitting into that big puzzle. Also today with me, we have Mirza, who’s one of our directors of product management here at Dixa, specifically in the AI and automation space. Mirza, can you introduce yourself, please?
04:11
Mirza
Thanks, Rob. Yeah, I’ve been with Dixa for about a year and a half now. As Rob mentioned, I focus primarily on our suite of automation products, and I’m very excited about all the possibilities that machine learning and AI have within the customer service space. This is what I spent more than half of my career focusing on. My kind of career in tech, started really working on machine learning products, and I’m a linguist by training. So Chat GPT in this conversation is super relevant and very interesting for me, and super excited to be here with you today.
04:42
Rob
Great. Thanks so much for joining us, and really excited to hear what you have to say on this, as well as one of our subject matter experts here at Dixa. All right, who is Dixa, anyways? I know a lot of you are joining us, hearing about Dixa maybe for the first time today, so we just wanted to give you a little bit of background. I promise this will be the short part. Dixa is a conversational customer service platform that makes it easy for customer loving brands to connect with their customers, regardless of the channel they come in on. We handle about 30 million conversations per year. These are all customer service conversations, so we deal with conversations a lot. And that’s what Chat GPT is all about. It’s about conversational AI. So this is really in our sweet spot. We have customers all over the world, and we’re always looking for new ways to leverage technologies to make these customer loving teams more efficient, make sure that they can deliver customer service that’s amazing.
05:37
Rob
And do it at scale. So that’s who Dixa is. That’s what we’re about. Conversations are what we live and breathe every day. And so that’s why this topic is super near and dear to our heart. So, moving on from Dixa to Chat GPT to the world that we’re in, let’s go back a little bit. Let’s take kind of the broader context and a little history before we get directly into Chat GPT. Mirza, I thought this would be a great piece for you to tackle. You want to run us through a little bit of the history around Generative AI, how we got to where we are today?
06:08
Mirza
Yeah, sure thing, Rob. Thank you. So, artificial intelligence as a concept is much older than you might be aware of. It’s sort of the concept the idea starts already in the first half of the 20th century, was popularized through movies and books, everything from Tin Man and The Wizard of Oz to the humanoid robot Maria in Fritz Lang’s metropolis. So, as early as 1930s, with this cultural backdrop and popular imagination, AI research then really kicked off in the 1950s, and there were a couple of seminal moments that I think defined the field all the way into the present day. So, in 1950, Alan Turing, who I’m sure you’ve all heard of, wrote Computing Machinery and Intelligence, essentially a paper that introduced what is still the logical framework for AI today the so called Turing Test, which you use to determine whether a machine has so called intelligence or artificial intelligence.
07:10
Mirza
The term itself the term artificial intelligence was coined six years later in 1956 by John McCarthy at a conference that was dedicated to discussing this new and budding field. So, there was already a lot of excitement as early as the 50s about artificial intelligence. The challenge was that computational power computing as such was incredibly expensive and simply not advanced enough yet. So, even though we already had the basic precepts of AI, that haven’t changed since the 1950s, it wasn’t really possible to do some of the major advances that today are sort of mainstream, and here we all are talking about them.
07:51
Rob
Or to chat with the punch card.
07:53
Mirza
Yeah, or to chat with the punch card. Sure. But over the next several decades, there was a lot of advancement made, both on the computing side and on theory side, when it came to artificial intelligence. So, the first Chat bot came to us as early as 1966. This was Eliza AI. Research, then funded by the US. Research agency DARPA, continued to develop different forms of AI language. And moving fast forwarding a little bit to the Kasparov, then world chess champion and grandmaster was defeated by IBM’s Deep Blue model. I think this was also another moment that captured popular imagination and brought the topic closer to all of us. You might remember that same year, Windows introduced speech recognition. This will age me a little bit, but I remember using it myself. I’m sure some of you in this call did as well. I was super excited about it.
09:01
Mirza
IBM again, I’m just traveling through time a little bit here. IBM in 2011 launched a model called Watson, still used today, by the way, that won Jeopardy. I think that was also a super interesting cultural moment. And in 2017, Google defeated the world champion Go with its alpha Go model. So a lot of this interesting stuff that happened over the course of 50 years captured the popular imagination and brought the topic of artificial intelligence into science fiction, into daily imagination. Today, the focus has shifted to what we’re calling generative AI. And if you look at it kind of from, of course, 50 years later, we can break it up into, let’s say, three major eras. They’re all part of artificial intelligence. So artificial intelligence, as such, essentially refers to machines mimicking cognitive functions and exhibiting intelligence. Machine learning, then, is sort of a subset. It’s an approach to artificial intelligence because it learns from data and learned experiences that then make data driven predictions based on those experiences.
10:15
Mirza
Generative AI, Chad GPT, all that stuff. Now, that falls under deep learning. This is a branch of machine learning where algorithms essentially work to model high level abstractions in data. Now, it’s a bit difficult to untangle that sentence that I just said, but fundamentally, what we’re trying to do is approximate the function of brain neurons. We’re forming so called neural networks. And I want to talk a little bit about I’ve also just thrown out a lot of expressions here. There’s a lot of expressions floating around talking about Chad GPT. So I want to get down and help us define the lingo. So we’re all on the same page. So what is generative AI, then? It’s essentially just a type of artificial intelligence. It’s an AI system. It’s a set of algorithms I think is the easiest way to think about it. And it can generate different kinds of data, such as images, video, audio, text, even 3D models.
