How Decathlon is using conversational AI in retail with Charles Guth, Customer Service Manager, Decathlon

How Decathlon is using conversational AI in retail with Charles Guth, Customer Service Manager, Decathlon 1600 1200 Kane Simms

Learn how Decathlon flipped the script on its customer service channels and built an entirely new customer service set up with AI.

Supporting Ukraine

The VoiceLunch Foundation is taking donations to help support the voice lunch and voice technology community in Ukraine. VUX World has, of course, donated. I plead with you to donate too.

Donate here

Presented by Deepgram and Symbl.ai

Presented by DeepgramDeepgram is a Speech Company whose goal is to have every voice heard and understood. We have revolutionized speech-to-text (STT) with an End-to-End Deep Learning platform. This AI architectural advantage means you don’t have to compromise on speed, accuracy, scalability, or cost to build the next big idea in voice. Our easy-to-use SDKs and APIs allow developers to quickly test and embed our STT solution into their voice products. For more information, visit:

Presented by Symbl.aiSee how easy it is to add simple but powerful call coaching and call tracking functionality to your customer experience solutions with Symbl.ai’s customizable Conversation Intelligence APIs. From calls to videos to text conversations — apply best in class contextual AI in no time by getting started for free.


AI-enabled customer service and sales

Charles Guth is the Customer Services Manager at Decathlon UK. Learn how Decathlon is using conversational AI; it’s use cases, value and learnings, including:

  • Why Decathlon decided to close its call centre and rely on messaging and chat channel
  • How the script was flipped, going from 70% email contact and 30% messaging, to the complete opposite
  • What use cases and channels Decathlon is utilising AI for
  • Implementation lessons learned and tips for messaging automation
  • And much more…


00:00 Intro and welcome Deepgram and Symbl.ai
Sign up to ‘Is AI coming for your job?’ webinar
Sign up to ‘The Easy Way to Integrate Voicebots with Your Customer Service Solution’ webinar
04:30 Welcome Charles Guth, Customer Service Manager, Decathlon
07:00 Beginnings of conversational automation at Decathlon
08:25 Finding unserved demand with AI
12:00 Resource reallocation
14:00 Closing the call centre
17:30 AI channel automation
20:40 How to foster customer adoption
24:00 Google My Business and AI
30:00 Technology selection
31:58 Learnings from implementing AI
34:50 Importance of customer input
37:25 Conversation design approach
40:00 Measuring customer satisfaction
42:42 Agent escalation approach
44:40 Measurable business value
47:44 Advice for customer service managers
49:50 Future plans
50:57 Outro


Register for ‘Is AI coming for your job?’ webinar

Sign up to ‘The Easy Way to Integrate Voicebots with Your Customer Service Solution’ webinar

Word for word

Kane 0:00
Hello ladies and gentlemen, boys and girls. Welcome to vux world, the podcast where we dive deep into all areas of natural language processing and conversational AI. Find the best in the business, find out how they do what they do, so that you can do what you do better. Thank you for joining me again. And thank you to everyone that joined us on our webinar last night with Diagram and elerian ai discussing how to create truly conversational AI in the voice channel. It was pretty well attended to be honest, and everyone stuck around until the last minute. We even had to overrun by 20 minutes to get through all the questions that people were asking. So thank you everyone for attending. Thank you deepgram, for helping us put it together and for obviously sponsoring this episode of the VUX world.

If you don’t know by now. deepgram is an industry leading automatic speech recognition provider. And if you’ve ever joined us last night at the webinar, you’ll have learned a bunch of different ways that you can actually use a deepgram to the speech recognition, a deep compromise, to tailor your use cases to your specific company and the way that your people and customers speak tailor that ASR model to give yourself the best possible chance of getting the most accurate input into your NLU system. Because if you feed rubbish into there, you’re stacking up a whole bunch of problems further down the line. So do check out deepgram.com/vuxworld if you are interested in learning more about that, that is deepgram.com/vuxworld. Another shout out to our presenting sponsor symbol AI. symbol AI is industry leading conversational intelligence capabilities. If you’re a developer, if you’re technically minded, you can use symbol AI to build a whole manner of different use cases that will allow you to access data within conversations that you didn’t even know existed. You can do things like topic summarization, Speaker diarization, you can use it for quality assurance, you can even build your own agent assist capabilities. Using the symbol AI technology, the possibilities are absolutely endless. Check out symbol.ai for more information, that’s symbol.ai. To learn more. Thank you for the symbol and Deepgram.

Next week on Tuesday, I’m talking at an audio codes webinar about how to easily and simply integrate your bots into any call centre platform, the last mile is the hardest thing to overcome, you can have a really fancy assistant that working in a prototype or on a staging environment. But getting it into your call centre is incredibly difficult to do, most organisations will actually host it on a different phone number. And with that you miss out on a whole bunch of context you miss out on data that the organisation has on the call a whole handover between agents get really cranky. And there’s a whole bunch more that you can do if you integrate properly into your call centre. So I’m going to be covering some high level strategic tips that you need to be taken on board when you’re thinking about automating calls in your call centre. And Ilan Avner will be showing you how to actually go about doing any Bot Framework integrated into any call centre, really, really simply and you could do it yourself. So join us next week, it’s going to be on Tuesday. And we’ll put the link down here in the chat if you’re tuning in live. And also it will be in the show notes, obviously, when this podcast is published on the various channels.

