Applications of Generative AI in Customer Experience | Episode 3 | Learning Kerv | Learning Kerv

Description

In this episode, we explore into the transformative power of generative AI in the realm of customer experience. Join us as we discuss the capabilities of AI and how it is reshaping the way businesses interact with their customers.

Key Highlights:

Key AI Capabilities: Discover the key features of generative AI that make it a game-changer in customer service, from natural language processing to real-time analytics.
Pre-Interaction AI: Learn how AI enhances customer experiences even before the first interaction, through predictive analytics and personalised recommendations.
Live Interaction AI: Explore how AI assists during live interactions, providing real-time support to both customers and agents, and improving the efficiency and accuracy of responses.
Post-Interaction AI: Understand the role of AI in post-interaction processes, such as feedback analysis, sentiment analysis, and continuous improvement of customer service strategies.
Compliance and Regulations: Delve into the critical role AI plays in ensuring regulatory compliance in customer interactions, safeguarding data privacy, and maintaining ethical standards.

Whether you’re a CX professional, a tech enthusiast, or just curious about the future of AI, this episode is packed with insights and practical examples of how generative AI is revolutionising customer experience. Tune in to learn how to harness the power of AI to deliver exceptional service and stay ahead in the competitive landscape.

If you want to talk to us further on this, please don't hesitate to contact us: Contact Us for Inquiries, Support, and Business Collaboration | Kerv

Hosted on Ausha. See ausha.co/privacy-policy for more information.