11:10
Mirza
Today, there’s hundreds of players in the generative AI space and hundreds of companies that are trying to build their own models and their own applications of the technology. Large language models, or LLMs, is again, another type of artificial intelligence. It’s a deep learning algorithm. It uses deep learning techniques to recognize, summarize, translate, predict, and generate data and content. And GPT then, or generative pretrained Transformer, is a type of LLM. It’s pretrained on a large data set of unlabeled text, and it’s able to generate novel humanlike responses, human like text. It’s essentially the evolution of something Google introduced in 2017, so called transformer models. And then lastly, but why we’re all here chat GPT. Well, Chat GPT is a chat bot first and foremost, but it’s a conversational interface. It’s a very complex chat bot in the sense it’s a conversational interface that’s built on top of OpenAI’s GPT-3 and GPT Four, which are large language models.
12:25
Mirza
All right, handing it over back to you, Rob.
12:29
Rob
Yeah. No, thanks, Samira. I’m glad that we gave the linguist the definitions here so I didn’t have to stumble over the big words. I think the takeaway for me here, right, is that Chat GPT is getting a ton of hype. It’s super interesting, but doesn’t mean it’s the only player in the space. And I think when we go through this presentation today, look, we’re going to be talking about a lot of different generative AIS, but we kind of use Chat GPT interchangeably with those. But I wanted to sort of foreground why we’re here today, and that’s because Chat GPT has been a massive technology hit. It’s been huge. It’s been one of the fastest growing technologies we’ve ever seen. Right, so over here, the graphic on the other side, on the right side of your screen, how long did it take? Chat GPT hit that big milestone, that million user milestone, only five days.
13:20
Rob
That’s incredible. An incredible number. We look at some older companies like Netflix, launched back in 99, took them almost three and a half years to get there. Even when we look at some newer companies, like Instagram, two and a half months. So Chat GPT is setting a massive record here for getting to 1 million users. And they didn’t stop there, right? They went to 100 million users in only two months after launch, which is also wild. It took TikTok, which was also a super hit, nine months to hit that same mark. And Instagram two and a half years. So we’ve really seen adoption just fundamentally change the game here. So the reason your CEO is knocking on your door, the reason your friends are talking about a dinner party, is because everyone is using it. And the reason everyone is using it is because Chat GPT makes AI available for everybody.
14:10
Rob
It’s a conversational interface for AI. So you can engage with these super complicated models without knowing a lick of code, without knowing any math, without having to be a mathematician or data scientist. So it’s really democratized AI brought it to the masses. So that’s why we’re here today, talking about Chat GPT. So we’ve seen how popular it is in the marketplace, but one thing I’m curious about is for you in the room, is Chat GPT something that you’ve used before? So we’re going to throw up a little poll here on your screen and we’d love for you to just go ahead, end the poll. Have you used Chat GPT to help you in your job? All right, already, right off the bat, we are seeing a huge where people are pounding. Yes, we’re seeing about 80%, 75%. Okay, now the no’s are coming in, so we’re about 25% no, 75%.
15:07
Rob
Yes. We’ll give it just another maybe 30 seconds here for people to vote and see where we end up. Yes, it looks like we’ve stabilized, I’m going to go ahead and share the results here. Unfortunately, it’s having a problem sharing it. But what we can see on my side, I guess you’ll just have to take my word for it, is that 70% of folks have said that they have used chat GPT to help them in their job, and 30% of folks have said no, that they have not used oh, people are saying they can see it. That’s good. Just an error on my screen. That’s great. Okay, you don’t have to take my word for it, but now you know I’m telling the truth. That’s a good thing. All right. Great to see that folks are using these new technologies. And for those of you that aren’t, we’re going to give you a little insight into maybe how you can think about using them.
16:05
Rob
And to do that, I’m going to kick it over to Tue, who’s our CX expert, who’s going to run us through some ways that at least we see Chat GPT affecting the customer service space. Tua?
16:18
Tue
Thank you, Rob. We’re going to go through five Chat GPT facts, and the first one that we’re going to talk about is how Chat GPT can help you learn from customer interaction. So what these models are really good at is making sense of large amounts of data. It can analyze the content of customer service conversations, and it can tell you what they’re about. And if you don’t already have a Chat GPT account, you can create one. And you can go to Chat GPT, and you can put in a trust pilot link of a brand that you are considering buying something from. And then you can ask the model, tell me what customers feel about this brand. So instead of having to read a lot of reviews, you can just get an aggregation of it. So it opens up some possibilities that allows you to make sense of data.
