Paul Atkins: And we’re seeing already just the power of having more agentic AI in there.
Kane Simms: How did you go about creating what you think the ideal experience should be?
Paul Atkins: Moving into Jupiter now, it’s a generated voice, right? So that the conversation, weirdly, is more flexible.
Kane Simms: So how do you go about measuring quality and measuring performance?
Paul Atkins: I think we had a Trustpilot review come in yesterday where someone was saying, this is, you know, I was going to, I was about to hang up and send an email. Basically, the customer said, but actually, you know, having gone through the conversation, I got the information I needed and I was out of there within the next 30 seconds.
Kane Simms: What do you know now that you didn’t know before that you would absolutely either recommend people do or don’t do, or what’s the thing or things when it comes to working with either PolyAI or voice AI projects that you didn’t know going into it that people should know?
Paul Atkins: There’s this preconception that voice as a channel is going to die at some point. I don’t think it will. To be honest, I think that people will always want to solve a problem with immediacy.
Kane Simms: This episode is with Paul Atkins, who is the head of business optimisation at Simply Health. For those that don’t know Simply Health, Simply Health is a health insurance provider in the UK. It’s been going for over 150 years. Does around about 500 million, about half a billion in revenue or so. It’s growing pretty well. I want to say aggressively — that sounds a bit harsh to say aggressively, but it’s growing incredibly well year on year. It does things like Denplan, for those of you that might have that, you might be familiar with Denplan, and it’s recently finished as the best ranked company in KPMG’s customer experience reports. Essentially, KPMG think that Simply Health provide the best customer experience in the healthcare sector. So how good is that? Anyway, we chat to Paul about the AI ventures that Simply Health has been going through over the last few years. We talk about voice AI automation. We talk about email automation, we talk about chat automation, and we talk about the changing role of the contact centre and customer experience. The jobs that need to be done are changing. The operating model is changing. What staff are doing is changing, and the role of AI amongst all of that is also changing the way that businesses operate. But I think for the better, and the results that Simply Health have achieved are absolutely fantastic. 30% of all calls into Simply Health right now are being handled by AI, and we talk to Paul about how they’ve made that happen and how you can make that happen in your organisation. So without further ado, boys and girls, this is Paul Atkins from Simply Health on the VUX podcast. Paul, hello there.
Paul Atkins: Hello, Kane. How are you doing?
Kane Simms: I’m very well, thank you. Very well. You look like — is that a golfing tee shirt you’ve got on?
Paul Atkins: It is. It is. It’s for the breathability in this weather, right? It’s for the cool, cool factor. You gotta, gotta keep cool.
Kane Simms: I’ll tell you what I got the other day. You’re gonna love this. This is — wait there, one second. This is not planned, obviously. I didn’t know you were gonna wear a golf top. Oh, but I’ve found probably the best golf shirt imaginable.
Paul Atkins: Happy Gilmore.
Kane Simms: Happy Gilmore! Love it.
Paul Atkins: How good is that? Oh, I love that. Go home, ball! I love that.
Kane Simms: Yeah, it’s absolutely brilliant. It’s a classic. Yeah, it’s an Australian golf brand called Golf Gods, and they sell all sorts of cool stuff like that. So there you go, Happy Gilmore jersey.
Paul Atkins: I will be looking that up shortly.
Kane Simms: And another observation which I also like is you have behind you what is probably my favourite, definitely my favourite CX book. One of my favourite books actually is The Effortless Experience. It’s absolutely —
Paul Atkins: A bit of a bible from back in the day for anyone that works in and around CX, continuous improvement, even all that kind of stuff. So yeah, The Effortless Experience. I think as a result of that, in a couple of companies that I previously worked for, like Expedia, we looked at introducing customer effort score off the back of that. And the organisations tend to go from that, like CSAT to NPS to customer effort in more operational spaces, transactional spaces. Yeah, and that book plays a big part in it.
Kane Simms: All right. Great holiday reading as well.
Paul Atkins: It is. It’s really good.
Kane Simms: It’s really good, yeah. It’s fantastic. So tell us about yourself then, Paul. So you’ve got a lot of experience in AI. You’ve kind of got also a lot of experience, you know, deploying AI at Simply Health. So maybe, for those that maybe don’t know Simply Health, for those a bit further afield, maybe it’ll be interesting for context to explain a little bit about what Simply Health does and then your kind of role within it. And we can get into, you know, what the goodness looks like in terms of the AI stuff you’ve been doing.