And lastly, last thing I’ll say is that we’re also running another webinar, the second vux world webinar, which is gonna be on April 20. With Kore AI and we’re gonna be talking about agent assist use cases the title is, is AI coming for your job? And the reality as we find out today is that no, that’s not actually the case. AI is helping organisations and helping agents in call centres be more productive and more productive costs. But when they do that, how can you use AI technologies to empower agents on the phone, give them next best action, help them understand and find the information that they’re searching for. And that’s what we’re going to be talking about on April the 20th. So please do join me and Raj CEO of of Kore AI on April 20. And the link to that is on our event page vux.world/events.

Now without further ado, we are going to be climbing into the Mind of Charles Guth just now who is the customer services manager at decathlon and decathlon has been using conversational AI are a bunch of different channels with a bunch of different use cases when we’re talking to Charles about the value that these AI assistants are providing to Decathlon, the learnings that he’s had over the course of implementing this with Decathlon, the value that exists and some of the challenges that he faced and how he overcome them. So without further ado, Charles, welcome to VUX world,

Charles 4:31
Good afternoon. You Okay.

Kane 4:34
Very good. Thank you. Yeah, thanks for bearing with us. I know we start a bit over typical delivery issues on this side, the day then it’s quite a sunny day today. So I think that everyone’s kind of just taking a very laid back approach to things so the delivery drivers turned up four and a half hours.

Charles 4:53
is very nice.

Kane 4:55
It’s nice. It is nice. Yeah, we should be doing this kind of enroute shouldn’t be in a park or something like that. For next time maybe, yeah, at least you’ve got a nice sunny background behind you, I look like I’m kind of trapped in space. But thanks for joining us, Charles, appreciate it. For those people who are not aware of yourself or maybe even not aware of the catalogue, perhaps it would be helpful to give us a little bit of background, tell us a bit about yourself. But what you do at decathlon, and for those that may not be familiar what the customer on what the customer does,

Charles 5:22
yeah, of course. So well, to be very simply, the customer is pretty much the biggest sports retailer in the world. So our goal is pretty much to share the benefits of sports with as many people as possible. So we sell a whole range of products, a whole range of services as well, to try and share all the pleasure and all the benefits of sports with Mr. UK people and every country that we’re in, basically. So that’s a quick introduction about Catherine, for me, as we said, and customer service manager for the UK, the online customer service manager. So managing classic channels, emails, chats, phones as well, even though we have a little bit of a different setup, maybe we can come back to that. And, yeah, basically, we started accelerating on live chat and automation in AI, about three years ago. And since then, we’ve basically changed our whole way of doing customer service. So I think it’d be very interesting to deep dive a little bit on that today.

Kane 6:37
Huh, nice, nice. So three years ago was this before the pandemic was because of the pandemic,

Charles 6:45
it was a little bit before. But the pandemic definitely made us accelerate a lot on the idea of the project that we had, so we had, when we started three years ago, the idea was to change this in about five years. When the pandemic came, we had to do it in six months. So he was challenging, but now we’re in a much better place.

Kane 7:07
Nice. Nice. And where did you start? Then you mentioned I mean, every channel you’ve mentioned, they email chat online form. They’re all conversational channels, even websites these days, they’re becoming conversational channels with live chat, and all that kind of stuff. So where did you begin the journey? What channel and what use cases were we looking at originally?

Charles 7:32
Well, if we take three years, the reason I took it three years ago is because we decided to change a little bit our vision of what customer service was going to be, and to introduce much more automation specifically on advanced tasks. And that’s when we decided to change a few things. So we basically started by checking Okay, what do we want to work on? And the answer is pretty simple, really, because the numbers speak for themselves. I’m pretty sure if we have people watching that work in customer service, they already know the answer. Where’s my order? How can I return something? Is your store open? And when all these questions that pretty much have zero added value for the team and for the customer, they are taking 80% of all time. So that’s how I basically started to basically.

Kane 8:27
Yeah, it’s interesting, isn’t it? Because it’s those things as well. They have the typical other constraints, which is that, to speak to somebody about them, you can only do that when you’re open, which is, you know, for, for people that work in that space, it’s an obvious benefit of having more 24/7 kind of access, but I think it’s sometimes it gets overlooked. But but the reality is that, you know, introducing those kinds of like, ability to be able to get an answer in a conversation or channel at any time of day, presumably allows you to tap into demand that might not have been there previously,

Charles 9:05
before and tap into something that wasn’t there. And also remove something that was there before and basically deflected somewhere that you don’t need to worry about so much. But even before we started all these projects. We didn’t realize that that much. You don’t necessarily think that when we’re not working people are still out there, because you’re just not used to it. But as a matter of fact, they are. And we do realize that very much today we stay with the KPIs and the numbers that we see. So yeah, it was very important to offer kind of a bigger coverage for customers and coverage that makes more sense as well.