17:07
Tue
To do that, you of course, need to have a strategy for why do you want to do it? But you also need and that’s even more important, a process for doing something with the output of it, because, unfortunately, we see a lot of data being collected and not a strategy for actually making that data actionable. And that’s something that we’re going touch upon later. How dangerous Chat GPT for your job? Somebody needs to do something with the data. So it’s not enough just to make sense of it. You also need to process it and then execute on the findings that you get here. The next fact that we have here is that Chat GPT can help you handle repetitive inquiries and tasks. And what we also know is that a lot of customers, they want to find the answer themselves. Like, they are reluctant to contact customer service because they have a bad experience with it.
18:01
Tue
It takes time. It’s hassling they just want to have their issue resolved we know that we don’t live in a perfect world and things can and will go wrong, and it’s about how those issues are resolved. What we do see is that because Chat GPT have made AI much more accessible, people are getting a bit more comfortable speaking with AI, and we see a broader adoption among people that are using it. And Mirza, you already talked about your experiences with chatbot, and I would love to hear your insights here.
18:36
Mirza
Yeah, sure. Toy the beauty of Chat GPT is that it’s essentially a chat bot on steroids. The large language model that underlies it enables it to do things that most traditional chat bots struggle with and that takes an insane amount of time to set them up to be able to do and handle certain use cases. Chat GPT sort of can learn on the fly and you can just feed the data and it can kind of figure out what to do with that. In a chat bot sense, its power is the power of every chat bot. It’s to deflect, it’s to automate conversations and to handle all of that repetitive, low value work that both your customers and your service agent don’t want to be doing with another human, where’s my order type thing or whatever use case you might have. The beauty of it. I think the huge advantage of it is that it’s just so much faster to set it up than any previous chat bot.
19:40
Mirza
And it’s faster, it learns faster, and it’s much more highly adaptable.
19:45
Rob
I’m glad you guys pointed this out because I think this is really one area where everybody just focuses in on saying, like, oh, Chat GPT is just going to take all of our conversations and it’s going to respond to all of our customers. But that’s not really one to me. That’s not a great use of the technology. It’s not all conversations that should be handled by a bot like this. And Mirta, maybe both of you guys have some insight on this, but there are times when customers want to connect with a service agent, right? This isn’t going to just say, okay, we don’t need service agents anymore. Maybe we can talk about some of those examples. I know we have a lot with our customers who say, look, there are times we need to talk to our customers. We want to talk to our customers.
20:33
Rob
We want to hear their voice on the phone. They want to hear our voice, whatever it is. Can either of you share a couple of examples of ways that this might not be a good fit?
20:43
Tue
I think from my point of view, there is a lot of organizations, especially if you have a subscription service, you need to do a lot to retain your customers. And we see a big correlation between retaining customers and the experience they have when they reach out to customer service. These brands are very much focused on consistently meeting customers expectations in every single interaction. But they can also see that when they speak to their customers are more likely to stay longer and spend more money with them. And that means that you can really prove the return on investment in CX in general, because things go wrong. They do that and it’s called the Service Recovery Paradox when you interact with customers because something goes wrong, if you meet customers expectations, you give them a good experience, they will remember how you made them feel at that point and they will be more likely to repeat purchase with you, spend more money.
21:41
Tue
Also tell their friends and family about the experience that they have. So it is really important to focus on delivering these good experiences because customers, they are very likely to move to another brand or another product after just one or two bad experiences. Mirzaza, what’s your experience here?
22:00
Mirza
I build chat bots for a living and even I’ve caught myself thinking, you know, wow, chat bots suck. And everyone has Chat GPT or not. It’s a chat bot. It can do everything. It’s not designed to do everything. And humans need a connection. Sometimes the most important element of customer service is the human connection. Of course, we’re always there to resolve the problem that a customer has, but certain problems are emotional for people and they require a kind of interaction that even a language model as advanced as what underlies Chat GPT cannot provide.
22:40
Rob
Right, so what I want to hear from both of you, right, is this is a tool that we can use to get rid of some of the low value, maybe easy to manage inquiries to where’s my orders that stuff. So that teams can devote the time and energy that they need to those high value conversations that are really meaningful for a brand as they try to create customer loyalty. I think that’s a great way to frame it. Yeah, thanks for that insight. Great. Next one.
23:09
Tue
The next one here is that of course Chat GPT can help you improve productivity. It’s something that we talked about in the conversation that we just had before. And here we can see that when you work with data in a conversational way where you can just ask a knowledge base without searching, you can just state your problem statement and then it can take everything that’s in your knowledge base and relay answer that makes sense. So that’s from the customer side, they don’t know that they are speaking to AI, they just have a problem and it’s resolved. And here we can talk about invisible AI, but we also want to apply that to the agent experience things that they would need to do manually. We’ll give some examples of that later, but just to mention a few, we can help them state what the contact reason is.
23:59
Tue
We can identify certain keywords or topics within the conversation that can help get the right conversation to the right agent in the right order. We can also identify booking numbers order numbers or something like that, things that agents would normally have to do themselves. Like I said, customers want to find the answer, but when they don’t are able to find it, then they need to find an agent that can help them and that agent needs to have the right tools available to them. But Rob, in your opinion, is Chat GPT a chatbot?