Paul Atkins: Happy days, yeah. So Simply Health is a healthcare facilitator, almost, right. So we’re a healthcare cash plan business primarily. We’re about just over 150 years old, so we’re twice as old as the NHS, which is quite interesting. Started from lots of mill owners in the northwest, I think, that would take employee contribution to then provide employees access to doctors as quickly as possible for the benefit of the working population, really. And so we’re a purpose led company. All our profits go back into the business, so we’re member owned, a little bit like the co-op. The model is quite similar. So there’s kind of no investment firms or shareholders or anything like that. And the purpose is to get access to healthcare to as many people in working Britain as humanly possible. Right. We’ve got about 2.5 million customers in doing that. So it’s quite a great organisation to work for in that sense. There’s a mission behind it, and the mission is quite a socially good one, and a good one for your conscious and that kind of thing as well. And so in serving those members, we build various financial products for them, both dental for end customers, and healthcare, chiropractors, mental health support, loads of different solutions. And then we work with B2B as well. So we work with corporates and large employers that want to provide us as an extra employee benefit almost, where you can claim back for various things that you might need healthcare for. So we’ve been a traditional cash plan business. We’ve been on a bit of a product-led journey over the last sort of certainly three years since I’ve been at Simply Health, where we’re now looking to be more proactive in helping people manage their health through kind of our app, Simply Health app, and kind of be more preventative and look at preventative healthcare, as well as introduce a broad network of partners for people to get discounted healthcare from, and also claim that back via their cash plan. So we’re looking at introducing more and more kind of financial support products in that healthcare environment, and that is to support the NHS. I think it’s widely publicised, kind of, the difficulties that the NHS has been on in the last few years, and so yeah, our product aims to kind of fill that gap, get people what they need as quickly as possible and help them just manage their healthcare more proactively.
Kane Simms: Nice.
Paul Atkins: So that’s a bit about the company. I am head of business optimisation there. So my remit is to support our group in its business activity, introducing tech-based tools to do that, whether that be like process optimisation and automation, whether that be AI-based tools for customer interactions or within a sales process. But for the first three years or so, I was head of digital services, which was based more in customer operations and customer services, and that’s where a lot of our AI-based deployments and triumphs came from, from that space. So off the back of that, I’ve moved into more of a group-wide function now, trying to replicate the same sort of success across various other elements of our business as well.
Kane Simms: Nice, wicked. So where did the AI journey start then at Simply Health?
Paul Atkins: So I think we started with an organisation called PolyAI, and that was because of our customer base, right? Traditionally, our cash plan customers are of the older demographic, from kind of the older cash plan businesses that see it as an insurance product that they pay into and hopefully never really have to claim a huge amount back from, and thus that demographic love to use the telephone to talk about their healthcare. Often to just have a nice chat on a Sunday after they’ve done the shopping or whatever. And so therefore we felt, in terms of trying to be efficient and trying to be able to help our business, you know, support the growth of our business and the scalability, we needed to deploy some functionality in the voice channel that was customer-centric, that didn’t introduce a hell of a lot of effort into that kind of voice experience, but also helped us gain some efficiency around some of the more transactional conversations and, you know, the FAQ type stuff. Ultimately, that was where we started. Since then, that’s evolved and has grown arms and legs. We can get into that as well. We then have Salesforce as our CRM partner, so within Service Cloud, we deployed things like omnichannel service replies. We’ve done some POCs with Agentforce and seen some success in there as well. So like work summaries, case summaries, yeah. The service reply stuff’s been brilliant as well. So we’ve done a little bit in that space. Looking at other channels, we’ve deployed a few of the bots in our chat channels as well. So live chat. So that’s really been kind of — it’s been quite service-centric initially, but now we’re starting to broaden out and look at other use cases and to build on the success that we did in the service space. Because it’s usually the easiest attainable ROI, the easiest place to do a POC, because it’s, you know, it’s efficiency, it’s widgets going down a pipe. Can you do more? How quickly can you do them? Can you ultimately free your paid employees who are brilliant, right? Our customer service scores are absolutely excellent, as are our Trustpilot results. Can we free those people up to do more complex tasks that require the emotional intelligence, the complex problem solving? So can we create capacity to spend more time delivering exceptional service in the cases that need it, by dealing with the most — the more transactional stuff — in more of an automated and AI-supported way. So those have been our kind of main starting blocks, I’d say.
Kane Simms: Nice, wicked. That sounds great. Yeah. PolyAI, I’m sure people listening are familiar with PolyAI. Shout out to PolyAI. We had Nikola on recently. Great, great, great company, great technology. Yeah, since the beginning, I mean, they’ve always been very unique in terms of their approach, you know, very much specialised in voice and stuff like that. So they definitely know their stuff. So that’s interesting. So I suppose it’s a fairly common place to start then in terms of, you know, get rid of the stuff that’s not adding value to either the user or the agent and have them focus on where the actual problems are and bring some of that kind of value into the IVR. So you said you started with FAQ based stuff, the straightforward stuff, and then you said it’s evolved. So what was the journey and where is it now?