Transcription

Paul Cox
Hello, and welcome to The Learning Curve, the podcast from Curve that delves into the latest developments in information technology and explores how organisations can put them to work for the good of their people, their customers, society and the planet. My name's Rufus Grigg, and in this series, with the help of some very special guests, we're looking at all things generative AI. So far, we've covered the background, the core technology and common techniques, and now we turn to real-world practical applications. Starting this week... with generative AI in customer experience. And I'm joined this week by two brilliant colleagues from Curve's very own customer experience practice called Curve Experience. Firstly, I've got Paul Cox. Paul is our principal CX consultant. Paul, tell us a bit about your role at Curve. So I spend my time talking to customers, trying to understand their contact center challenges, their business strategy, and then helping them to use our experience to guide them best to use that technology to reach their business goals.
Rufus Grig
Brilliant. Thanks, Paul. And without outing you too much on an age perspective, you've spent quite a bit of time in the CX and contact center world, I understand.
Paul Cox
I'm coming up to my 25 year anniversary of life in contact centers. So yes, I've been here for a while.
Rufus Grig
And yet you maintain your youthful vigor and appearance.
Paul Cox
So I still love it as much as I did when I started. Yeah,
Rufus Grig
there's a portrait in your attic somewhere that looks pretty shocking. but great to have you with us, Paul. And I'm also joined by Sean Lindsay, who's Principal Solution Architect also with Curve Experience. Sean, just give us an idea of your role and your day-to-day.
Sean Lindsay
Thanks, Rufus. So my role is Principal Solutions Consultant here at Curve Experience, and I predominantly work on the Genesys Cloud, so customer experience platform. And my day-to-day role is working with customers enhancing their CX journeys. So understanding their flows, their struggles, the pain points that they might be having, and how the contact center platform can essentially help them. And again, a key focus and hot topic of all areas is generative AI and AI in general. So therefore, again, using those platforms to enhance their customer service.
Rufus Grig
That's brilliant. Thank you very much, Sean. So as I hope you can hear, we've got absolutely the right people with us to steer us on this episode through how generative AI applies to customer experience. Before we actually delve into the applications, I'd really just like to explore what are the key capabilities of generative AI that are making most of the running. So Paul, first, can you pick something for me that's helpful?
Paul Cox
So generative AI is all based around large language models, and the key is in the language piece. So generative AI is unbelievably good at understanding what people are saying, and then generating language. That could be from one language to another. So translating from French to German, it could be just understanding what people say so that it can work out or the agent can work out what to do with that intent. Or it could be to generate some text, like a knowledge article or indeed a response to a question.
Rufus Grig
Brilliant. Thank you, Paul. Sean, anything from you?
Sean Lindsay
Yeah, so there's quite a few products that I'd like to mention. So in terms of generative AI, you've got... Key features like speech-to-text analytics that kind of leverages an AI model, specifically focusing on an intent miner, what we would call an intent miner, which gathers or analyzes conversations. So it'll be historical conversations. And with the use of AI itself, we are analyzing those conversations to look for topics and phrases, again, taking that automation to the next level, saving the business time and money, essentially, in that sense. And then one other element that we would like... to kind of highlight is going to be around summarization elements. So again, knowledge articles being surfaced to agents in real time, as well as the just general AI models, so AI bots for voice and digital.
Rufus Grig
Okay. So interestingly, although clearly we think of generative AI as creating text, what I'm actually hearing from both of you is it's the ability of gen AI or large language models in particular to understand what's being said that seems to be one of the more... important and useful pieces. So when we had a bit of a chat before this episode, we thought we'd structure our discussion into the sort of different phases of an interaction. So starting on what happens before an interaction, maybe even trying to prevent an interaction from needing to take place. We'd look at how Gen AI can apply to fully automated interactions where there's no human involved. Moving on to how AI can be used to assist agents with live interactions, whether they're talking or chatting by text with a customer. Then looking at what happens post-interaction, once the customer has hung up the phone or disconnected from the chat, what happens there. And then we're going to finally have a little bit of a look about the role of generative AI in assuring regulatory compliance and quality. We'll start with that pre-interaction world. And Paul, I guess given that... In many ways, most contact centre interactions these days, or a lot of them certainly, are really indicative of a failure of someone to be able to transact something online. It's often a sign that something has gone wrong or the self-service wasn't quite good enough. How does generative AI help keep people away from the contact centre altogether?
Paul Cox
I guess the first thing that's really important that gen AI can do for us or can improve for us is... understanding what it is that went wrong. Everyone who's heard me speak before knows that I'm a little bit obsessed with the importance of triage. So the first thing that Jenny and I can really help us with is when somebody says to us, here is my problem, then using that language capability to really understand what they mean and therefore what it is that needs to be done to help them. The second part is then to understand what the answer might be. And traditionally, in recent years, we use knowledge bots to understand somebody's intent, and then map that onto a potential answer. So someone says, you know, I'm having problems configuring my new iPhone. And therefore, we put up an article about that, and play that back to them. If we're using generative AI, then one of the things that this gives us on top of the traditional model is being able to have a more nuanced answer. So we might listen to what somebody's asking and realize that actually there are two or three different articles in our knowledge base that might be able to answer their question. Generative AI can read those articles and then generate a new response that's personalized just for them to answer that question. And all of that in answer to your question adds up to a much higher probability that we'll be able to solve their challenge and remove the need for them to speak to a live human being. So therefore reducing the cost to serve.
Rufus Grig
So would I be right in summarising that as a better way of understanding precisely the challenge that the customer has and then the ability to construct a more personalised answer to help solve their specific problem?
Paul Cox
Yeah, absolutely. Absolutely.
Rufus Grig
Okay, great. So I can absolutely see how that... does actually remove a fair number of potential needs to interact. So Sean, Paul mentioned triage. How in that sort of setup call, once we determine that a customer does need to speak or chat with a live agent, what aspects of generative AI are helpful in getting the best outcome in that scenario?
Sean Lindsay
It's a great question. So as Paul alluded to, when we're going through that journey of the IVR and triaging the user to get through to the team. There's a range of elements that we would utilize in terms of AI and generative AI. So one of the key elements that we might want to look at as a first glance is the reason for contact. And based on the reason for contact, we would then leverage a predictive routing, so it's an AI routing element, which then analyzes the conversation based on the reason for contact and based on the conversation and looks for the right team to handle that interaction. So instead of that customer going through the IVR, selecting it, journey, so selecting one or selecting an option within the IVR itself or message flow, interacting with a bot, the system will analyze that conversation, leverage AI, and distribute that interaction to the right team with the right skills. Now, we can take that a little bit further as if we understand the reason that we know the system says, okay, based on the predictive routing model, we know it needs to go to team A. We can then also look at conditional skills to take it a level deeper. To say, yes, we need to go to team A, but within that team, I'm looking for a very specific skill. So we can then tag that interaction with the specific skill sets and look for the right skilled agent. And that's all leveraging AI and all happens within milliseconds of the platform. So again, offering a great customer journey in that IVR stage, that understanding stage, and triaging that or transferring that to the right team with the right skills to handle that query.
Rufus Grig
And that's brilliant. Thanks, Sean. Let's have a bit of a think about... live chat before we get into talking about agents. I know when ChatGPT first sort of raised its head and entered public consciousness, there was a lot of talk about how it felt very like the sort of interaction we might have with a company on a sort of live chat scenario. Paul, perhaps you could explain sort of what's going on there with a live chat and is generative AI really up to the job there? Does it have a role to play? Is it going to improve the likely outcomes or experiences for chat?
Paul Cox
So I think we have to consider the different types of chat that we might have. When somebody makes contact with a bot, either it's because they have a question or because they have a request. In the scenario where they have a question, then Gen AI's ability to understand the triage piece we were talking about, and its ability to formulate those responses from multiple articles and personalize the response is going to give a much more human feel than a traditional bot would, where it is just giving you a canned response. It's looked at what you've said, it's trying to map it to an article. And then it's handed you back the article. So it is very much, if you've asked the standard core question, fantastic. If you have deviated very slightly from what the knowledge author had in mind when he wrote the article, then it's going to feel a bit weird. It's going to feel a bit unnatural. So whereas Gen AI is going to understand your knowledge articles, and it's a really important point that you can absolutely focus it on, say, only give me answers. from this knowledge base, from these articles. Do not go out to the internet, go out to your training and make stuff up. Only get information from here, but it will use that to construct a new answer.
Rufus Grig
That's what we talked about in an earlier episode as grounding and making sure the model is grounded in using only the approved and authored text for your brand and for your service line.
Paul Cox
Yeah, absolutely. Absolutely. For any AI geeks out there. This is a grounding method called RAG. So this is retrieval augmented generation, specifically that we're talking about here. So that's number one, where we're trying to answer a question and it definitely is going to make a more human experience, a better customer experience there for relatively low risk if we take this approach with a request where I say, I want to book a holiday, for example. And it's going to take me through a number of steps and ask me questions. There is an argument that it might be a slightly more human-like interaction. But actually, you know, really to do this sort of thing, you have a number of set pieces of information that you need to capture. And using Gen AI for that kind of experience, I would say the gains are relatively minimal compared to just going through a flow, a traditional flow. where the bot asks you one question after another.
Rufus Grig
Great. Thanks very much, Paul. Really interesting to see how chat, or certainly that automated chat, is evolving and how Gen AI is breathing sort of extra capability. But Sean, let's take us to the scenario where the automated chat can't do it. We are having a live interaction between a customer and an agent, whether that's speaking with speech or whether it's a chat where the agent is typing back. How is generative AI helping to assist that agent in the handling of that call or that chat session?
Sean Lindsay