24:34
Rob
Chat GPT is a conversational interface, right? But there is a lot of technology that underlies that we leverage in ways that are not chatbot like. So I know at Dixa and we’ll get into this a little bit later, we can share some of the strategies that we’re using, but it’s not always using Chat GPT as a chat bot. Sometimes it’s using the underlying models and technologies to do some really interesting things. But Tue, I think there’s something really interesting that you pointed out here, and that’s a customer service. It’s a two sided experience, right? There’s the customer’s experience, there’s the agent’s experience. Both of those come together to create those great sort of loyalty level experiences that were talking about just a minute ago. Right? So it’s important, I think, when we look at new technologies like this, that we are also investing in agent experience, making that better, because that ultimately affects customer experience as well.
25:25
Rob
So I think that’s a really nice point to take away from this one.
25:30
Mirza
One of our recent guests referred to Chat GPT as your intern. So I think that’s also an interesting way to look at it.
25:38
Rob
Yeah, it can be a great helper, but you don’t always want to put it in the board meeting, I guess.
25:44
Tue
Yeah, you’re absolutely right, Rob. And when we talk about how Chat GPT can help you stay ahead of the curve, it very much relates to the agent experience because we know that happy agents create happy customers. And something that makes agents unhappy is to move across a lot of different applications, having to look for order information about process description for how to solve a specific issue. They just want to be able to focus on the customer. Why did the customer contact you? Why did they contact you before on the same other channels? Also, who is the customer? What was the last order? How much money have they spent? Pulling out information that allows agent to deliver a personalized experience at scale and then also helping them to know how to solve the customer’s issue? Looking at the question from the customer and providing contextual information that can make it easier and faster to solve the customer’s issue in the right way and also use this to offer the next best action, maybe not just answering the question the customer had, but also the follow up question.
26:50
Tue
And technology can actually be used in a lot of ways to eliminate a lot of the friction for agents so they can focus on delivering good experiences.
26:59
Rob
Yeah, and I think it’s important here. It does help them build skills over time. Right? And I think that’s really important. We talked a little bit about agent experience on the last slide, but we also have to remember, agent experience is more than just creating good customer experiences. It’s also creating stability within your organization. It’s making sure that those agents don’t look for a job somewhere else where they have better tooling. Right. I know if I’m a customer service agent and I have one tool that basically puts me in an Iron Man suit and allows me to do my job better, I’m going to be much more engaged with that than I would be somewhere where they’re not utilizing that sort of technology. So I think that’s a really.
27:39
Tue
Great point here, number five, and I guess that’s why most of you are here today. Debt GBT is not here to replace you. It’s something that can assist you. It is not a silver bullet that can solve all issues that you have. Like I said, you need to have people that can implement these last language models and more importantly, somebody who can do something with the output. And it is a very important tool that you can add to your toolbox. I’ve been working@saxo.com Selling Books, Denmark’s largest online bookstore. When I joined Saxo, we had 80% of our questions that were about where is my order? And over the span of three and a half years, we did a lot of improvements, very minor improvements. But when you aggregate them together, it results in big results. And it could be something like, where is my order?
28:34
Tue
Link on the front page. It could be the expected delivery date in the order confirmation email. It could be redesigned order confirmation email with a big red button saying, track your order here. It could be a timeline so you can see where it is. And all of this was something that we identified based on the conversations that were coming in. To be able to make those changes, you need to have buy in from your sales organization, from your marketing department, and from it. So you cannot do it on your own in customer service. And that’s why these models can help you make sense of the data, but it’s still up to the entire organization to do something with it. So you need to have a strategy for collecting the information and also doing something with it when you have found out what you need to improve.
29:22
Rob
Yeah, really great points to so we just went through five ways that we think Chat GPT could be shaping the new world for customer service organizations. But again, interested in what the audience has to say here. What do you think? Are we worried? Should we be worried? We somewhat worried. Not worried at all. Where are we at? Okay, we have answers starting to come in here. And while those answers come in, just a quick reminder for me. We’re halfway through now. We’ve said a lot of stuff. You might have some questions. You’re more than welcome to pop those questions into the Q and A, and we’ll get those when we get to the answer here. Okay, so it looks like we’ve filtered in most of the responses about 70% of you have answered already. So maybe we’ll give this just two more seconds, and then I’ll end the poll here.
30:20
Rob
And this is interesting. I think this is more interesting than expected. So let me share the results with all of you, and hopefully you can see these now. But what I’m seeing on my side is that 8% of you say, yes, I’m worried that Chat GPT could affect my role. It could threaten my job. 42% are somewhat worried. Great. I think those are the people that this webinar is really for, to say, look, we’ve talked about some ways that we think Chat GPT could help humans, could replace humans. And then 50% say, not worried at all. I just came to this webinar for giggles because I wanted to hang out with these guys. So glad to see that some of you are not at all worried. So, really great response from the audience. We all, I think, are concerned about the kind of errors that humans can commit.
31:12
Rob
So we wanted something that was a little more reliable than the audience here. So we also asked, Jatbt, are you coming for my service job? And here’s what it said. Look, Chat GPT says, I’m not here to replace you, but to assist you with various tasks. And I think this goes into a lot of the things that you talked about. Tue customers will always need human touch and empathy. But can Chat GPT take care of some repetitive things? Can it help generate content? Could it provide answers to some common customer questions? Can it be a tool? Chat GPT is convinced that it can be. So that’s its opinion. Tue we’ve heard that you kind of are along the same lines. Mirza interested to see what you think on this question.