Paul Atkins: Yeah. So yeah, it’s evolved a fair bit. So yeah, we started with the FAQ stuff, testing, building out the models, all that kind of really solid stuff. And we did it in a really stepped way. So we created actually around the same time an AI forum within the organisation that was tech supported, you know, got legal in there because we’re a regulated organisation as well, right? So there’s, you know, there’s quite a lot of hunger to do this type of stuff, but also a bit of trepidation about what could happen if things go wrong. And we’ve seen that happen in other organisations, right? So we started really, really small. We started with just — it was almost auto triage, like IVR type routing stuff. So what is it that you’re after? Do I get you to the right place? And then we started to layer that up slowly but surely, as well as doing test and learn on smaller volumes that we would funnel into a new journey, right? So we started to layer up with stuff like sending claim forms out. And then we started to look more on adding in more FAQs for customers. So we started with the big hitters and moved down from there, and did the old 80/20 around rolling that stuff out. And that saw great success, great resolution or containment, whatever your flavour is of what you want to call that. And, yeah, really successful deployment. And then as we kind of got more into the relationship, PolyAI started to talk to us about Jupiter and their new platform, right? Which is this, I guess, quite low-code, no-code front end that’s based around prompt engineering. In my team, I’ve got a few people that can do that sort of thing and are very good at it. And that then opens the doors to do, you know, dynamic real-time journey management, right? And then all of a sudden you’re thinking, okay, can I get in, you know, can I link into some of our financial systems and more deeply into Salesforce so that customers can talk more specifically about their product and I can serve them back information. We’ve actually just had a brilliant, relatively non-invasive API journey go live around entitlement. So essentially we have a product table. We retrieve the customer’s account number from Salesforce and their product reference, look it up against the product table, which covers about 70% of our population, and can tell them specifically that they are or aren’t covered for a certain thing. So without getting too granular into customer information, we will get there. And we are doing that. There are layers that we can peel back that feel a little bit safer, and those journeys are brilliant. I’ve heard some of the calls just the other day, absolutely fantastic, and we’re up to around 30-odd percent resolution now, and we’ve got more API enabled stuff to come in the voice channel.
Kane Simms: Is that 30% of all calls coming in, or 30% of the use cases that you’ve deployed?
Paul Atkins: So it’s 30% of all calls coming in.
Kane Simms: Wow.
Paul Atkins: Yeah, that’s brilliant. Yes. And we were in and around 25% last year with the old version of PolyAI. Yeah, or earlier this year — last year into earlier this year. And we’re seeing already just the power of having more agentic AI in there is making for more organic conversations. That barge functionality, which PolyAI like to talk about, which is brilliant. So we’ve gone from voice actors that we would script lines into, right? And there was the gentleman. So Denplan is our other brand. That’s more for dentists, that sits under Simply Health Group, and we had a lovely chap called Tom who had a fantastic voice, who all of the Denplan customers and our Denplan staff absolutely loved his voice. But with moving into Jupiter now, it’s a generated voice, right? So that the conversation, weirdly, is more flexible than what we did have in the early days of PolyAI, which was pre-recorded lines and a bank of verbiage that we could use. And when we recognise utterances and stuff like that, which again was great and, you know, got us a lot of the way there, but this is just way more dynamic now. The conversation feels more organic. It can flow. Customer can interrupt. The voice will generate the response to match the tone of the customer and the language you use. It’s just way more intuitive and easier to change, right? Instead of having to get back in the studio, record more voice lines, raise the change ticket, get it in, test it, all that kind of stuff. Now we can just in our front end change, you know, change a customer journey, change your response, change some of the knowledge grounding. And it’s done, right? Which therefore opens the door to our product teams who want to do some testing, right, on tweaking journeys and building out product improvements. Well, now all of a sudden we can, you know, filter a percentage of customers down into the test route, and they can have a different experience that product wants to measure. So for like product agility and test and learn — that real test and learn culture — all of a sudden opens those doors, right, to be a great channel to learn things about our customers from and do that in a relatively safe environment that we can change, switch off almost immediately, and have more oversight and control over.