Yeah, very good question. So within the platform itself, generative AI plays a huge role in assisting the agent either on a digital channel or a voice channel with the articles, so the knowledge-based articles that are present within the system. So surfacing that knowledge to the agent in real time based on the customer interaction, so what they're saying, the text. So the customer could be talking about a delivery, for example, or where is my delivery or something more specific. Now, based on the conversation and the wording being used, the phrases, we will then be able to surface articles to the agent in real time, leveraging generative AI. This has huge benefits because here we are again, providing that knowledge to that agent, again, taking that guesswork out of everything, lowering average handle times, and again, empowering the agents with the information that they need. They no longer need to go and look at back office systems for the relevant articles or relevant documentation to try and find the answer. The system and generative AI itself is now surfacing all of that content over to the live agent in real time.
Rufus Grig
And that works not just with chat, but can that work with a voice call as well? Is it listening and understanding the voice conversation too? Yes,
Sean Lindsay
absolutely. So that's both on digital interactions and voice interactions. So exactly what I've just explained across both channels.
Rufus Grig
Okay, no, that's brilliant. I can't remember if it was you or Paul earlier also mentioned the ability of a large language model to translate. How are we seeing that sort of language translation play out here?
Sean Lindsay
Yeah, this is quite a big topic. So let's start with the web messaging or digital channels first. So first of all, when we're considering an interaction coming in, let's say a customer was coming in from Germany, for example, they go through the IVR, so the bot journey or voice IVR, and then we transition. over to a live agent. Now, unfortunately, there's no live agents that speak German. So what we can do here is we can transition that over to the next best skilled agent. and that could be an English speaking agent. And what we can then leverage is translation tools on both voice and digital. So again, when we're talking about large language models, we're looking at the accuracy of the translation. So here in the, starting with web messaging, we can leverage a text-based translation, meaning customer will type a message. It'll automatically translate to the agent's native language. Agent can then see that text, understand that text in their language, and then ultimately respond. back in English, which will then automatically translate that into the customer's native language. In this example, it'll be German. Now, when we're leveraging the AI model, so the generative AI, we've also got to take into consideration the accuracy, as well as that consistency of receiving interactions and again, using that data for training purposes. That'll then increase the accuracy of those overall translations.
Rufus Grig
That's really interesting. It takes me back to the Hitchhiker's Guide to the Galaxy and the the Babel fish in the ear, the technology that seemed, when I read that as a child, so remote and now we're actually using it day in, day out. That's fascinating. Okay, so I can absolutely see how we're seeing massive benefits in terms of supporting the agent during that live interaction. What about post-interaction? The customer's hung up or the chat session's finished. Paul, what sort of technology is being put to use there?
Paul Cox
Sure. So this is actually one of the most exciting areas for me, because it's something that we're seeing being used very quickly. So it's one of the first parts of Gen AI to be adopted. The first one that we're seeing is summarization. So taking the transcript of the call or the chat, we've been for a while able to take that transcript and push it into a CRM system. But for the next agent, to have to read through what potentially could be a five or 10 minute calls worth of transcription and try to understand what was going on is really tough. I'm sure all of you out there as consumers have called a contact center, identified yourself, and then immediately been put on hold while the agent reads through your notes. And then there's nothing that makes a worse customer experience. So summarization allows us to condense that transcription into a short passage. paragraph that's easy for the next agent to understand, and then push that summary into the CRM to give not only a reduced handling time by shortening the wrap-up, but also to give a much better experience to the agent on the next call and a much better experience to the customer. The other thing that really helps is the outcome coding. So one of the things that we've struggled with for years is that If you give an agent five or six outcomes, they generally pretty consistently choose the right one out of those six at the end of every interaction and code interactions in a consistent way that gives you some good stats. But really and truly, you don't want just a very general one out of six. You really want one out of about 20 or 30 or perhaps even more. And where contact centers have tried to expand that list or make it a two-tier list, the accuracy and the consistency of agents selecting the right option tends to fall proportionately. With AI, the AI can look at that transcription, understand the key topics from the interaction, and then automatically code what that interaction was about. And it might be that it's about two or three things and it can identify those key topics that were discussed during the interaction. And that is a much more accurate way of really understanding what the interaction is about. That is a huge leap forward in terms of MI. So,
Rufus Grig
Sean, the final part of our interaction journey where everything's finished and we're into the quality and compliance management. Can you just talk a little bit about some of the applications in? that area of CX.
Sean Lindsay
Yeah, absolutely. So this is a very vital piece and this has huge benefits within the business. When we're talking about compliance, there's a range of elements that we can consider within the CX platforms. So once that interaction has been completed, we have the transcript and we have the reason for contact. So we know the category. Now, when we're talking and focusing on compliance or quality assessment, we can then look at quality assessment forms that need to be completed traditionally. by supervisors. Now with leveraging AI and generative AI, we can now look at automating that quality assessment scorecard. So what this does is it looks at the sentiment analysis, so based on the sentiment that you have configured, and also looking at the positive and negative parameters around the sentiment, as well as your topics. What that'll allow us to do is to auto score each of those questions in the quality assessment form based on the phrases used during. the conversation from both the agent side and the customer side. This is great because again, we're saving time. And again, if we do have an element where that needs to be reviewed by a supervisor that can absolutely be done so, AI is doing that work. You can scan through that, check it, make sure it is in fact correct. And you can optimize those journeys and those phrases using that intent miner that I referred to earlier. So the intent miner again, is analyzing those conversations, taking all the guesswork out of understanding or guessing what customers may say, and looking for those phrases that customers are using within the contact center. Now, we can also take that a step further where we're looking at detection of fraudulent patterns, as an example. So based on the phrasing used during a conversation, we can flag up key elements. So key phrases, maybe focusing on fraudulent elements there, and we can actually highlight those in the conversations and where those took part within that conversation. You can then add that into... processes within the CX platforms.
Rufus Grig
Really interesting. I love the phrase you use about taking the guesswork out. I suppose if you've got 100,000 interactions to look at and you're sort of sampling, the chances of finding what you're looking for are quite needle in a haystack. Whereas if we've understood really what's going on in each call and interaction, it's easier to hone in on the areas where you think you might have issues or certainly need to look at the quality. Really interesting.
Sean Lindsay
Absolutely. I mean, when we're talking about compliance and also knowledge itself, I mean... we want to analyze the conversations because guessing what customers might say is really hard and very time consuming.
Rufus Grig
Now, that's great. Well, I think we've been through a whole journey of an interaction from trying to remove the need for a live interaction to how do we support them? And then how do we ensure the quality and compliance after? Just, I suppose, to wrap up all, you know, you're with CX, customer experience leaders day in, day out. Just how much interest or fear or excitement is there about Gen AI? How real is this?
Paul Cox
I think there are two extremes that I encounter fairly regularly, actually. The one extreme is where we're trying to help people at the beginning of their AI journey with maybe something that isn't using any generative AI, but they've read a lot of stories in the media about hallucinations. And they are terrified of any kind of AI that might be customer facing because they're worried it's going to start making stuff up. And I think, Rufus, you told me a story the other day about a McDonald's self-service offering people bacon-flavored ice cream. And it's stories like that that really stick in people's minds. And they kind of feel like any kind of chatbot or voicebot is going to be dangerous for their brand potentially.
Rufus Grig
Interesting that that's still a very prevalent attitude. And I suppose we are still only sort of 18, 24 months on from this all becoming mainstream. But there's excitement too. You have the people at the positive end of the scale.
Paul Cox
Absolutely. And sometimes too much. So there are definitely people who I talk to who in their mind have decided that the sort of scripted intent-based bots that are most prevalent out there today are last year's technology, and they're not really interested in it anymore. And therefore, if they're going to implement any kind of bot from scratch, they only want to do generative AI, because that's the new version, and the old version would be a bad investment. So what we try to do is encourage people to think of Gen AI as a turbo booster on top of their traditional AI. It's something that makes traditional AI. just that little bit better, that little bit more human, that little bit more personalized, rather than it's got to be Gen AI or it's got to be traditional AI, and you've got to choose between the two. But there is an awful lot of misinformation out there, which is what happens when you have this sort of media hysteria. I think we've had over ChatGPT over the last six months or so. We do a lot of myth busting with people. before we can start to strategize in terms of what's right for them.
Rufus Grig
And I suppose there may be people out there who'd like the idea of bacon flavored ice cream. And thanks, Paul. And Sean, you're working with the technology. Keep us grounded in reality. How ready for primetime is this? Are people really doing this day in, day out?
Sean Lindsay
From my point of view and working in the CX industry, we're seeing it being rolled out and we're seeing it becoming available in the platform each week, coming out at an incredible pace. And there's some real great features that are out there today. To specifically answer your question, I think you've got to be very careful where we leverage generative AI and how contained that is. There is some scary stories on the market where people are introducing generative AI into their let's say, web messaging interfaces where customers are going onto the web chat and manipulating the answers that the generative AI is feeding back, which can hurt their brand. So yes, it's in the space, it's in the CX contact sensors today, and that is rolled out in a controlled manner. If you are introducing a generative AI platform, you want to have that in a controlled environment because it has the ability to really make up answers that may not be correct if it's not contained, surfacing old content. which can then again hurt your brand, maybe make promises to customers that actually are no longer possible. So again, you've got to be very careful of that. So my view of it is the contact center platforms are rolling this out and the features are incredible. So yes, I would say it's absolutely ready when rolled out in the right way.
Rufus Grig
So make sure it's rolled out in the right way. Make sure you work with a partner like Curve, who can help you and steer you and keep everything grounded and protected for your brand. Paul and Sean, thank you so much for your time today. It's been really fascinating listening to everything you've had to say about the use of generative AI in customer experience. If you've been interested in what we've said, then please do get in touch. Tell us what you think. You can find out all about Curve by visiting curve.com. Please do listen out for the next episode. You can subscribe. You can tell all your friends. Again, thank you very much to Paul and Sean for your contributions. Thank you for listening. And until next time, goodbye.