32:03
Mirza
I tend to look at Chad GPT large language models, AI in general, as a Swiss knife. And the Swiss knife is useless if you don’t know how to wield it. It’s useless if you don’t know how to apply it in different situations. And that’s essentially what this is. We are definitely at a precipice, at a turning point. There’s no doubt about it. And I think the technology will require all of us to become prompt engineers, or at least prompt aficionados, whatever your flavor. But it’s not going to take away as many jobs, at least in my prediction, as the rumor mill says it will. It will, however, make a lot of jobs better. It will make you more productive. It will make you more effective and efficient and it will enable you to do things that you previously probably couldn’t have done at a certain speed or at a certain quality level that doesn’t just apply to service jobs.
32:59
Mirza
I think that applies across the board.
33:02
Rob
Yeah, just to answer from my side too, look, one of the biggest trends that I’ve seen in the service industry over the last three or five years is that customer service is becoming not just a nice thing for businesses to have, but many businesses are saying customer service is our number one differentiator. It is the main reason that customers shop with us and not our competitors. And so what I’m really interested in is to see how those types of companies use AI to make sure that the resources that they have in the business today can go above and beyond where they’re at and provide even higher levels of service to folks. So I think to me, it’s a huge enabler, and I think there are a lot of great examples at how it’s already doing that. And I think for the next part of the webinar, we’re going to look at some of those examples.
33:53
Rob
Now, one thing I wanted to point out, look, my job is in product and engineering, right? So I’m interested in what can AI do for our product people? What can it do for our developers, for our engineers, for the people that are writing code? And so I’m looking at not just chat GPT, but I’m looking at other AI models that can help enable our organization to build its capacity. And so one of the things that we’ve looked at is a product called Copilot. Copilot is built by GitHub. It helps developers code. It’s your Copilot while you’re coding, it helps you write code. You can ask it questions, you can ask it to reformat things, you can ask it to generate code. It’s super interesting. So GitHub did this survey where it asked a bunch of developers to complete a specific set of tasks. Half of them used Copilot, half of them did not.
34:47
Rob
They studied the result. And what I really took away from this survey was not just that the people that use Copilot were more productive, which is cool, but it was actually in the satisfaction that they derived from doing their job using this tool as part of it. So you can see at the top of this chart, almost 90% of engineers that used Copilot said, I am more productive. So that’s great. Everybody loves productivity. No one’s going to fault that. But the part that I love is this next bit. 75% said the job was more satisfying. 60% said that it was more fulfilling. 60% said they were less frustrated. A huge percentage said they were faster with repetitive tasks. There was less mental effort, less time searching, and more time just in the flow of doing their job. And so when we look at how we can leverage generative AI, there’s a lot more than just productivity at stake here.
35:46
Rob
It’s satisfaction with your job, it’s the ability to do your job with excellence. And I think that’s what we’re really interested in. So, Tue, I want to kick it back over to you. I talked a little bit about developers here. It’s kind of a departure from what we’ve been talking about in customer service. Maybe you can bring us back, tell us a little bit more at how we can use some of these technologies in the customer service space.
36:09
Mirza
Yeah.
36:10
Tue
Thanks, Rob. We really want to help agents become super agents, and part of that is creating effortless experiences. And to do that you need to have agents that have an effortless experience in their job. And it’s not too far from what Rob showed in the previous slide where we take away the repetitive tasks that you have to do every single day. When they are gone, you have more job satisfaction and that ultimately result in creating a better experience. And some of the things that you can do with these model is to identify the contact reasons why customers are contacting you. You don’t have to select it in eleven different drop down boxes to fill out every single time because the model understands the intent, the purpose of why the customer is reaching out and can add that you don’t need to search the whole text or back and forth to find numbers if it’s already there in the text somewhere.
37:06
Tue
You can teach the model to find this order number, tracking number, booking number, reservation number. It can also help you graph replies. It knows what the customer is asking. It also knows, based on the articles in your knowledge base, what you are supposed to write. Of course, you need not to send that in many cases directly to the customer, but it has to go through a human. So you validate that it’s absolutely correct. And it also allows you to put a little bit of you into the response because customers want to feel like they are interacting with a real human being and we’re going touch on that a little bit later. We crave that human connection and we want to automate what can be automated, but we also want to speak to a real human when we need to. It can also help you to optimize some of the processes, automatic tagging and categorization of conversations and then also doing summaries of conversations.
38:06
Tue
And here I think it’s important also to note, to follow up on what Rob and Mirzazvah said, that we are at the beginning of an exponential growth. I don’t think anybody had expected us to be where we are today with Chat GPT just twelve months ago. But look, 612 months into the future because we already see models, training models. So this is going to accelerate and it’s about finding out, how can we use these models, chat, UBT, whatever name you’re using to solve one case really well and then implement it and find other ways within your organization, customer service, other departments where it can solve something. Just don’t try to use Jat to solve everything because then it will most likely fail, but find the niche where it solves something perfectly.