Kane Simms: Absolutely. One of the great stories from PolyAI back in the day was one of their customers is called Landry’s. It’s like a casino and restaurant chain in America. And the guy who — I’ve had him on the podcast before, I’m pretty sure of it, Brian Jefferson, who kind of runs the contact centre there. Shout out to Brian. He was telling us that — this was ages ago, this was back in the day — that the single thing that made them want to go with PolyAI over any other provider was that when they had them basically build them a demo, because they were based in Texas, PolyAI chose a Texas accent for the demo, and he said that was brilliant because that’s exactly how our customers speak. And it’s the best experience. And I thought that was really interesting. I mean, I’ve always thought that the voice is important, but never really considered how important until, you know, the next generation now of synthetic voices, because there’s so many of them available, they’re so good quality, and you’ve got so much flexibility now. Whereas before it was like you had 12 Google voices, 12 Amazon voices, and you got to choose between one of them, and most of them are American. So you’ve got like three English voices. Yeah. And it’s like you’re really stuck. Whereas now there’s so much choice that you can really craft the right experience that matches your use case, your users, their mental model, their attitudes, all that kind of stuff. I wonder whether — I mean, did — was there much thought that went into that? Was that you? Was that PolyAI? How did you go about creating what you think the ideal experience should be?
Paul Atkins: It was us as an organisation, right? So you’re absolutely right. There’s a ton of choice there. And so we got brand involved as well because it’s a bit like, what’s the voice of your brand? When I was at Virgin Media, we did quite a bit of work on that. Like, what’s the voice that represents us the best? So we kind of put it to our organisation community, right? And tested a few and took votes in and all that kind of stuff, because again, accent’s really important. I’ve worked in customer contact for a couple of decades now, right? And worked up in the northeast, here in the northwest, and in London and the southwest as well. And I think kind of accent and how that’s perceived as an identity for your organisation, it’s quite important, and it’s kind of what do you want to put across as your persona or the image that most people, or a lot of people, are going to conjure up when they’re listening to you. So we did lots of tests. We put it to votes, we got our brand teams involved in it.
Kane Simms: Did you get Tom involved? Did you clone Tom’s voice?
Paul Atkins: We didn’t clone Tom’s voice in the end. No, there was quite a vocal Tom fan club that came out in force saying, bring back Tom. Wherever you are, Tom, there’s a lot of love at Simply Health for you and Denplan in particular, and you’ll always be welcome should you ever want to visit HQ. But yeah, so we settled on something relatively neutral and clear. I think actual clarity was quite a big one as well. I mean, you’ll have read all of the studies about accents and what works, what doesn’t work, and all that kind of stuff. And it’s, you know, it’s quite interesting to read, but I think actually just being able to get done what you need to get done in the clearest way possible probably has still got to be your biggest factor. Right? But our brand principles were to be approachable. So you’ll see, if you ever go on my LinkedIn page, you’ll see a banner at the top that is our motto, which is simplifying healthcare, which is like a load of tangled wires into one single wire. So clarity almost is our brand identity. So therefore, having a bit of clarity, being able to make sense of things, being approachable and warm, they were kind of the guiding principles behind our choice. And we went with something relatively neutral. That was — as you’d expect, I was an advocate for the northern accent, but I’m the minority in my organisation, unfortunately. So, yeah, even though we’re a legacy Northwestern company as well, right back to the roots 150 years ago. So yeah, that was a really interesting exercise, and it’s just really interesting how polarised people are about this type of thing as well, and how subjective it can be in certain instances. But, yeah, I think if we got Tom back in, though, I think that would work as well.
Kane Simms: Nice. And how — you know, how do you go about, you know, measuring success? Because you’ve almost got two criteria, haven’t you? You’ve got the one that you began with, which is can we free up time? Can we free up people, can we focus on, you know, people with more complex needs and stuff like that. And the other is that those that are having the experience with these AI solutions, making sure that they can get what they need and they can have a nice experience, an effortless experience. So how do you go about measuring quality and measuring performance?