Kane 9:54
Interesting. You mentioned the future of customer service there. I wonder whether you can kind of share it and it was a little bit more on, you know, you mentioned kind of trying to remove some of those, like non value added tasks. And so I wonder if you can explain a little bit more about Decathlon ‘s vision for its customer service and customer experiences today. And now that you’ve been doing this for, for a little while?

Charles 10:14
Yeah, well, if we, if we take you back to the beginning, the idea was, was pretty much to make our processes more efficient, internally, but mainly for customers. So on the questions that I mentioned, where’s my order? Etc, etc. So not headphones. But the idea behind that is okay, once we do that, what is the point behind this? Because we don’t want to remove people. We have people with a lot of talents. That’s why we recruited them. So how can we use them better than just answering those non advisory questions? And that’s the vision that we have today is we’ve worked a lot on the after sale spot, thanks to automation. One of the reasons that we have a good setup on after sales Now is all the implementation that we’ve done in that part, now is how do we reinvest this. And this is a massive project here at Catherine UK, but also for the decathlon in the world . Obviously there’s a pandemic, the online share of all our sales has gone up. It’s the same for the retailers. But the support that we offer, hasn’t gone to ask quick. And that’s the problem. And today, we have great support when you’re going so if you need any advice on the technical products, and we have a lot of those, you’ll find someone to help you. If you go online, you’ll find a lot of details on one of the product pages, but you won’t necessarily find someone to help you as quickly and as efficiently as in store. And that’s the big problem today. And we need to close that gap and the leverage that the automation is given us to try and reallocate resources is going to go towards that.

Kane 12:07
Very interesting. So the AI assistants are taking up some of the demands of some of the more simple, simple, I suppose use cases. Where are you reallocating those resources? Is it more in the kind of like white glove services, live chat capacity for complex use cases and stuff like that? I wonder if you can shed a bit of light on where agents and resources are being reallocated now that you’ve got.

Charles 12:31
So basically what we’re building today, at the moment, we were until six months ago, let’s say we had classic customer service, we sold the customer service agent replying to order deliveries and things like that. Today, we are building a second customer service, if I can say it is going to be dedicated only to products. Okay, so only to answer technical questions about bikes, tents, or kayaks, to first help people make the right choice when they need to buy something. And seconds, help people repairing or building or being more technical wisdom products. Today, we don’t have that, except for bikes. That’s the first team that we built. Because obviously, during the pandemic, everyone wanted a bike. We had big needs. But the idea is this is where we’re going to reallocate those resources. How can we use people that we don’t need anymore on defining tasks, to help more customers to sell more products and to help customers when they have an issue with their product? So technical questions that a chatbot might not be able to answer, because it’s really too complex sometimes. And you do need a human at some point. But yeah, basically saying more products helps more customers.

Kane 13:57
Nice. Nice. So what are the channels that you kind of operate today from a? Well, most let’s start with it from a customer service point of view. You mentioned email, chat, website, phone, as of any of the channels that are not necessarily AI enabled, but any of the channels that we’ve missed out on that you utilize for customer service.

Charles 14:16
Well, apart from those three, we have custom social media, basically. So Facebook, Twitter, and we use while chat, WhatsApp, Google Business messaging as well. And what’s interesting is that basically, during the pandemic, the first thing that we did was close our phone lines. It was because to be honest, we didn’t really have a choice, it was too much too much to handle. We were not ready. I don’t think anyone was. But the very interesting thing is that at the same time we really pushed automation with our partner with different integration for API for order tracking, etc, etc. To make sure that we offered an alternative To our phone lines, that was pretty enough so that customers wouldn’t need it. That was well, about two years ago, march 2020. To this day, we still haven’t reopened our phone once, once our customer satisfaction has gone up 10 or 12 points overall. So that says a lot about the change of customers’ expectations, in terms of which channels do you need, but also the power of automation. When it comes to deflecting and changing the way you do things, basically.

Kane 15:37
Wow. So you see, you don’t have any phone lines open at all?

Charles 15:40
No, not at all. We do have the ones in store for store specific questions. For our online customer service. We do not have any phone line open. And we are handling about five to 6000 contacts every week.

Kane 15:56
Wow, wow, that is impressive. I’ve heard a couple of companies in the past closing their phone lines from an almost brute force point of view, go and use the website, you know, stop calling those. Whereas what’s interesting, and what ends up happening there is that you’re dealing or you’re left with what is likely an inadequate website, because there’s a reason why people are calling in the first place. So it’s interesting how you were kind of not closing the phone lines specifically to try and force people into other channels. But because you just genuinely couldn’t cope with the demand. But then also

Charles 16:35
does Yes, sides you were to deal with different after because we managed to basically offer a good alternative to

Kane 16:41
people. So I’m saying that’s what I was trying to get around to is that you you didn’t cut all close your phone lines and basically say go and deal with the rest of the channels that we have you close your phone lines, and then went and invested specifically in those other channels with an automation lens, which is quite a it’s quite a unique perspective. And, you know, in hindsight, you probably say quite a brave thing to do.