38:55
Rob
Yeah, I think that’s a great point to we should always look for the job to be done right before we decide the tool to use. So I think sometimes, especially when we have exciting new technologies, we get a little overexcited. So thanks for pointing that out here as well. I wonder if you could give us just like, maybe one concrete example of how Dixa is doing this today, using these technologies to turn agents into super agents.
39:21
Tue
Very often in retail we see customers reaching out about an issue and sometimes they reach out multiple times. They might send an email and then they send another email and you need to understand what the previous conversation was about. Sometimes they speak to a chatbot and chatbots fail sometimes, and then they are passed on to a human. And something that creates friction for customers is having to either feel that they’re switching channel or repeating information. So one easy lift that we can do is take long conversations and condense them into a summary. So I can easily see what the conversation that was handed over to me from a bot or what the customer contacted me about yesterday or two weeks ago. I don’t need to read a lot of messages back and forth, I can just click one button. Then I see the information about what the customer reach out and that makes it easier for me to help the customer, which result in a better experience for the customer.
40:19
Tue
So that’s one way where we can leverage these model to again turn agents into super agents because they know what the conversation was about without having to read it, because this is what the model is really good at.
40:31
Rob
Yeah, I think it’s super powerful. I mean, I know for me, my least favorite thing is having to say ten times why I’m contacting support. And so this is definitely a silver bullet for that sort of thing. So it’s really cool to see this coming very soon, live into Dixa, but we talked a lot about the strengths of AI, but I think one thing that’s important to say is that, look, this isn’t all sunshine and rainbows, right? This is still a technology that we’re struggling to understand, we’re struggling to make work. And sorry for the startling picture of the frog, but I think this is a great example, right? When we think about the problems that AI has, sometimes it does something we call hallucination. And so one AI image generator got this prompt, I just want a frog, please. And it got a frog it may not be the frog, that it’s not the frog of your dreams.
41:30
Rob
It’s not the response you were expecting. Right. But I think we could all say that this is a frog, and I think that’s a really great thing to think about when you want to employ AI in your service. Where are the places that someone just wants a frog, and then they get this, and that’s an experience that means they’re no longer your customer. So we do need to use some caution when we use AI because it’s just not always accurate, and I wanted to talk a little bit about why that is. So, these are probability models, right? And they’re based on humans, and humans also have weird, complicated beliefs about maybe what a frog is. But one example that I really liked that made a lot of sense to me, and hopefully it makes sense to you is if you go in and you ask Google what happens when you smash a mirror, essentially the same thing happens when you go in and you ask Chat GPT what happens if you smash a mirror.
42:29
Rob
At the bottom here, you see some different responses that you get, and those responses are based on the size of the model, the number of parameters that the model is using. Right. So, with a relatively small model, you get a response that’s correct, but it’s not that informative. What happens if you smash a mirror? Okay. You smash a mirror, like, yes, technically, this is true, but you didn’t really wow me. This isn’t great. The next one, you make it a little bit bigger, and you get a response that’s fairly correct. So, a mirror is glass that reflects light. So if you smash it, maybe you can’t see anything at all. Okay, great. The model gets bigger. You smash the mirror, it shatters into a million pieces. Okay? Probably true. It’s a little more illustrative. That feels more like natural speech to me, which is what this model is really geared to deliver.
43:18
Rob
And then finally, you go for the one that is the most sort of natural, speech based, biggest model. And it tells me, if you smash a mirror, you have seven years bad luck. Now, that feels really wrong to us, right? That’s an incorrect answer. At least most of us believe that’s an incorrect answer. But if you ask Google, it thinks the same thing. If you smash a mirror, you will get seven years bad luck. It’s not a strange result. Turn up on Google, right? So, these large language models do mimic human behavior in some interesting ways, and sometimes that makes them hallucinate answers that aren’t necessarily true just because they sound like natural text or they sound like they could be the right answer, because that’s what the model is really geared to deliver. So accuracy continues to be an issue here. It’s something that we’ve looked a lot at.
44:09
Rob
Dixa Mirza I think a little later, you’re going to talk about how we address this, but the problem with these models doesn’t necessarily end with accuracy. And Mirza, as our in house expert for these technologies, I wanted to kick it over to you to talk about some of the other issues that we’re seeing with Chat GPT. Not necessarily them particularly, but large language models in general.
44:35
Mirza
Yeah. So what happens when you kiss the AI frog? Does it turn into a prince? Not sure because of a couple of problems that exist with the technology today and some of them that have a very clear impact on diversity and inclusion as a topic and as I think we’re all striving for it, but also a number of ethical concerns that come with the technology as we strive and attempt to understand it. So the first thing Rob already mentioned hallucinations. We also refer to those as coherent nonsense, because large language models can produce everything from racist content to entirely made up statements that are presented as fact. This is a real challenge of the technology that many companies, including OpenAI, is working really hard to overcome, primarily through reinforcement learning and a number of other techniques that involve humans making the models better and smarter, essentially.