Paul Atkins: So yeah, so we’ve got — we’ve had obviously legacy QA functions and have legacy QA functions and that call listening quality. So test, test, test, right, is the biggie. And then don’t be afraid to roll back if we need to, you know, if we need to tweak something and it’s not performing as we would expect it to. But in that journey as well, we’ve learned stuff. So with the ramp up onto Jupiter, for example, we started with 10% of calls. We went to 15, 20 and so on and so forth until we reached 100. And we uncovered things that were, you know, that didn’t work. We uncovered edge cases, we uncovered bugs, and we tested the living daylights out of it. My team in particular listened to over a thousand calls in a relatively short space of time, because they were passionate about it. Because it’s our baby to an extent, right? And we know that we’re one of the front runners on the new platform as well. So we wanted to do a really thorough job for our own benefit, but as well to add the value into the partnership of getting PolyAI the information that they want about performance and about usability and about reliability. So a ton of core listening. It sounds really old school to be saying it, given the subject of this podcast, right? But I think early days, you kind of need to put that effort in up top. Now obviously we’ve got automated QA elements in there. There’ll always be a bit of human in the loop for that for us, just to make sure that we’re keeping, you know, keeping as safe as possible. And obviously with that kind of AI governance framework in mind as well. But I think that’s the real key. Resolution’s great. And you can have targets for resolution. You know, our demand forecasters — shout out Mark — we’ll always have a, you know, demand forecast, and it needs to resolve this amount because, you know, of all of that kind of stuff, that resource profiling across the year. But actually, we’re only discovering what the art of the possible is. So we’re almost bracketing where we want to get to by the end of the year, and already PolyAI is starting to creep up and overperform, you know, in some areas that we’re — you know, better than we were anticipating. So there’s that performance bit. But you’re right, the quality is the main thing, and certainly in complex organisations. Ours is like a hybrid of healthcare provision and financial services. So it’s like at the intersection of two pretty heavily regulated industries, and so actually making sure that the quality delivered is of a standard and maintains our reputation as well. We’ve just been published in one of the consultancy KPMG papers as the best health insurer, as a leader in that. I think we were 35th or 31st in the world. And we were a heavy leader based against, you know, your axes and all that kind of stuff. They were followers. There was clear water between us. So we’ve got quite an excellent reputation to uphold. And so the quality is the main thing, right? And hence why a lot of the effort went into listening to those calls, finding the edge cases, finding the bits that didn’t quite work, and then tweaking and retesting. As you evolve with these solutions, that bit gets quicker, right? As the trust builds, as you get more proficient in making the tweaks, squashing the bugs, all that kind of stuff, that gets quicker. But I think early on you’ve got to be ready to put some old-fashioned graft in on making sure that the solution is working to a quality that you desire. I don’t think there’s any way out of that.
Kane Simms: Yeah, I think you can automate things to give you very broad kind of trends, if you like. But certainly in those first early days of going live, the team, you want to be all over it anyway because you want to see for yourself, you know, live calls, real people. Let’s have a listen. Let’s see how it goes. And partly you’re kind of there because you’re trying to find out the little, you know, the long tail stuff that you didn’t work out and the things that happened that you didn’t anticipate, and you want to try and fix things. But also when you hear a call that just goes right, it’s just brilliant. It’s —
Paul Atkins: Pretty satisfying. Yeah, when you actually listen to it yourself —
Kane Simms: It’s different to just looking at it on a dashboard that says, you know, 90% of calls were handled or resolved or whatever it might be. To actually experience it yourself is brilliant. And also when you read the transcript or you listen to the calls and you find issues, it allows you to really empathise with the user, especially with voice, because you can hear them. Oh, for f***’s sake! Yeah, you can hear it, and you’re like, oh, we don’t really want to do that to people, you know? So there’s a little bit more kind of empathy involved in how you fix the problem, you know?
Paul Atkins: And I’ll tell you what, when it does go well, right, and we’ve had quite recently some presentations to the board and to our ExCo, where we’ve brought the customer into the room and played those interactions, where we’ve tried to design something so like this entitlement journey and it’s gone well, you know. A lot of testing and hard work behind that, but it’s gone well. And then you’ve listened to a call, and it’s like, that’s what I pictured when we designed this journey. That’s the interaction that we pictured. It feels great. And it’s really nice to then show that to the rest of your organisation and play that. And, you know, but there is a lot of graft that goes into making that happen, right?
Kane Simms: Absolutely. Well, last thing on the voice. Voice is a very hot topic at the minute, and lots and lots of companies are interested in doing it. What do you know now that you didn’t know before that you would absolutely either recommend people do or don’t do, or what’s the thing or things when it comes to working with either PolyAI or voice AI projects that you didn’t know going into it that people should know?