Charles 17:03
Yeah, well, that’s that’s what I was mentioning. At the beginning. That was always the idea. But obviously, in our mind that transition was going to take much longer interests, the COVID. And the pandemic just changed, changed everything. So we had to accelerate along that.

Kane 17:24
Interesting. So you mentioned WhatsApp, Google Business messaging, social media, email, live chat. Yeah. What’s one of those channels? Where is there currently today? A kind of AI assistant presence.

Charles 17:43
So we have Messenger on WhatsApp, and on Google Business messaging? Yeah.

Kane 17:52
They all handle the same use cases? Is it the same assistant in each channel? Is there some variation?

Charles 17:56
Yes. We have the same YouTube channel. Yeah.

Kane 18:00
Interesting. And where did you start? So let’s go. So WhatsApp, Facebook Messenger. And Google My Business? Yes. Correct. Interesting. Interesting. Where did you start them? Which one did you start with?

Charles 18:13
We started with messenger, because at the time when we started, we didn’t have WhatsApp, we only had messenger and Google’s business messaging. And the messenger was talking about 95% of the conversation. So that’s basically where we started. And the idea was to try and build a customer journey that is simple and quick and efficient at the same time, so that people can find the answers quickly, and, and try to deflect as much as possible during COVID, all the questions that we were getting all the time.

Kane 18:47
Interesting. We’ve got some questions coming through. I’ll get to these in a minute. Tatiana just wants it to end interestingly. So so. So you began with Facebook Messenger. And that was handling 90% of conversations

Charles 19:02
between just between messenger and Google Business messaging moments, okay. Overall, overall, when we started, we had emails and, and Messenger. He was in Google Business messaging, the live chat bots. So messenger, and Google Business messaging was talking about how safe 25 to 30%. And the email was taking all the rest.

Kane 19:25
Right. Okay, that makes sense. So, presumably then started with those kinds of straightforward use cases, they do it on those channels. Where’s my delivery? General q&a, that kind of stuff.

Charles 19:38
Yeah, so we’ve already had FAQs on our website. So we pretty much use those changes, of course to text to make it a little bit shorter and adapted to a kind of live chat frame. But that was the base. And then we did a few integrations as well so that customers can track the order automatically. For example, All, and a few questions about bikes availability and this kind of thing, which was the hot topic at the moment. The store locator as well, I think at the time, was probably the biggest one because the stores were opening and closing, some of them were partially opened. Some of them were not open at all. And it was changing all the time. So the fact that we were able to have that store integration on our life charts and updated almost in real time, it was massive help for our customers first and for our customer service. Second, because they had to deal with a lot less of the squares.

Kane 20:37
Interested in how you got there? So like part of the well, if you look at another Tatiana had a question about Amazon, Alexa and google assistant and stuff like that. And I have a lot of experience in working with those in the past. And part of the problem with those channels is that you need to try and persuade people to stop doing something on another channel and start taking up that behavior on a new channel, which is a smart speaker. With the call center, for example, you’ve got that adoption kind of baked in. People call a phone number, if they hit a bot, then they hit a bot, they’re likely to just talk to it. Curious about how you can like when you implemented Facebook Messenger and Google My Business, but did you do anything to kind of like, try and encourage the adoption of those channels? Or was it just because there was natural usage there that people just started using it?

Charles 21:24
Well, so if you go on our contact us page, and it hasn’t changed much since then, one thing that we’ll do to try and redirect the traffic to those channels is very simple. You have the list of two different contact options. So at the time it was messenger, Google Business messaging and email. Under messenger and business messaging, we just said we reply within one hour, and under email will reply within 24 hours, just to try and drive that traffic towards the channels where we knew we had the automation in that capacity to try and deflect some of the contacts that we were getting.

Kane 22:03
That’s good, because emails are historically just a nightmare to deal with. Anyway, honestly, you don’t really want to encourage that.

Charles 22:09
Yeah, and the thing about email is, even if we reply in the same amount of time it is still going to be a back and forth. Those are going to take them much longer than the back and forth on chat. If the conversation doesn’t go too human, so justice much, much more efficient, you can do several chats at the same time for emails to be as user friendly. So it’s an event for customers, I think. It’s just that you probably need new habits. Less and less people use emails, more and more people don’t use whatsapp or messenger or live chat or so I think there’s a bit of both, we managed to redirect some of the traffic and I think it was just natural as well, just by people have it’s

Kane 22:53
interestingly, out of those channels, WhatsApp Messenger and Google Business messaging, where would you say kinda like most of the usages, like on that Contact Us page, there’s a list of them where most people kind of gravitate into.

Charles 23:08
So we’ve got about, I would say, 10 percent going to Google Business messaging, and WhatsApp and messenger, or pretty much equal 45 percent, each more or less. The thing is, when you go on our contact page, the widget at the bottom right, where the live chat button is, is the messenger widget. And within the options, you have WhatsApp and etc, etc. So basically, customers have both. And what is important is that we added a guest option on Messenger widget, because not all customers are comfortable with putting their Facebook or Messenger account on home on any website. And yet, right now, they have about four to 5% of contacts each within the live chat scope.