Chapters

Introduction to the speakers
37sec
Overview: Key capabilities of Generative AI
2min
The role of AI in CX: Pre-interaction with customers
5min
The role of AI in CX: Live-interaction for customers
9min
The role of AI in CX: Live-interaction for agents
12min
The role of AI in CX: Post-interaction with customers
16min
The role of AI in Regulatory Compliance: Post-interaction with customers
19min
How ready is AI for customers?
22min

Description

Key Highlights:

Key AI Capabilities: Discover the key features of generative AI that make it a game-changer in customer service, from natural language processing to real-time analytics.
Pre-Interaction AI: Learn how AI enhances customer experiences even before the first interaction, through predictive analytics and personalised recommendations.
Live Interaction AI: Explore how AI assists during live interactions, providing real-time support to both customers and agents, and improving the efficiency and accuracy of responses.
Post-Interaction AI: Understand the role of AI in post-interaction processes, such as feedback analysis, sentiment analysis, and continuous improvement of customer service strategies.
Compliance and Regulations: Delve into the critical role AI plays in ensuring regulatory compliance in customer interactions, safeguarding data privacy, and maintaining ethical standards.

If you want to talk to us further on this, please don't hesitate to contact us: Contact Us for Inquiries, Support, and Business Collaboration | Kerv

Hosted on Ausha. See ausha.co/privacy-policy for more information.