Paul Atkins: I think people’s receptiveness to getting resolution, whether it be AI or not AI. I think a lot of us had preconceptions that there was going to be a hefty population of our customers that would just hang up, and there are some, or a hefty population of our customers that would say, speak to someone, speak to someone, speak to someone, which, you know, a lot of people do, right? They overpower the agentic capability and go get routed straight through. I think, weirdly, what we found was that there was a large receptiveness to it. We had, I think, a Trustpilot review come in yesterday where someone was saying, this is, you know, I was going to — I was about to hang up and send an email. Basically, the customer said, but actually, you know, having gone through the conversation, I got the information I needed, and I was out of there within the next 30 seconds. And then on reflection, I thought, well, that was actually a really good experience. So there is, I think, there is an opportunity. What we didn’t know was there was an opportunity in that moment, in that interaction, to almost convert customers into interacting with an AI voice capability. I didn’t ever see that happening. I thought there would be people that would tolerate and get, you know, get through the journey to get the information they needed or to do the thing that they needed to do, or they would just straight override and shoot through to an advisor in person. Those things do happen. But there was another population that you could, you know, you could tell were a bit sort of wary. But then as they had the experience, they’d got the information they needed, and they were on their way, getting on with the rest of the day, and have then been converted to that. And a lot of our repeat callers that we had are happy to repeat call and use PolyAI, which is something that I never — I didn’t see happening, right? I thought there was just going to be the polarised camps that were into it. And those people would continue to use voice because, you know, they would never go into a digital channel. And there was forever going to be that friction there. So that, and the results we’ve had in terms of resolution being higher than we thought it would be, or certainly thought it would be at this point. So I think you’ve got to have an openness to give it a go. I think as long as your parameters are in place of how you experiment, how you ramp up volume, how you find those edge cases we talked about, squash those bugs and stuff. I would say to anyone that’s kind of reluctant, give it a try, right? I think there’s a big preconception out there that voice AI is not really anything that people want to necessarily spend any money on or develop within their organisation, because there’s this preconception that voice as a channel is going to die at some point. I don’t think it will, to be honest. I think that people will always want to solve a problem with immediacy and know that there’s something happening. And that call to action of picking up a phone and kind of directly orchestrating that thing happening as a human, I think that drive will always be there, especially when it’s more complex things. So if you can solve more of those complex things and demonstrate that efficiency in that moment, then you have a real chance of getting real value out of the tool itself as well, right? So yeah, I think there’s a few things there.
Kane Simms: Nice, nice. And so you mentioned also you’ve kind of been experimenting with Agentforce and chat on the chat side. Is that the same use cases as on the voice channel, is it different use cases? How does that look?
Paul Atkins: Yeah. So it’s roughly the same use cases. So we’ve done some POCs with Agentforce. Weirdly, now that we’re kind of looking at PolyAI and their Jupiter platform, we’re looking to see if there’s any chat capability going to be happening there and can we get into that. I think there’s the knowledge bit in the background of not having too much stuff in disparate places, not being able to curate it. We can get into that. But yeah, we’ve done some really interesting stuff with Salesforce, with Agentforce, with their Einstein layer that was kind of the precursor to Agentforce. So we’ve introduced stuff like conversation summaries or work summaries, which, you know, summarise an interaction and do all of that bit for an agent on the telephone so that they don’t have to write case notes and all that kind of stuff. Tons of efficiency in that, right? I think getting the user adoption around that and kind of the trust within your own employees around that’s quite a biggie to make that fly. As you mentioned, yeah, chat Agentforce, same use cases, FAQs. I think there it becomes more about where do you want to exist to access that channel, right? Because the customer can pick up the phone, they’re in the channel. With chat, it’s where do you want to go? We’re quite lucky because we’ve got FAQ pages, which I call a cul-de-sac. So it doesn’t really have a ripple effect on any of the other journeys. Once you go there, you’re going there to get something. And so we popped our chat buttons up on that page because we knew customers were going down that journey. It was a relatively small percentage of kind of web queries, like 17% or something like that. So we knew roughly the volumes of people visiting that page. We could put our web chat button there and do a little bit of test and learn, right, on kind of the things customers wanted to talk to us about, how we curate the knowledge base that sits behind that. So we moved from quite an old school decision tree-based chatbot, which is interesting and does a thing, but can be quite frustrating, right? Because it’s almost just a representation of the FAQs that’s on the page that they are on anyway, right — into a more dynamic conversational bot. And we saw natural resolution uplifts straight away off the bat of being able to surface information in a more conversational way, in a more dynamic way. So that was a really good POC. So we’re now kind of at the crossroads of how we develop that POC and look at our suite going forward. Yeah, really interesting one on chat. Quite a different channel as well, right, to voice, in that it does tend to be a little bit more transactional.
Kane Simms: Yeah.
Paul Atkins: We’ve also got email service replies. So understanding an inbound contact from a customer, being able to draft a response, human in the loop presses go on it, reducing our email processing time from about 12 minutes to two —
Kane Simms: Wow.
Paul Atkins: Straight off the bat, which is great. That’s brought us a lot of goodness and upside, and yeah, that’s a brilliant bit of functionality, right? I think almost if you’re quite email reliant in terms of your organisation — we have quite a lot of emails coming in because we have contact us, all that kind of stuff. We work with corporates as well, and brokers who tend to send emails and fire off. I think what will be interesting into next year is as we further examine WhatsApp, which for me always feels like a bit of a — it’s an asynchronous channel, but it almost feels like a halfway house between chat and email in a sense. It will be interesting to see, on the B2B side of things. I think it’s a great place to start, right? Because you’ve got scalability. Feels like it’s white gloves, but it’s easier to resource almost because it’s asynchronous in a sense, and then the organisation feels like they’ve got an open line to you at all times, even though it’s resourced slightly differently. Smoke and mirrors behind the scenes. So I think that will be an interesting evolution of chat, and then we’re looking at in-app messaging as well. So I think that will arrive at some point this year. And we want to look at what functionality we want to ship in the box with that in terms of agentic response. So there’s tons going on in the Salesforce space as well.