Kane 23:56
Interesting, that’s really interesting. I think Google My Business is a really sort of untapped area. I’ve never actually you’re the first person I’ve spoken to who’s actually implemented something in that channel. And I’ve thought, I’ve thought for a long time that it’s got tremendous value. Because what do people do when they Google a business? The first thing that comes up is the Google My Business listing on the right hand side, you’ve got the opening times and all that kind of stuff. And I’ve always thought that Google Google messaging, Google Business messaging option is a fantastic idea.

Charles 24:28
It’s very good because it has a lot of benefits. The first one for us is that people in stores have a million things to do. So they are not as available as we can be in the customer service to answer your customers questions on the phone or even if they have their own WhatsApp or something. And customers most of the time have simple questions about the different services that we have in store for workshops or for restraining rackets or things like this. We have all this information. So if we’ve got any questions coming in about any store, we can just answer in a minute, we’re probably in the store, it might take a little bit longer, which is a shame. And all this, for customers and for people in store is its massive value. Because, again, for people in store, it takes away some less added value tasks, let’s say than what they already have to do. So they can focus on the customers that are present in the store or on making sure that the store is good for customers. And for us, it doesn’t necessarily change our job in any way. So just just a win win for everyone.

Kane 25:40
That’s wicked now. Yeah, that’s quite nice. Actually, how you’ve got a centralized team dipping into localized kind of use cases, it’s quite an interesting scenario

Charles 25:51
That takes a lot of work to make sure the centralized team, which is us, has the right information. We have a lot of contacts with the store to make sure that all the information we have about them, the different services that they have, or the click and collect availability or something like that is accurate. Because obviously we didn’t want to give wrong information to our customers.

Kane 26:17
Yeah, of course. Yeah. Yeah. So so how did you approach the sort of design of these things I’m assuming, you know, being the customer services manager, I’m assuming that it was fairly straightforward for you to identify, you know, the use cases that you wanted to tackle first, given given the strategy of trying to minimize the kind of non value added contacts and stuff like that. What was your approach in the actual design and development of this stuff?

Charles 26:43
So basically, at the very beginning, it was very much focused on after sales, because we would that was the main concern pretty much at the time. So we had FAQs, and then we had different options for all the return deliveries, etc, etc, with different integration on each floor. That lasted for about two and a half years. And about a year ago, we said, All right, now we’re good on after sales. So how can we change things a little bit, so that we can not only work on after sales, but also use that automation to drive business, drive traffic and drive business. So we basically changed the architecture of the chatbot to make it much more simple for the customers.

So if you go on our chat bot, now, you will have basically three points of entry. Either one, the first one is called my older, where you will either be able to track your order, or to organize the return of the order. Because those two questions are 90% of questions that are coming in about orders, you will have a second one that is going to be on products. And this one we are going to work very hard to develop as much as possible. So in this one, again, you will have two options: either help with my product that you already have, or buy a new product. So help with my product, the idea is to give as much information to customers so that they can be able to use it. For example, if you put in the bike section, you will have information on how to build your bike, which spare parts you can use for the specific item in your bike that you tried to draw, and you have the same for fitness machines for electronics, etc, etc.

In buy products. The idea is that a customer is going to come and say I’m looking for this. And we’re going to show him a carousel of 10 products with some product information. And depending on the conversion rate, for example, we can just shoot just a matrix of how those products are going to be displayed. And use an Add to Cart button. So that basically we can try and use that to start driving sales and driving revenues through the Chatbot. Because at the end of the day, in that flow, we currently have an Add to Cart button. But if you synchronize further and you have an integrated payment solution, you could basically have customers that are going to buy a product using only automation, which is going to be amazing because it gives you another point of entry within your business that doesn’t even require human intervention.

Kane 29:16
And something that can be extended and repurposed into an infinite number of channels basically providing the support the payments capability, you know,

Charles 29:25
yeah, exactly. That’s what works for everything. Any child, basically we have the automation, we can just build that build that build that flow, as we said as long as they have the capability of having integrated payment within

Kane 29:40
and you go nice, nice, wicked and so on. So using the header technology which was acquired by HootSuite not too long ago we’ve had Anna Ralph on the podcast from here to really good episode for those of you tuning in if you’re interested in how to break into conversation design and you know, learning from people who’ve done are taking different paths into it, then definitely check out the episode we did with Honorata was definitely worth checking out. And we talked a bit about her day there as well. Was there a kind of what was the logic behind arriving here? We were already a Hootsuite customer, like, where did that journey begin?

Charles 30:15
So it began three years ago, actually, that’s when we started working together, we were working with another supplier at the time. And our vision was to try and build more and more automation. In more and more efficient automation. It wasn’t at the time, the other partner we’re working with was less strong than a day on that point. So that’s why we had a meeting with hayday. They presented us with why . What was their vision at the time, on automation, after sales, sales, etc. And we kind of recognise our vision as well, on what they were trying to build in what they were trying to do. So that’s how we ended up working together. And now yeah, it’s been three years. And I think in those three years, we kind of went in the direction that we set originally, which is great. So now we just need to build the next three years together.