45:39
Mirza
The other thing, since there are humans involved, is inherent biases. Large language models are hoovering data from the Internet, and they’re kind of creating an average of that data. So that means that human bias is simply unavoidable. It is, however, often presented to sound like it’s the truth in a response by a language model. Sometimes this is very direct, sometimes it’s more subtle, like it is a human interaction and conversation. The other thing is that we’re at the beginning of this, so it’s impossible to predict what the impact on society this will have. We’re in a bit of an AI arms race right now. OpenAI kind of they dove right into it. Google has been developing the same technology for a long time. So has Facebook. So have others. They were a lot more reluctant to launch it. And the models themselves are so large at this point that we can’t understand what they’re actually doing with the data.
46:46
Mirza
They sometimes hallucinate more than just facts. They hallucinate whole languages, which is quite a challenge for us. And then, lastly, that brings about a number of ethical concerns. Generative AI. As we talked about earlier, can create all kinds of content, including video. This means that people can use generative AI with malicious intent, creating deep fakes, creating fake news, spreading misinformation, or mimicking humans, their voices, their knowledge, whatever, for fraud and scams. So these are real challenges that we as an industry are trying to tackle in a number of different ways.
47:30
Rob
Yeah. And maybe you can talk just a little bit about how we’re tackling this at Dixa, I think from maybe just kind of a broad perspective that other people might be able to think about when they use AI in their applications. Yeah.
47:44
Mirza
So at Dixa, what we’re trying to do is guide ourselves, use a number of principles that guide us in how we’re building, and using generative AI as tech. The first major principle I’ve mentioned it earlier is all about human connection. So we’re building technology that enables human connection. Customers often identify with the brands they love, they’re looking to connect with your brand, and they aspire to be a part of the vision you’re creating. Dixa is all about creating personal relationships. So our tools need to put that connection first when adopting new technologies, where we are putting customer experience first and agent experience right behind it. We know customers have better experiences where they connect with a happy agent and that agents are, as tool also talked about earlier, more powerful advocates for your brand when they’re also given compelling experiences themselves. So what we’re trying to do is we look for opportunities for automation to work hand in hand with humans, but never to get rid of human connection completely.
48:51
Mirza
This also helps us validate, verify, and make sure that some of these problems and concerns that I talked about earlier can be overcome because there’s a human to kind of do, let’s say, a QA, so to say. The other thing is our principle of garbage in, garbage out. Most AI driven tech is trained on a corpus of raw and unvaluated data. At Dixa, our AI strategy is different, where we’ve made substantial investments in knowledge and quality assurance to ensure that we have access to highly structured and highly clean data that we can train our models on. So a model trained on substandard conversations will deliver an output that is just that substandard. That’s why integrated knowledge and QA are really critical to any sane automation strategy. QA isn’t just important for model training, it also provides a mechanism to make sure that your automated customer interaction lives up to its in person counterparts.
49:55
Mirza
And it also means that you can identify where automation is failing your customers and work to improve that. Then lastly, I like the wording on this one. It’s more power than wow. Look, not every technology is successful at launch. Some great launches end in spectacular failures. That’s why we’re more interested in the power technology delivers than the wow factor behind it. When looking to leverage this kind of tech, we’re focusing on the value it creates for users today, not the promise of tomorrow. So essentially what that means is we’re being very cautious about the hype, but at the same time trying to stay quick to adapt wherever we can prove value and some kind of staying power for the technology. This hopefully ensures that we’re on the bleeding edge, but that you, our customers, are safe from the splatter. Back to you, Rob.
50:53
Tue
Yeah.
50:54
Rob
Good way to end that one. Great. That’s it. That’s all we have prepared for you today. Thank you. All. So much for joining us here. Few takeaways look generative AI. It is a force multiplier for your organization. It is something that makes you better, that makes you stronger. Leveraging this technology can actually help you achieve the goals that you’re trying to get to. It’s not a replacement for you. It’s a way to make your team better. So we are going to stick around for a little bit of Q and A. But before I get into the Q and A, I would be very remiss if I didn’t invite you all, especially those of you who are in Europe. We have our flagship European event. It’s our Dixa connect. It’s where we get together with customers, prospects, experts in the CX space here at our office in beautiful, sunny Copenhagen.
51:49
Rob
So if you would like to join us, you can register below. Just hit that QR code. I’m going to leave it on the screen for a second so that you can do that. You can also visit us at our website if you’d like to learn more. And I think I’m just going to leave it here since this way you have access to that QR code while we hit a few of the questions that have come from you. So let’s see. All right, so here’s a great question, and, Mirza, maybe I’ll let you answer this one. So, we have one from one of our customers here. We’re already using Dixon in our organization. How can you help us start to leverage these technologies? Mirza, you want to share?
52:35
Mirza
Yeah, sure. So, over time, we will be launching a number of things that will enable you to leverage this technology, whether you’re aware of it or not. Of course, we’re all very transparent about this, but if you have specific questions about how you can use it in your day to day, please reach out to us. I’m always available, so find me through your customer success manager or account manager, and I’m happy to talk to you directly about any challenges or questions you might have about the technology. But again, ourselves very interested and keen on the tech. So more to come.