Kane Simms: Nice. Interesting. I mean, I’ve heard a kind of mixed response about Agentforce, but it sounds from what you’re saying there actually that it’s actually pretty good. And the email use cases are great because people can just be buried in emails all day, you know. So taking that time down is absolutely amazing.
Paul Atkins: Yeah, yeah. We have people using templates and stuff, or pre-stored templates and then tweaking them and stuff. And the queries can be quite complex in a sense, but we had to take, I guess, organisationally, we had to take a stance of what’s good enough as a response, because, I mean, our customer service advisors are brilliant, right? Best in class. So they almost have a temptation to over service. I don’t really like that terminology, but they’ll have a temptation to give a really, really, really bespoke answer to a customer. Whereas I think if you’re in a growing business like we are, you’ve got to think, okay, well what’s excellent and what’s good and what’s good enough between that excellent and good in terms of a response to this query type. And so I’ve got a business readiness team that works in my group that worked really hard on that, like speed to happiness for our people in using the tool. And actually the adoption and then adherence to using the tool, right? So how can we optimise it for them? How do we present it better on that console, for example? How can we kind of move, behaviourally, our people from using these templates and then over-engineering from the template upwards and spending 12 minutes eyeballing a service reply, and saying, that’s good enough. That gives the customer what they need. Off you go. So that took a little while. Actually, it took a little while to ramp up to that because it was a real behavioural change.
Kane Simms: As I said, not from the over servicing, but providing incredible service every single time and being fastidious in that. There has to be a little bit of tolerance into, okay, you know, how can we serve more customers, get faster at doing that and therefore be better in that way? You know, rather than spending, you know, probably a little bit of an excessive amount of time on a service reply. So yeah, that was a really good one. We were the first organisation in the world to use service replies to email a customer back from Salesforce.
Paul Atkins: Yeah. Nice.
Kane Simms: That’s really good. I suppose the other thing there is that, you know, for the people who are actually sending the emails, there’s a little bit of change management to go on there, isn’t there as well? Because all of a sudden there’s something they used to do that they’re not doing anymore. They’ve got to develop trust in the system, trust in the outcome, you know, all that kind of stuff. And that kind of whole — you know, it’s a different one, it’s a different way of working. And two, you know, it’s probably that what an AI model produces is probably not as good as they would handcraft themselves. Correct. You know, it’s a little bit like in any management role, isn’t it? There are people who you manage who you probably would be able to do a better job than. But that’s not the point. You know, you can’t do everything all the time, and so you’ve kind of got to just, you know, as you said, like, does it answer the email? Is it good enough? Let’s go, then. You know.
Paul Atkins: I think that’s where teams like business readiness teams, right, and a lot of people are talking — you know, you’ll have seen quite a lot of the events in our world and a lot of people are starting to talk about business readiness and take it quite seriously now. Whereas it’s been, in the past, in a contact centre environment, for example, it was just about kind of the conduit between a large platform migration or whatever, you know, a large change thing happening, and, you know, your base population of users ultimately. It’s, I think it’s evolved now as a role in of itself, in that it can perform as a role within a transformation unit. Effort to be great programme managers, great road mappers, the ability to translate what’s coming up and what that means for people’s day-to-day. Liaise with training on what that means and build, you know, collaborative solutions, but also for, you know, the often overlooked bit of transformation, which is: can you get it to stick? Can you get it to be adopted? Can you get it to be adhered to? And then can you evolve it, right, with a great user community that are pouring that feedback into the functionality or the change or the tool that you’ve delivered, right? And so that role, I think, the volumes ramped up really on the importance of that role. I would say in the last couple of years, from what it used to be, which felt quite transactional. Actually, they’re doing this, this is what it means, I need to communicate that to you. It’s way more than that now. My readiness team are, you know, qualified programme managers. Some of them are doing their AI apprenticeships. It’s become more of a rounded value-add role in any organisation, I would say, that’s going through any sort of transformation and AI journey. And so stuff like hypercare and speed to happiness that they measure for their users, the champions groups that they work with in assessing features and assessing viability. What is good enough? You know, the question you’re asking is what is good enough here? Well, we used our people to help tell us that, right? And the level of automation that we would get in, and how we would build that into our responses and our knowledge bases that sit behind it — all that user input, business readiness are the conduit for all of that stuff. So, yeah, I think that’s been a real key component of our success in deploying some of this stuff and being the first out of the blocks in some instances too.