Kane 31:10
Nice. Nice. And so is it your team that kind of practically uses the tools? Is it kind of like a managed service? How does the actual physicality of making this stuff happen?

Charles 31:24
Yes. So we’ve got separated by channels. So we’ve got one team working specifically on emails once in working specifically on live chat on live channels, let’s say? Because within the tool, you have all the live channels that we mentioned to Google My Business, WhatsApp Messenger, all these are within Haiti. So my team is working on a day to day basis, on on all the conversations that are coming in from those channels on a

Kane 31:52
nice, nice, what was the what was the kind of from three years ago beginning that journey around automating these conversations? What would you say some of the kinds of learners that you might have that you might be able to share with people who might not have done it yet or haven’t maybe built something for there was a something’s chat channel specific. I haven’t done anything for Google My Business yet or WhatsApp or whatever. Like, I wonder if we get into some of the learnings that you and your team have found from designing and implementing these conversations?

Charles 32:20
Yeah, I think the three points that I mentioned before are very important to being quick, being relevant and being efficient. When we built the first time, our chatbots. We had, I think some of these points, maybe not all of them. But the first one, the quickness. Nobody wants to spend 20 minutes going round and round in circles on a chatbot. So we tried to make a flow in different flows that don’t take too long for the customers’ finances. And I think that’s very important.

And the two others being relevant and being efficient, they are going hand in hand. So the most important thing is to build the right path for your customers. But then, again, once you’ve done that, what you do next and you need to analyze weekly, to try and understand why are you Kirsten, why is your customers feedback on those paths that you created? Which ones are good, which ones could be improved and how they can be improved. And that’s the work that we are doing in Katherine and hand in hand with Haley to try and every week analyze while the conversations have the best rating, while the conversations have the worst one. And well, the customer saying so we have basically a thumbs up and down system, which is very simple and very classic, focus too much say if something has been helpful know, if I have one, one specific flow that has 60% of negative feedback, okay, while the customer is saying behind, so it’s just about analyzing the conversations to find out the keywords so the customers are mentioning, and then just feeding the NLU on what, what was the misunderstanding on that point? And how can we try to avoid it in the future?

So it’s, it’s very important and not very easy to build the right path. And but once you’ve done that, the word basically just starts because then behind you, you need to follow very, very closely what’s going on. Otherwise, you might just not be aware that things are going wrong for some of your customers.

Kane 34:44
Or maybe all of them as you know all of them. It’s very hard because that’s where a lot of people fall down or at least I think things have gotten better these days. But definitely over the over the years you’d have you know organizations that with the Probably something having not tested it with a customer group or what have you. So they’re totally blind leading up to actually going live with it. And they go live with something without any analytics capabilities. And so you’re almost kind of just let an employee loose on the website with absolutely no visibility or no monitoring whatsoever. Yes,

Charles 35:18
yeah. And it’s interesting, because actually, yeah, I think the analytics bar is, pretty much, the main key. It’s without you, just as you said, you’re blind, and you can’t, you can’t do anything. And one of the very important points that you mentioned is getting the feedback directly from your customer and testing with the customer. And that’s something that we did in September last year, we invited some customers in our offices to have a workshop specifically on the channel topic. And we learned a lot of things. And I think it was definitely following this specific workshop, that we had a clear idea of what our new chatbot path was going to be. And it was great, because they told us things that for them were super obvious, that we would have never thought about because when you have insights every day, all the time, you just miss things that are obvious for your customers. That’s why it’s super important, not only to get feedback through analytics, but also direct feedback. And it’s great to talk to your customers as well.

Kane 36:28
Yeah, definitely, I think as we, every designer who’s done any amount of user research, which I would hope all of them have. But it’s one of the very first things that you find out is that basically most of your decisions are completely assumptions. And that’s true with graphical user interfaces, but it’s compounded with a conversational user interface where you’re only ever designing one half of the conversation. And your conversational partner doesn’t exist. And so you need to get it in front of them. Because it’s the same as the principles is the same as testing with a user group or any other product as well as that. It doesn’t matter how empathetic the design team believes themselves to be, you’re not actually a customer in that buying context with the life that they have the situation they’re difficult to release, it was very difficult to do. Interesting.

So you obviously built up a good amount of capabilities and skills and experience in creating these conversation assistants. On a chat channel, it’s quite easy to, to kind of like use guardrails, you know, so you mentioned that at the beginning, if someone enters the top of the kind of the bar at the beginning, they can go down the is it a new product thing is, is a product, you know, a post sales kind of conversation? What’s your approach to that kind of conversation design element? Do you use a lot of like suggestion chips and buttons? Do you rely on more freeform dialogue, like wondering if you can talk us through some of the design decisions that were made in terms of how your book communicates.