53:15
Rob
Yeah, I think that’s a great answer. And look, we’ve already introduced our first feature that uses the technology behind Chat GPT to auto generate answers based on knowledge base articles. So if you’re not on the waitlist for that already, you can definitely do that. Please ask your CSM about that. And as you saw today, summarization is also coming soon. So these technologies are appearing in Dixa as part of Dixa. And, of course, as Mirza said, if you have unique ways that you’d like to leverage these, please reach out and let us know, and we can work with you to make sure that happens. Okay, let’s see. I think maybe we have time for just one more question here. And let’s see. One of the questions I have here is, could this be used to do quality auto scoring. So sentiment prediction models? Something like that.
54:18
Rob
Mirza, do you want to take that question?
54:20
Mirza
Sure. Short answer?
54:22
Rob
Yes.
54:23
Mirza
Sentiment detection is part of what language models are able to do. This is essentially the beauty of what they do because they’re able to analyze human interaction and understand it and predict not just what the next best answer should be, but also what the meaning and sentiment behind an interaction is. And maybe, rob, very quickly, there’s one question that has quite a few likes in the Q and A. So I think perhaps it would make sense to also try to tackle this one. It says, how will you navigate the data information the Chat GPT gives out in relation to GDPR and incorrect or offensive information?
55:08
Rob
I think Tue raised his hand for this one. Tue, do you want to take that?
55:12
Tue
Yeah. We know that there are two things that we need to be aware of that we cannot share personal or sensitive information with actors that don’t always have the best intents. We need to ensure that it’s only shared anonymously and with agreement where it’s not used for anything that we don’t want it to. And that’s still a gray area now. So that’s also why we are exploring how we can do it in a safe way. When it comes to incorrect information, we are also looking at it. If you feed the model information that is coming from your knowledge base, then you can trust the information. But if you also open it up to take information that is coming from other sources, then you can’t trust it anymore. Because like Rob said, it doesn’t know right from wrong. And it can be very authoritative in the way that it explains something that makes you believe it, even though it’s not true.
56:05
Tue
So we are following this really closely. We are also speaking to our customers to understand what will make them feel secure in actually using this technology within their environment. So this is a joint effort, I think, in general in the AI industry for us as a vendor, but also together with our customer.
56:26
Mirza
I think I’d also very briefly add to that Chad GPT and GPT Four, that’s just one out of many models out there. And what we’re doing is experimenting with a variety of models that are and approaches to using those models so that there’s less and fewer hallucinations. So there are ways in which you can kind of limit this output and make them function a bit better.
56:56
Tue
Tracy has a good follow up question to this. What are the biggest risks of generative AI? How come some people say that it could threaten humanity? How would that happen?
57:07
Mirza
Well, wow.
57:09
Rob
Okay.
57:09
Mirza
I mean, that’s a huge question. I am happy to answer it. The honest truth is that we don’t know what the biggest risk of generative AI are because we are still at a point where, like I said, the amount of data that goes into that we feed into a model is so vast that we as humans cannot comprehend the amount of information. We also sometimes cannot comprehend what so called emergent properties. This is what you call it in the field. For example, Google’s barred model. They asked it a question in I forget which language it was, but they asked it a question in a language that the model was not trained in. It only gave it a few words, a few sentences, and the model was somehow able to answer in the language. It was never taught. It was never trained to speak. So we still don’t entirely understand how that happened.
58:07
Mirza
And that’s what we call an emergent property. So I think that’s possibly where sometimes the conversation about the threat comes from is because we don’t always understand how it does certain things and how it’s able to produce things. So it’s kind of the stuff of Sci-Fi, I suppose.
58:25
Rob
Yeah. And just super simply, right. We have something that can generate stuff that’s connected to the Internet and everything is connected to the Internet. You put a technology that’s connected to the Internet that can generate code or malicious code could hack things. Who knows, right? Okay, we’re going to have to end it there. We’re already at time, sorry. The end of humanity was kind of a rough place to end it. Don’t think it’s going there, at least not today. So today still focused on how we can leverage this great technology to make everyone more successful in the customer service field. If you have more questions about it, please don’t hesitate to reach out to us here at Dixa. You can visit our website, hit us up on social, reach out to us through your CSM if you’re already a customer or just give us a holler on LinkedIn.
59:16
Rob
All of us fine folks are there. Thank you again so much for attending. Have a really pleasant day, evening wherever you are, and we’ll talk to you again soon. Thanks so much.
59:28
Tue
Thank you for joining us.
59:29
Mirza
Thank you, everyone.
59:32
Ciaran
Thanks for listening today’s episode of Customer Friendship Conversation. We hope you enjoyed this break from our usual pace. The show will return to normal next episode when I’ll be sitting down with another expert in the field of customer experience. If you’ve enjoyed the show, then make sure you’re following us on your podcast platform of choice. It means that you’ll get notified each time we release a new episode, so you won’t miss out on any of the other amazing customer friendship heroes we’ll be showcasing in the coming months. Of course, a rating for review is a huge help to the show, so we always massively appreciate those as well. And if you’re interested in learning more about customer friendship, then head to Dixa.com to discover everything you want to know about customer experience as it’s meant to be. I’m Ciaran Nolan. And thanks for listening.