Transcription

Paul Cox
Hello, and welcome to The Learning Curve, the podcast from Curve that delves into the latest developments in information technology and explores how organisations can put them to work for the good of their people, their customers, society and the planet. My name's Rufus Grigg, and in this series, with the help of some very special guests, we're looking at all things generative AI. So far, we've covered the background, the core technology and common techniques, and now we turn to real-world practical applications. Starting this week... with generative AI in customer experience. And I'm joined this week by two brilliant colleagues from Curve's very own customer experience practice called Curve Experience. Firstly, I've got Paul Cox. Paul is our principal CX consultant. Paul, tell us a bit about your role at Curve. So I spend my time talking to customers, trying to understand their contact center challenges, their business strategy, and then helping them to use our experience to guide them best to use that technology to reach their business goals.
Rufus Grig
Brilliant. Thanks, Paul. And without outing you too much on an age perspective, you've spent quite a bit of time in the CX and contact center world, I understand.
Paul Cox
I'm coming up to my 25 year anniversary of life in contact centers. So yes, I've been here for a while.
Rufus Grig
And yet you maintain your youthful vigor and appearance.
Paul Cox
So I still love it as much as I did when I started. Yeah,
Rufus Grig
there's a portrait in your attic somewhere that looks pretty shocking. but great to have you with us, Paul. And I'm also joined by Sean Lindsay, who's Principal Solution Architect also with Curve Experience. Sean, just give us an idea of your role and your day-to-day.
Sean Lindsay
Thanks, Rufus. So my role is Principal Solutions Consultant here at Curve Experience, and I predominantly work on the Genesys Cloud, so customer experience platform. And my day-to-day role is working with customers enhancing their CX journeys. So understanding their flows, their struggles, the pain points that they might be having, and how the contact center platform can essentially help them. And again, a key focus and hot topic of all areas is generative AI and AI in general. So therefore, again, using those platforms to enhance their customer service.
Rufus Grig
That's brilliant. Thank you very much, Sean. So as I hope you can hear, we've got absolutely the right people with us to steer us on this episode through how generative AI applies to customer experience. Before we actually delve into the applications, I'd really just like to explore what are the key capabilities of generative AI that are making most of the running. So Paul, first, can you pick something for me that's helpful?
Paul Cox
So generative AI is all based around large language models, and the key is in the language piece. So generative AI is unbelievably good at understanding what people are saying, and then generating language. That could be from one language to another. So translating from French to German, it could be just understanding what people say so that it can work out or the agent can work out what to do with that intent. Or it could be to generate some text, like a knowledge article or indeed a response to a question.
Rufus Grig
Brilliant. Thank you, Paul. Sean, anything from you?
Sean Lindsay
Yeah, so there's quite a few products that I'd like to mention. So in terms of generative AI, you've got... Key features like speech-to-text analytics that kind of leverages an AI model, specifically focusing on an intent miner, what we would call an intent miner, which gathers or analyzes conversations. So it'll be historical conversations. And with the use of AI itself, we are analyzing those conversations to look for topics and phrases, again, taking that automation to the next level, saving the business time and money, essentially, in that sense. And then one other element that we would like... to kind of highlight is going to be around summarization elements. So again, knowledge articles being surfaced to agents in real time, as well as the just general AI models, so AI bots for voice and digital.
Rufus Grig
Okay. So interestingly, although clearly we think of generative AI as creating text, what I'm actually hearing from both of you is it's the ability of gen AI or large language models in particular to understand what's being said that seems to be one of the more... important and useful pieces. So when we had a bit of a chat before this episode, we thought we'd structure our discussion into the sort of different phases of an interaction. So starting on what happens before an interaction, maybe even trying to prevent an interaction from needing to take place. We'd look at how Gen AI can apply to fully automated interactions where there's no human involved. Moving on to how AI can be used to assist agents with live interactions, whether they're talking or chatting by text with a customer. Then looking at what happens post-interaction, once the customer has hung up the phone or disconnected from the chat, what happens there. And then we're going to finally have a little bit of a look about the role of generative AI in assuring regulatory compliance and quality. We'll start with that pre-interaction world. And Paul, I guess given that... In many ways, most contact centre interactions these days, or a lot of them certainly, are really indicative of a failure of someone to be able to transact something online. It's often a sign that something has gone wrong or the self-service wasn't quite good enough. How does generative AI help keep people away from the contact centre altogether?
Paul Cox
I guess the first thing that's really important that gen AI can do for us or can improve for us is... understanding what it is that went wrong. Everyone who's heard me speak before knows that I'm a little bit obsessed with the importance of triage. So the first thing that Jenny and I can really help us with is when somebody says to us, here is my problem, then using that language capability to really understand what they mean and therefore what it is that needs to be done to help them. The second part is then to understand what the answer might be. And traditionally, in recent years, we use knowledge bots to understand somebody's intent, and then map that onto a potential answer. So someone says, you know, I'm having problems configuring my new iPhone. And therefore, we put up an article about that, and play that back to them. If we're using generative AI, then one of the things that this gives us on top of the traditional model is being able to have a more nuanced answer. So we might listen to what somebody's asking and realize that actually there are two or three different articles in our knowledge base that might be able to answer their question. Generative AI can read those articles and then generate a new response that's personalized just for them to answer that question. And all of that in answer to your question adds up to a much higher probability that we'll be able to solve their challenge and remove the need for them to speak to a live human being. So therefore reducing the cost to serve.
Rufus Grig
So would I be right in summarising that as a better way of understanding precisely the challenge that the customer has and then the ability to construct a more personalised answer to help solve their specific problem?
Paul Cox
Yeah, absolutely. Absolutely.
Rufus Grig
Okay, great. So I can absolutely see how that... does actually remove a fair number of potential needs to interact. So Sean, Paul mentioned triage. How in that sort of setup call, once we determine that a customer does need to speak or chat with a live agent, what aspects of generative AI are helpful in getting the best outcome in that scenario?
Sean Lindsay
It's a great question. So as Paul alluded to, when we're going through that journey of the IVR and triaging the user to get through to the team. There's a range of elements that we would utilize in terms of AI and generative AI. So one of the key elements that we might want to look at as a first glance is the reason for contact. And based on the reason for contact, we would then leverage a predictive routing, so it's an AI routing element, which then analyzes the conversation based on the reason for contact and based on the conversation and looks for the right team to handle that interaction. So instead of that customer going through the IVR, selecting it, journey, so selecting one or selecting an option within the IVR itself or message flow, interacting with a bot, the system will analyze that conversation, leverage AI, and distribute that interaction to the right team with the right skills. Now, we can take that a little bit further as if we understand the reason that we know the system says, okay, based on the predictive routing model, we know it needs to go to team A. We can then also look at conditional skills to take it a level deeper. To say, yes, we need to go to team A, but within that team, I'm looking for a very specific skill. So we can then tag that interaction with the specific skill sets and look for the right skilled agent. And that's all leveraging AI and all happens within milliseconds of the platform. So again, offering a great customer journey in that IVR stage, that understanding stage, and triaging that or transferring that to the right team with the right skills to handle that query.
Rufus Grig
And that's brilliant. Thanks, Sean. Let's have a bit of a think about... live chat before we get into talking about agents. I know when ChatGPT first sort of raised its head and entered public consciousness, there was a lot of talk about how it felt very like the sort of interaction we might have with a company on a sort of live chat scenario. Paul, perhaps you could explain sort of what's going on there with a live chat and is generative AI really up to the job there? Does it have a role to play? Is it going to improve the likely outcomes or experiences for chat?
Paul Cox
So I think we have to consider the different types of chat that we might have. When somebody makes contact with a bot, either it's because they have a question or because they have a request. In the scenario where they have a question, then Gen AI's ability to understand the triage piece we were talking about, and its ability to formulate those responses from multiple articles and personalize the response is going to give a much more human feel than a traditional bot would, where it is just giving you a canned response. It's looked at what you've said, it's trying to map it to an article. And then it's handed you back the article. So it is very much, if you've asked the standard core question, fantastic. If you have deviated very slightly from what the knowledge author had in mind when he wrote the article, then it's going to feel a bit weird. It's going to feel a bit unnatural. So whereas Gen AI is going to understand your knowledge articles, and it's a really important point that you can absolutely focus it on, say, only give me answers. from this knowledge base, from these articles. Do not go out to the internet, go out to your training and make stuff up. Only get information from here, but it will use that to construct a new answer.
Rufus Grig
That's what we talked about in an earlier episode as grounding and making sure the model is grounded in using only the approved and authored text for your brand and for your service line.
Paul Cox
Yeah, absolutely. Absolutely. For any AI geeks out there. This is a grounding method called RAG. So this is retrieval augmented generation, specifically that we're talking about here. So that's number one, where we're trying to answer a question and it definitely is going to make a more human experience, a better customer experience there for relatively low risk if we take this approach with a request where I say, I want to book a holiday, for example. And it's going to take me through a number of steps and ask me questions. There is an argument that it might be a slightly more human-like interaction. But actually, you know, really to do this sort of thing, you have a number of set pieces of information that you need to capture. And using Gen AI for that kind of experience, I would say the gains are relatively minimal compared to just going through a flow, a traditional flow. where the bot asks you one question after another.
Rufus Grig
Great. Thanks very much, Paul. Really interesting to see how chat, or certainly that automated chat, is evolving and how Gen AI is breathing sort of extra capability. But Sean, let's take us to the scenario where the automated chat can't do it. We are having a live interaction between a customer and an agent, whether that's speaking with speech or whether it's a chat where the agent is typing back. How is generative AI helping to assist that agent in the handling of that call or that chat session?
Sean Lindsay