Charles 37:56
So at the moment, we’ve got to be of both, we have some buttons, so suggestions basically, for customers, but all the way through the conversations that we always have the opportunity to just send the message instead of clicking on something. So throughout the conversation, the customer will always have the possibility either to click on something, or to send a message, that’s great, because that leaves basically a kind of free will to the customer. And we don’t want the customers to get stuck at any point. And that’s also why in the process of building it, we have a power transfer button, you will be everywhere, not everywhere, but in strategic places. So that we make sure that the customer is not stuck. And I think that’s very important. Because today, if you want different chatbots. In some of them, you do end up stuck sometimes. And it’s very annoying. And it’s made not all the time, but sometimes on purpose. So you’re going to find the top to agent button at the end. The very, very end of the flow is the same thing when you’re on the phone and you have to keep 1 million sales automated and only one of them out of 100 is going to take you to someone. And it’s very, very annoying.

So for us, we obviously look at the automation rates, but still being conscious that we made it on purpose so that we don’t have a 90% of automation rate, because it’s great to have 90%. But what’s behind it if you have 90% instead of 50%? But the 40% difference is just people that are gonna leave frustrated and send you an email instead. Then there’s no point.

Kane 39:40
Yeah, exactly. That’s the whole my gripe is with the concept of containment rate, which is a bit of a fallacy. As you said, if you contain 90% of conversations, but 50% of them are bad. Then it’s kind of defeating the object, you know, exactly. How do you approach getting into that level of detail then I’ve seen a lot of different Whereas of doing this from tagging to kind of like having specific intent on the back end of conversations to rating conversations, was it good or bad? Did this answer your question? Yes or no? Lots of different ways that you can kind of, you know, assume that somebody’s got to the end of a conversation, and therefore, it’s been successful. What are your thoughts on how to measure the quality of actual individual compensation? And I’m wondering as to how you’re measuring the success of it.

Charles 40:26
So different things. The first one is, customers have the opportunity to rate their conversations on every conversation they have with each other. And obviously, we have access to this data behind the idea now that it’s regrouped within categories, right? So for example, delivery questions, all the questions or returns questions, etc, etc. And within those categories, we have a little bit more granular analysis.

So in delivery, we have delivery policies, or delivery, etc, etc, etc. Right? And the idea is, all right, I’m gonna analyze what is the main category? That is the worst rate, it’s within that main category? What is the subcategory that has the worst rating that is taking the score down? And within that subcategories? Can I extract those conversations and go and watch myself? What has been happening with those customers? So that’s because maybe we haven’t understood one, one very simple example that I can give you is last month, we had an FAQ that was called, How can I cancel my order? Right? It’s a very simple FAQ, there was a real link to a video that explained how to cancel your order, etc, etc, very simple. By analyzing those conversations, we realized that literally 100% of customers going in through, we’re going there because they hold their order had already been canceled. So basically, the information that was there was completely irrelevant. And we saw that because of those ratings, subsystems down, we changed the FAQ completely, so that it’s more relevant to orders cancellation and how to cancel something.

So it’s a very simple, one simple No, it’s, it’s a straightforward job to do. But it’s very important to do it and to do it regularly to make sure that you keep up to date on the customer’s expectation, basically.

Kane 42:32
Nice, nice. You mentioned transfers a few times. Another way that you can monitor the success of a conversation is looking at where those transfers are being made, and whether it’s at the kind of appropriate point or not. I’m wondering what you’re kind of. It sounds as though you have a fairly liberal approach to agent escalation, which I think is the right approach to have if someone ultimately is not getting what they want, and they want to speak to someone, then the best thing to do is to let them do that. I’m wondering whether you can talk us through some of the decisions that you’ve made as far as how the whole agent transfer and escalation stuff works?

Charles 43:04
Yeah. So why it’s, it’s interesting, because we were coming back to the same topics a few times. Because basically, originally, the agent transfer was made only to our customer service. So do different options that we’re giving to customers to choose where they want it to be transferred, for basically customer service reasons, so same order, deliveries, returns, etc. With that new vision that we’ve had for the past year of developing sports reports online, we change this and now the customer, when transferring to an agent, is going to choose not a reason for customer service. So we can have replicated the real beat the organization that we have in store per sport category with one customer service wants, and the customer will either come to customer service, which is us, or Tuesday sports and will talk directly to an expert. So it’s kind of changing the mindset of the OB of saying, Alright, we’re not only a customer service anymore, we are a customer service on one side, and the sales or technical support. Pull on the other side, and the customer can have the choice between those two.

Kane 44:23
Nice, nice. So what has the kind of impact been so you had a situation where COVID happened? You get a flood of stuff happening online, as many retailers have, you have phones ringing off the hook to the point where you know, it’s impossible to handle the demand. You’ve invested a lot in one ultimate in a lot of these conversational channels and then to drive kind of adoption of these channels. You’ve got a lot of experience in building out the conversations within those channels and you know, reflecting the new state As with the organization, as far as kind of sales and support is concerned, how are you measuring the success of this high level? So rather than like, you know, the percentage of conversations that we’ve got are bad, like, what is the business value that this thing is delivering? And how are you measuring that?

Charles 45:17
So as I was mentioning earlier, when we launched, we had about 25 to 30% of conversations going through the live channels scope, and 70% or 75%, through emails. At the time, in the 25% of the conversation, live channels conversations, we had about 10 to 15% automated when we very, very started, right. So that’s pretty much nothing on the overall of conversations, where we are now is that we switched completely, we have 70 to 75%, going through live channels, so your messenger, WhatsApp, business messaging, etc. And the rest, so 25 to 30%, on emails, out of those 70 to 75%, we have about 40 percent, though fully automated. So if you think about it, we went from an overall 3%, maybe of all of our conversations automated to right now, on the overall customer service more than 30%. So that’s more than 10 times the rate of automation, which means that we’ve been able, as I was saying before, to reallocate resources from our old customer service, let’s say, to more interesting projects.

So one of them being the product, sales and support, but also other projects as well from people that have moved from customer service to our communications team, or our marketing team, or all the different teams that we’ve got in the office. So right now we have a little bit less people that we used to have in customer service. But the people that were there and are not there are still within the catalog environment. And they’re just doing more interesting things. And even for the people that we have any customer service. Their job is much more interesting,

Kane 47:14
though. Interesting. That’s really good. That was really good stuff. What would y’all kinda like? What would any of your pointers or bits of advice be to people who are looking to explore the same kind of thing we’ve spoken about, kind of advice around conversation design? And so I’m thinking more along the lines of, you know, you spoke well about use cases, starting with simple use cases, having urgent escalation in the right kind of places, you know, how to kind of get people to adopt certain channels and stuff like that. And we’ve covered a lot of best practices. But is there any other kind of advice that you would give someone in your position? Who has yet to do this?

Charles 47:54
Well, I think the first advice is probably don’t, don’t be shy. It’s a big step to take. Betting, let’s say on, on automation, for your company. It’s not easy, and, and the task can seem quite big, at first, and it is, but down the line, it’s 100% worth it. So I would say don’t be shy. And the second one, I think it’s very important as well, is to find the right partner for you to find a partner who is sharing your vision. We all have different companies and all different visions. And it’s exactly the same for AI or live chat for all those companies. And I think it’s very important to make sure that you’re aligned on what you’re trying to do and the timeframe you’re trying to do it in because otherwise, you might get issues down the line with your partner, which was viewed. So I think it’s something not to not to overlook, to properly look at you first and what you’re trying to do, and then look at the market and see what’s wrong with your strategies.

Kane 48:59
That’s very good advice because I’ve seen so many organizations who are switching out chatbot providers. In fact, you mentioned that you even went through the same thing. You know, that a lot of hasty decisions are made when there’s a rush to kind of climb on the AI bandwagon. And it’s like any vendor will do or the bosses look at them and all of a sudden you kind of get into a relationship where you know, you end up either choosing the wrong technology type, you know, if you are predominantly chat based and you all of a sudden want to extend into a voice channel, having the wrong partner can can potentially provide something to think about there or you might provide you might have a partner that doesn’t really isn’t able to handle use cases in your industry as well as another who might have specific domain knowledge I hear there, for example, was always about retail. NLU is retail focused, you know, so absolutely, yeah, that’s very good. That’s very good. So final steps, then what is the kind of you mentioned about the advancement of the use cases and stuff like that? What’s the plan and what would you like to see happen over the next kinda two Last month, is it more transactional different channels? Like what’s your what’s your thoughts?

Charles 50:05
Yeah, so we got to develop a difference. Add some more channels, for example, within mentioned Instagram today, because we obviously do have Instagram within the customer, UK, but it’s not yet within our scope. But at some point, it’s going to make sense to integrate it, especially when we’re going to have more and more sales going through their heyday. And that’s the second point really, I think with hayday. Now we are in the process, for example of switching the products database that we have so that we have products, database sites much more accurate, because one of the big strategies that we’ve got is to push the sales through automation, and through our teams that we’re building product experts. So for 2022, it’s going to be the D major project.

Kane 50:56
Nice, wicked. Well, best of luck. I find it on the front of her day, as I said before, big fan of Anna and the team and it sounds as though you’re having some real success. So congratulations on it all and look forward to seeing where it goes.

Charles 51:11
It’s meaningless. It’s been one year and we’ll see.

Kane 51:15
Let’s do it. Let’s do it. Yeah, that sounds good to me. Cool. Well, Charles, thank you. This has been absolutely immense. Thank you so much for joining us. Thank you for tuning in on the podcast. And if you’ve been watching it live pod you didn’t get to those questions. I think we might be circling back around them on YouTube. a little while. But yeah, just enough to remember. Don’t forget to join us on Tuesday for the webinar, audio codes. The links will be in the show notes and they’re in the chat. And also on the 20th of April for the webinar on how to empower your agents with AI assist capabilities with Kore AI on April 20. So that that is that if you haven’t yet subscribed, please do so via VUX.world/subscribe to get all invitations to order these immense discussions for people like Charles every single week, sometimes twice a week, in fact, mostly twice a week, helping you craft conversational agents and NLP solutions that deliver real value for your business. So thank you very much for joining us. And thank you again, Charles. It’s been a pleasure.

Charles 52:09
Thank you, too. Cheers.

    The world's most loved conversational AI event is back
    This is default text for notification bar
    Share via
    Copy link
    Powered by Social Snap