Yeah, very good question. So within the platform itself, generative AI plays a huge role in assisting the agent either on a digital channel or a voice channel with the articles, so the knowledge-based articles that are present within the system. So surfacing that knowledge to the agent in real time based on the customer interaction, so what they're saying, the text. So the customer could be talking about a delivery, for example, or where is my delivery or something more specific. Now, based on the conversation and the wording being used, the phrases, we will then be able to surface articles to the agent in real time, leveraging generative AI. This has huge benefits because here we are again, providing that knowledge to that agent, again, taking that guesswork out of everything, lowering average handle times, and again, empowering the agents with the information that they need. They no longer need to go and look at back office systems for the relevant articles or relevant documentation to try and find the answer. The system and generative AI itself is now surfacing all of that content over to the live agent in real time.
Rufus Grig
And that works not just with chat, but can that work with a voice call as well? Is it listening and understanding the voice conversation too? Yes,
Sean Lindsay
absolutely. So that's both on digital interactions and voice interactions. So exactly what I've just explained across both channels.
Rufus Grig
Okay, no, that's brilliant. I can't remember if it was you or Paul earlier also mentioned the ability of a large language model to translate. How are we seeing that sort of language translation play out here?
Sean Lindsay
Yeah, this is quite a big topic. So let's start with the web messaging or digital channels first. So first of all, when we're considering an interaction coming in, let's say a customer was coming in from Germany, for example, they go through the IVR, so the bot journey or voice IVR, and then we transition. over to a live agent. Now, unfortunately, there's no live agents that speak German. So what we can do here is we can transition that over to the next best skilled agent. and that could be an English speaking agent. And what we can then leverage is translation tools on both voice and digital. So again, when we're talking about large language models, we're looking at the accuracy of the translation. So here in the, starting with web messaging, we can leverage a text-based translation, meaning customer will type a message. It'll automatically translate to the agent's native language. Agent can then see that text, understand that text in their language, and then ultimately respond. back in English, which will then automatically translate that into the customer's native language. In this example, it'll be German. Now, when we're leveraging the AI model, so the generative AI, we've also got to take into consideration the accuracy, as well as that consistency of receiving interactions and again, using that data for training purposes. That'll then increase the accuracy of those overall translations.
Rufus Grig
That's really interesting. It takes me back to the Hitchhiker's Guide to the Galaxy and the the Babel fish in the ear, the technology that seemed, when I read that as a child, so remote and now we're actually using it day in, day out. That's fascinating. Okay, so I can absolutely see how we're seeing massive benefits in terms of supporting the agent during that live interaction. What about post-interaction? The customer's hung up or the chat session's finished. Paul, what sort of technology is being put to use there?
Paul Cox
Sure. So this is actually one of the most exciting areas for me, because it's something that we're seeing being used very quickly. So it's one of the first parts of Gen AI to be adopted. The first one that we're seeing is summarization. So taking the transcript of the call or the chat, we've been for a while able to take that transcript and push it into a CRM system. But for the next agent, to have to read through what potentially could be a five or 10 minute calls worth of transcription and try to understand what was going on is really tough. I'm sure all of you out there as consumers have called a contact center, identified yourself, and then immediately been put on hold while the agent reads through your notes. And then there's nothing that makes a worse customer experience. So summarization allows us to condense that transcription into a short passage. paragraph that's easy for the next agent to understand, and then push that summary into the CRM to give not only a reduced handling time by shortening the wrap-up, but also to give a much better experience to the agent on the next call and a much better experience to the customer. The other thing that really helps is the outcome coding. So one of the things that we've struggled with for years is that If you give an agent five or six outcomes, they generally pretty consistently choose the right one out of those six at the end of every interaction and code interactions in a consistent way that gives you some good stats. But really and truly, you don't want just a very general one out of six. You really want one out of about 20 or 30 or perhaps even more. And where contact centers have tried to expand that list or make it a two-tier list, the accuracy and the consistency of agents selecting the right option tends to fall proportionately. With AI, the AI can look at that transcription, understand the key topics from the interaction, and then automatically code what that interaction was about. And it might be that it's about two or three things and it can identify those key topics that were discussed during the interaction. And that is a much more accurate way of really understanding what the interaction is about. That is a huge leap forward in terms of MI. So,
Rufus Grig
Sean, the final part of our interaction journey where everything's finished and we're into the quality and compliance management. Can you just talk a little bit about some of the applications in? that area of CX.
Sean Lindsay
Yeah, absolutely. So this is a very vital piece and this has huge benefits within the business. When we're talking about compliance, there's a range of elements that we can consider within the CX platforms. So once that interaction has been completed, we have the transcript and we have the reason for contact. So we know the category. Now, when we're talking and focusing on compliance or quality assessment, we can then look at quality assessment forms that need to be completed traditionally. by supervisors. Now with leveraging AI and generative AI, we can now look at automating that quality assessment scorecard. So what this does is it looks at the sentiment analysis, so based on the sentiment that you have configured, and also looking at the positive and negative parameters around the sentiment, as well as your topics. What that'll allow us to do is to auto score each of those questions in the quality assessment form based on the phrases used during. the conversation from both the agent side and the customer side. This is great because again, we're saving time. And again, if we do have an element where that needs to be reviewed by a supervisor that can absolutely be done so, AI is doing that work. You can scan through that, check it, make sure it is in fact correct. And you can optimize those journeys and those phrases using that intent miner that I referred to earlier. So the intent miner again, is analyzing those conversations, taking all the guesswork out of understanding or guessing what customers may say, and looking for those phrases that customers are using within the contact center. Now, we can also take that a step further where we're looking at detection of fraudulent patterns, as an example. So based on the phrasing used during a conversation, we can flag up key elements. So key phrases, maybe focusing on fraudulent elements there, and we can actually highlight those in the conversations and where those took part within that conversation. You can then add that into... processes within the CX platforms.
Rufus Grig
Really interesting. I love the phrase you use about taking the guesswork out. I suppose if you've got 100,000 interactions to look at and you're sort of sampling, the chances of finding what you're looking for are quite needle in a haystack. Whereas if we've understood really what's going on in each call and interaction, it's easier to hone in on the areas where you think you might have issues or certainly need to look at the quality. Really interesting.
Sean Lindsay
Absolutely. I mean, when we're talking about compliance and also knowledge itself, I mean... we want to analyze the conversations because guessing what customers might say is really hard and very time consuming.
Rufus Grig
Now, that's great. Well, I think we've been through a whole journey of an interaction from trying to remove the need for a live interaction to how do we support them? And then how do we ensure the quality and compliance after? Just, I suppose, to wrap up all, you know, you're with CX, customer experience leaders day in, day out. Just how much interest or fear or excitement is there about Gen AI? How real is this?
Paul Cox
I think there are two extremes that I encounter fairly regularly, actually. The one extreme is where we're trying to help people at the beginning of their AI journey with maybe something that isn't using any generative AI, but they've read a lot of stories in the media about hallucinations. And they are terrified of any kind of AI that might be customer facing because they're worried it's going to start making stuff up. And I think, Rufus, you told me a story the other day about a McDonald's self-service offering people bacon-flavored ice cream. And it's stories like that that really stick in people's minds. And they kind of feel like any kind of chatbot or voicebot is going to be dangerous for their brand potentially.
Rufus Grig
Interesting that that's still a very prevalent attitude. And I suppose we are still only sort of 18, 24 months on from this all becoming mainstream. But there's excitement too. You have the people at the positive end of the scale.
Paul Cox
Absolutely. And sometimes too much. So there are definitely people who I talk to who in their mind have decided that the sort of scripted intent-based bots that are most prevalent out there today are last year's technology, and they're not really interested in it anymore. And therefore, if they're going to implement any kind of bot from scratch, they only want to do generative AI, because that's the new version, and the old version would be a bad investment. So what we try to do is encourage people to think of Gen AI as a turbo booster on top of their traditional AI. It's something that makes traditional AI. just that little bit better, that little bit more human, that little bit more personalized, rather than it's got to be Gen AI or it's got to be traditional AI, and you've got to choose between the two. But there is an awful lot of misinformation out there, which is what happens when you have this sort of media hysteria. I think we've had over ChatGPT over the last six months or so. We do a lot of myth busting with people. before we can start to strategize in terms of what's right for them.
Rufus Grig
And I suppose there may be people out there who'd like the idea of bacon flavored ice cream. And thanks, Paul. And Sean, you're working with the technology. Keep us grounded in reality. How ready for primetime is this? Are people really doing this day in, day out?
Sean Lindsay
From my point of view and working in the CX industry, we're seeing it being rolled out and we're seeing it becoming available in the platform each week, coming out at an incredible pace. And there's some real great features that are out there today. To specifically answer your question, I think you've got to be very careful where we leverage generative AI and how contained that is. There is some scary stories on the market where people are introducing generative AI into their let's say, web messaging interfaces where customers are going onto the web chat and manipulating the answers that the generative AI is feeding back, which can hurt their brand. So yes, it's in the space, it's in the CX contact sensors today, and that is rolled out in a controlled manner. If you are introducing a generative AI platform, you want to have that in a controlled environment because it has the ability to really make up answers that may not be correct if it's not contained, surfacing old content. which can then again hurt your brand, maybe make promises to customers that actually are no longer possible. So again, you've got to be very careful of that. So my view of it is the contact center platforms are rolling this out and the features are incredible. So yes, I would say it's absolutely ready when rolled out in the right way.
Rufus Grig
So make sure it's rolled out in the right way. Make sure you work with a partner like Curve, who can help you and steer you and keep everything grounded and protected for your brand. Paul and Sean, thank you so much for your time today. It's been really fascinating listening to everything you've had to say about the use of generative AI in customer experience. If you've been interested in what we've said, then please do get in touch. Tell us what you think. You can find out all about Curve by visiting curve.com. Please do listen out for the next episode. You can subscribe. You can tell all your friends. Again, thank you very much to Paul and Sean for your contributions. Thank you for listening. And until next time, goodbye.

Chapters

Introduction to the speakers
37sec
Overview: Key capabilities of Generative AI
2min
The role of AI in CX: Pre-interaction with customers
5min
The role of AI in CX: Live-interaction for customers
9min
The role of AI in CX: Live-interaction for agents
12min
The role of AI in CX: Post-interaction with customers
16min
The role of AI in Regulatory Compliance: Post-interaction with customers
19min
How ready is AI for customers?
22min

Embed

Introduction to the speakers

37sec

Overview: Key capabilities of Generative AI

2min

The role of AI in CX: Pre-interaction with customers

5min

The role of AI in CX: Live-interaction for customers

9min

The role of AI in CX: Live-interaction for agents

12min

The role of AI in CX: Post-interaction with customers

16min

The role of AI in Regulatory Compliance: Post-interaction with customers

19min

How ready is AI for customers?

22min

Transcription

Paul Cox
Hello, and welcome to The Learning Curve, the podcast from Curve that delves into the latest developments in information technology and explores how organisations can put them to work for the good of their people, their customers, society and the planet. My name's Rufus Grigg, and in this series, with the help of some very special guests, we're looking at all things generative AI. So far, we've covered the background, the core technology and common techniques, and now we turn to real-world practical applications. Starting this week... with generative AI in customer experience. And I'm joined this week by two brilliant colleagues from Curve's very own customer experience practice called Curve Experience. Firstly, I've got Paul Cox. Paul is our principal CX consultant. Paul, tell us a bit about your role at Curve. So I spend my time talking to customers, trying to understand their contact center challenges, their business strategy, and then helping them to use our experience to guide them best to use that technology to reach their business goals.
Rufus Grig
Brilliant. Thanks, Paul. And without outing you too much on an age perspective, you've spent quite a bit of time in the CX and contact center world, I understand.
Paul Cox
I'm coming up to my 25 year anniversary of life in contact centers. So yes, I've been here for a while.
Rufus Grig
And yet you maintain your youthful vigor and appearance.
Paul Cox
So I still love it as much as I did when I started. Yeah,
Rufus Grig
there's a portrait in your attic somewhere that looks pretty shocking. but great to have you with us, Paul. And I'm also joined by Sean Lindsay, who's Principal Solution Architect also with Curve Experience. Sean, just give us an idea of your role and your day-to-day.
Sean Lindsay
Thanks, Rufus. So my role is Principal Solutions Consultant here at Curve Experience, and I predominantly work on the Genesys Cloud, so customer experience platform. And my day-to-day role is working with customers enhancing their CX journeys. So understanding their flows, their struggles, the pain points that they might be having, and how the contact center platform can essentially help them. And again, a key focus and hot topic of all areas is generative AI and AI in general. So therefore, again, using those platforms to enhance their customer service.
Rufus Grig
That's brilliant. Thank you very much, Sean. So as I hope you can hear, we've got absolutely the right people with us to steer us on this episode through how generative AI applies to customer experience. Before we actually delve into the applications, I'd really just like to explore what are the key capabilities of generative AI that are making most of the running. So Paul, first, can you pick something for me that's helpful?
Paul Cox
So generative AI is all based around large language models, and the key is in the language piece. So generative AI is unbelievably good at understanding what people are saying, and then generating language. That could be from one language to another. So translating from French to German, it could be just understanding what people say so that it can work out or the agent can work out what to do with that intent. Or it could be to generate some text, like a knowledge article or indeed a response to a question.
Rufus Grig
Brilliant. Thank you, Paul. Sean, anything from you?
Sean Lindsay
Yeah, so there's quite a few products that I'd like to mention. So in terms of generative AI, you've got... Key features like speech-to-text analytics that kind of leverages an AI model, specifically focusing on an intent miner, what we would call an intent miner, which gathers or analyzes conversations. So it'll be historical conversations. And with the use of AI itself, we are analyzing those conversations to look for topics and phrases, again, taking that automation to the next level, saving the business time and money, essentially, in that sense. And then one other element that we would like... to kind of highlight is going to be around summarization elements. So again, knowledge articles being surfaced to agents in real time, as well as the just general AI models, so AI bots for voice and digital.
Rufus Grig
Okay. So interestingly, although clearly we think of generative AI as creating text, what I'm actually hearing from both of you is it's the ability of gen AI or large language models in particular to understand what's being said that seems to be one of the more... important and useful pieces. So when we had a bit of a chat before this episode, we thought we'd structure our discussion into the sort of different phases of an interaction. So starting on what happens before an interaction, maybe even trying to prevent an interaction from needing to take place. We'd look at how Gen AI can apply to fully automated interactions where there's no human involved. Moving on to how AI can be used to assist agents with live interactions, whether they're talking or chatting by text with a customer. Then looking at what happens post-interaction, once the customer has hung up the phone or disconnected from the chat, what happens there. And then we're going to finally have a little bit of a look about the role of generative AI in assuring regulatory compliance and quality. We'll start with that pre-interaction world. And Paul, I guess given that... In many ways, most contact centre interactions these days, or a lot of them certainly, are really indicative of a failure of someone to be able to transact something online. It's often a sign that something has gone wrong or the self-service wasn't quite good enough. How does generative AI help keep people away from the contact centre altogether?
Paul Cox
I guess the first thing that's really important that gen AI can do for us or can improve for us is... understanding what it is that went wrong. Everyone who's heard me speak before knows that I'm a little bit obsessed with the importance of triage. So the first thing that Jenny and I can really help us with is when somebody says to us, here is my problem, then using that language capability to really understand what they mean and therefore what it is that needs to be done to help them. The second part is then to understand what the answer might be. And traditionally, in recent years, we use knowledge bots to understand somebody's intent, and then map that onto a potential answer. So someone says, you know, I'm having problems configuring my new iPhone. And therefore, we put up an article about that, and play that back to them. If we're using generative AI, then one of the things that this gives us on top of the traditional model is being able to have a more nuanced answer. So we might listen to what somebody's asking and realize that actually there are two or three different articles in our knowledge base that might be able to answer their question. Generative AI can read those articles and then generate a new response that's personalized just for them to answer that question. And all of that in answer to your question adds up to a much higher probability that we'll be able to solve their challenge and remove the need for them to speak to a live human being. So therefore reducing the cost to serve.
Rufus Grig
So would I be right in summarising that as a better way of understanding precisely the challenge that the customer has and then the ability to construct a more personalised answer to help solve their specific problem?
Paul Cox
Yeah, absolutely. Absolutely.
Rufus Grig
Okay, great. So I can absolutely see how that... does actually remove a fair number of potential needs to interact. So Sean, Paul mentioned triage. How in that sort of setup call, once we determine that a customer does need to speak or chat with a live agent, what aspects of generative AI are helpful in getting the best outcome in that scenario?
Sean Lindsay
It's a great question. So as Paul alluded to, when we're going through that journey of the IVR and triaging the user to get through to the team. There's a range of elements that we would utilize in terms of AI and generative AI. So one of the key elements that we might want to look at as a first glance is the reason for contact. And based on the reason for contact, we would then leverage a predictive routing, so it's an AI routing element, which then analyzes the conversation based on the reason for contact and based on the conversation and looks for the right team to handle that interaction. So instead of that customer going through the IVR, selecting it, journey, so selecting one or selecting an option within the IVR itself or message flow, interacting with a bot, the system will analyze that conversation, leverage AI, and distribute that interaction to the right team with the right skills. Now, we can take that a little bit further as if we understand the reason that we know the system says, okay, based on the predictive routing model, we know it needs to go to team A. We can then also look at conditional skills to take it a level deeper. To say, yes, we need to go to team A, but within that team, I'm looking for a very specific skill. So we can then tag that interaction with the specific skill sets and look for the right skilled agent. And that's all leveraging AI and all happens within milliseconds of the platform. So again, offering a great customer journey in that IVR stage, that understanding stage, and triaging that or transferring that to the right team with the right skills to handle that query.
Rufus Grig
And that's brilliant. Thanks, Sean. Let's have a bit of a think about... live chat before we get into talking about agents. I know when ChatGPT first sort of raised its head and entered public consciousness, there was a lot of talk about how it felt very like the sort of interaction we might have with a company on a sort of live chat scenario. Paul, perhaps you could explain sort of what's going on there with a live chat and is generative AI really up to the job there? Does it have a role to play? Is it going to improve the likely outcomes or experiences for chat?
Paul Cox
So I think we have to consider the different types of chat that we might have. When somebody makes contact with a bot, either it's because they have a question or because they have a request. In the scenario where they have a question, then Gen AI's ability to understand the triage piece we were talking about, and its ability to formulate those responses from multiple articles and personalize the response is going to give a much more human feel than a traditional bot would, where it is just giving you a canned response. It's looked at what you've said, it's trying to map it to an article. And then it's handed you back the article. So it is very much, if you've asked the standard core question, fantastic. If you have deviated very slightly from what the knowledge author had in mind when he wrote the article, then it's going to feel a bit weird. It's going to feel a bit unnatural. So whereas Gen AI is going to understand your knowledge articles, and it's a really important point that you can absolutely focus it on, say, only give me answers. from this knowledge base, from these articles. Do not go out to the internet, go out to your training and make stuff up. Only get information from here, but it will use that to construct a new answer.
Rufus Grig
That's what we talked about in an earlier episode as grounding and making sure the model is grounded in using only the approved and authored text for your brand and for your service line.
Paul Cox
Yeah, absolutely. Absolutely. For any AI geeks out there. This is a grounding method called RAG. So this is retrieval augmented generation, specifically that we're talking about here. So that's number one, where we're trying to answer a question and it definitely is going to make a more human experience, a better customer experience there for relatively low risk if we take this approach with a request where I say, I want to book a holiday, for example. And it's going to take me through a number of steps and ask me questions. There is an argument that it might be a slightly more human-like interaction. But actually, you know, really to do this sort of thing, you have a number of set pieces of information that you need to capture. And using Gen AI for that kind of experience, I would say the gains are relatively minimal compared to just going through a flow, a traditional flow. where the bot asks you one question after another.
Rufus Grig
Great. Thanks very much, Paul. Really interesting to see how chat, or certainly that automated chat, is evolving and how Gen AI is breathing sort of extra capability. But Sean, let's take us to the scenario where the automated chat can't do it. We are having a live interaction between a customer and an agent, whether that's speaking with speech or whether it's a chat where the agent is typing back. How is generative AI helping to assist that agent in the handling of that call or that chat session?
Sean Lindsay
Yeah, very good question. So within the platform itself, generative AI plays a huge role in assisting the agent either on a digital channel or a voice channel with the articles, so the knowledge-based articles that are present within the system. So surfacing that knowledge to the agent in real time based on the customer interaction, so what they're saying, the text. So the customer could be talking about a delivery, for example, or where is my delivery or something more specific. Now, based on the conversation and the wording being used, the phrases, we will then be able to surface articles to the agent in real time, leveraging generative AI. This has huge benefits because here we are again, providing that knowledge to that agent, again, taking that guesswork out of everything, lowering average handle times, and again, empowering the agents with the information that they need. They no longer need to go and look at back office systems for the relevant articles or relevant documentation to try and find the answer. The system and generative AI itself is now surfacing all of that content over to the live agent in real time.
Rufus Grig
And that works not just with chat, but can that work with a voice call as well? Is it listening and understanding the voice conversation too? Yes,
Sean Lindsay
absolutely. So that's both on digital interactions and voice interactions. So exactly what I've just explained across both channels.
Rufus Grig
Okay, no, that's brilliant. I can't remember if it was you or Paul earlier also mentioned the ability of a large language model to translate. How are we seeing that sort of language translation play out here?
Sean Lindsay
Yeah, this is quite a big topic. So let's start with the web messaging or digital channels first. So first of all, when we're considering an interaction coming in, let's say a customer was coming in from Germany, for example, they go through the IVR, so the bot journey or voice IVR, and then we transition. over to a live agent. Now, unfortunately, there's no live agents that speak German. So what we can do here is we can transition that over to the next best skilled agent. and that could be an English speaking agent. And what we can then leverage is translation tools on both voice and digital. So again, when we're talking about large language models, we're looking at the accuracy of the translation. So here in the, starting with web messaging, we can leverage a text-based translation, meaning customer will type a message. It'll automatically translate to the agent's native language. Agent can then see that text, understand that text in their language, and then ultimately respond. back in English, which will then automatically translate that into the customer's native language. In this example, it'll be German. Now, when we're leveraging the AI model, so the generative AI, we've also got to take into consideration the accuracy, as well as that consistency of receiving interactions and again, using that data for training purposes. That'll then increase the accuracy of those overall translations.
Rufus Grig
That's really interesting. It takes me back to the Hitchhiker's Guide to the Galaxy and the the Babel fish in the ear, the technology that seemed, when I read that as a child, so remote and now we're actually using it day in, day out. That's fascinating. Okay, so I can absolutely see how we're seeing massive benefits in terms of supporting the agent during that live interaction. What about post-interaction? The customer's hung up or the chat session's finished. Paul, what sort of technology is being put to use there?
Paul Cox
Sure. So this is actually one of the most exciting areas for me, because it's something that we're seeing being used very quickly. So it's one of the first parts of Gen AI to be adopted. The first one that we're seeing is summarization. So taking the transcript of the call or the chat, we've been for a while able to take that transcript and push it into a CRM system. But for the next agent, to have to read through what potentially could be a five or 10 minute calls worth of transcription and try to understand what was going on is really tough. I'm sure all of you out there as consumers have called a contact center, identified yourself, and then immediately been put on hold while the agent reads through your notes. And then there's nothing that makes a worse customer experience. So summarization allows us to condense that transcription into a short passage. paragraph that's easy for the next agent to understand, and then push that summary into the CRM to give not only a reduced handling time by shortening the wrap-up, but also to give a much better experience to the agent on the next call and a much better experience to the customer. The other thing that really helps is the outcome coding. So one of the things that we've struggled with for years is that If you give an agent five or six outcomes, they generally pretty consistently choose the right one out of those six at the end of every interaction and code interactions in a consistent way that gives you some good stats. But really and truly, you don't want just a very general one out of six. You really want one out of about 20 or 30 or perhaps even more. And where contact centers have tried to expand that list or make it a two-tier list, the accuracy and the consistency of agents selecting the right option tends to fall proportionately. With AI, the AI can look at that transcription, understand the key topics from the interaction, and then automatically code what that interaction was about. And it might be that it's about two or three things and it can identify those key topics that were discussed during the interaction. And that is a much more accurate way of really understanding what the interaction is about. That is a huge leap forward in terms of MI. So,
Rufus Grig
Sean, the final part of our interaction journey where everything's finished and we're into the quality and compliance management. Can you just talk a little bit about some of the applications in? that area of CX.
Sean Lindsay
Yeah, absolutely. So this is a very vital piece and this has huge benefits within the business. When we're talking about compliance, there's a range of elements that we can consider within the CX platforms. So once that interaction has been completed, we have the transcript and we have the reason for contact. So we know the category. Now, when we're talking and focusing on compliance or quality assessment, we can then look at quality assessment forms that need to be completed traditionally. by supervisors. Now with leveraging AI and generative AI, we can now look at automating that quality assessment scorecard. So what this does is it looks at the sentiment analysis, so based on the sentiment that you have configured, and also looking at the positive and negative parameters around the sentiment, as well as your topics. What that'll allow us to do is to auto score each of those questions in the quality assessment form based on the phrases used during. the conversation from both the agent side and the customer side. This is great because again, we're saving time. And again, if we do have an element where that needs to be reviewed by a supervisor that can absolutely be done so, AI is doing that work. You can scan through that, check it, make sure it is in fact correct. And you can optimize those journeys and those phrases using that intent miner that I referred to earlier. So the intent miner again, is analyzing those conversations, taking all the guesswork out of understanding or guessing what customers may say, and looking for those phrases that customers are using within the contact center. Now, we can also take that a step further where we're looking at detection of fraudulent patterns, as an example. So based on the phrasing used during a conversation, we can flag up key elements. So key phrases, maybe focusing on fraudulent elements there, and we can actually highlight those in the conversations and where those took part within that conversation. You can then add that into... processes within the CX platforms.
Rufus Grig
Really interesting. I love the phrase you use about taking the guesswork out. I suppose if you've got 100,000 interactions to look at and you're sort of sampling, the chances of finding what you're looking for are quite needle in a haystack. Whereas if we've understood really what's going on in each call and interaction, it's easier to hone in on the areas where you think you might have issues or certainly need to look at the quality. Really interesting.
Sean Lindsay
Absolutely. I mean, when we're talking about compliance and also knowledge itself, I mean... we want to analyze the conversations because guessing what customers might say is really hard and very time consuming.
Rufus Grig
Now, that's great. Well, I think we've been through a whole journey of an interaction from trying to remove the need for a live interaction to how do we support them? And then how do we ensure the quality and compliance after? Just, I suppose, to wrap up all, you know, you're with CX, customer experience leaders day in, day out. Just how much interest or fear or excitement is there about Gen AI? How real is this?
Paul Cox
I think there are two extremes that I encounter fairly regularly, actually. The one extreme is where we're trying to help people at the beginning of their AI journey with maybe something that isn't using any generative AI, but they've read a lot of stories in the media about hallucinations. And they are terrified of any kind of AI that might be customer facing because they're worried it's going to start making stuff up. And I think, Rufus, you told me a story the other day about a McDonald's self-service offering people bacon-flavored ice cream. And it's stories like that that really stick in people's minds. And they kind of feel like any kind of chatbot or voicebot is going to be dangerous for their brand potentially.
Rufus Grig
Interesting that that's still a very prevalent attitude. And I suppose we are still only sort of 18, 24 months on from this all becoming mainstream. But there's excitement too. You have the people at the positive end of the scale.
Paul Cox
Absolutely. And sometimes too much. So there are definitely people who I talk to who in their mind have decided that the sort of scripted intent-based bots that are most prevalent out there today are last year's technology, and they're not really interested in it anymore. And therefore, if they're going to implement any kind of bot from scratch, they only want to do generative AI, because that's the new version, and the old version would be a bad investment. So what we try to do is encourage people to think of Gen AI as a turbo booster on top of their traditional AI. It's something that makes traditional AI. just that little bit better, that little bit more human, that little bit more personalized, rather than it's got to be Gen AI or it's got to be traditional AI, and you've got to choose between the two. But there is an awful lot of misinformation out there, which is what happens when you have this sort of media hysteria. I think we've had over ChatGPT over the last six months or so. We do a lot of myth busting with people. before we can start to strategize in terms of what's right for them.
Rufus Grig
And I suppose there may be people out there who'd like the idea of bacon flavored ice cream. And thanks, Paul. And Sean, you're working with the technology. Keep us grounded in reality. How ready for primetime is this? Are people really doing this day in, day out?
Sean Lindsay
From my point of view and working in the CX industry, we're seeing it being rolled out and we're seeing it becoming available in the platform each week, coming out at an incredible pace. And there's some real great features that are out there today. To specifically answer your question, I think you've got to be very careful where we leverage generative AI and how contained that is. There is some scary stories on the market where people are introducing generative AI into their let's say, web messaging interfaces where customers are going onto the web chat and manipulating the answers that the generative AI is feeding back, which can hurt their brand. So yes, it's in the space, it's in the CX contact sensors today, and that is rolled out in a controlled manner. If you are introducing a generative AI platform, you want to have that in a controlled environment because it has the ability to really make up answers that may not be correct if it's not contained, surfacing old content. which can then again hurt your brand, maybe make promises to customers that actually are no longer possible. So again, you've got to be very careful of that. So my view of it is the contact center platforms are rolling this out and the features are incredible. So yes, I would say it's absolutely ready when rolled out in the right way.
Rufus Grig
So make sure it's rolled out in the right way. Make sure you work with a partner like Curve, who can help you and steer you and keep everything grounded and protected for your brand. Paul and Sean, thank you so much for your time today. It's been really fascinating listening to everything you've had to say about the use of generative AI in customer experience. If you've been interested in what we've said, then please do get in touch. Tell us what you think. You can find out all about Curve by visiting curve.com. Please do listen out for the next episode. You can subscribe. You can tell all your friends. Again, thank you very much to Paul and Sean for your contributions. Thank you for listening. And until next time, goodbye.

Chapters

Introduction to the speakers

37sec

Overview: Key capabilities of Generative AI

2min

The role of AI in CX: Pre-interaction with customers

5min

The role of AI in CX: Live-interaction for customers

9min

The role of AI in CX: Live-interaction for agents

12min

The role of AI in CX: Post-interaction with customers

16min

The role of AI in Regulatory Compliance: Post-interaction with customers

19min

How ready is AI for customers?

22min