Kane Simms: That’s really good. Yeah. I think the, you know, the role changes as well, doesn’t it. Because the more AI solutions you have, someone needs to manage them. Someone needs to look after them, improve them, analyse them, audit them, all that kind of stuff. And the contact centre is a good place to do that because they are the closest to the customer. They understand what good looks like. They understand the business as well, which — yeah, it makes sense. What was I going — oh, yeah. Knowledge. You were talking about knowledge earlier. And, you know, with these different sorts of solutions, you have voice, you have chat, you have emails, and who knows, you know, what kind of comes next. All of them, you know, potentially having crossovers in use cases, potentially having crossovers in knowledge, all that kind of stuff. How are you thinking about that, that knowledge management problem?
Paul Atkins: For me, it’s the next frontier, right. Because we’ve got some great stuff being brought out, right. There’s some great companies — you speak to a lot of them — and some great products out there that can, you know, deal with some really chunky use cases. In the background, though, how do you power those things, so you can have some of the sexiest AI kit available on the market, and you can build it and deploy it? But the knowledge that powers it, the answers that you give to your customers — your product might change, your prices are probably going to change, terms and conditions change, business processes, you know, in any organisation change on a daily basis. So how do you make sure that the engine that powers all of that cool AI stuff is as current as it can be, and it’s curated? Now, remember back in the day, you used to have, you’d have like a knowledge base and you’d have a knowledge team, I think. Remember back at Virgin Media, we had a knowledge team, and a couple of other organisations I’ve worked in as well. And you know, in reality you’ve got, you know, you’ve got your AI that’s customer facing, you’ve got your kind of internal AI, your knowledge that your kind of customer-facing representatives have access to. You’ve got product knowledge. So you’ve got product teams developing really cool new stuff and making tweaks to the product and its pricing structure almost all the time. And that will change the answer that you’re going to give to your customers. So how actually do you curate all of this stuff? Do you do it centrally? Do you have a federated model that’s got some, maybe even some AI, you know, kind of database joined functionality in there that keeps things up to date in a cascaded way? Really, really interesting subject, right? Because I think that’s kind of where we’re getting to now, where a lot of organisations have matured in what they have and how they use it in an AI sense, but as you add and layer in more functionality and evolve it, how do the other parts of your business that access ultimately the same centralised bit of knowledge, how do they help author, curate and access it in the same way? And how is it all kept up to date? You know, going at the pace of change that most organisations want to go to? So I think actually we had some speakers at the session that you and I were at that were talking about kind of LLM native wikis.
Kane Simms: Yeah.
Paul Atkins: So this kind of concept of contribution from lots of places with semantic synthesis, stuff that’s like continuously drafted that people can query. So as I was referencing, a bit of AI in between these models that sit, or these knowledge bases that sit. And I think that’s really interesting. I think it’s a really interesting departure from what we’ve had in and around keyword searches and, you know, centralised curation and knowledge article suggestions and kind of more of the old school stuff, which is, you know, it’s great. It’s served the purpose for, you know, the couple of decades I’ve been in this world. But I think the knowledge base bit and the powering of the AI solutions is almost where we might bounce back to, to look at, because it will ultimately hamper the speed at which and the agility at which you can deploy AI solutions and keep them as accurate and as, you know, as pacily evolved as you’d want to as an organisation. Kind of got to look at the engine and lift the bonnet up a little bit at some point. And I think that’s where we’re hurtling towards, right? So those forward-thinking AI companies will be the ones that are thinking about that right now and starting to solution it and kind of approaching organisations and saying, right, you’ve got, you know, you’ve got your product database, you’ve got your internal knowledge database, you’ve got kind of your FAQs, you’ve got all of your training literature, you’ve got your customer-facing FAQs as well. You’ve got your internal knowledge for your staff and training material, product, legal terms, Ts and Cs, and all that kind of pricing structures. Here’s a solution for you to keep all that continuously curated, continuously updated in one place so anyone, internal or external, can reference them at the right journey point and at the right touch point. I think whoever cracks that is on to an absolute winner.
Kane Simms: Yeah, absolutely. And I think there’s going to be new roles there as well. And I think, you know, some organisations have like content people, content designers and stuff like that that are starting to try and do this thing together. But as you’ve kind of said, the problem is they don’t have the tools. So they’ve got a kind of fair idea of what they want and what they need, but they don’t have the tools to be able to do it. It reminds me a little bit of, you know, when I used to do a lot of work in government, there used to be, you know, as you can imagine, very much siloed systems, depending on what service you need. It’s all sitting in different systems. And so you had, like, one division has, like, a customer data set because it’s all for one use case, but another department has another system where the end user has to create a completely different account to access those services. And so now you’ve got two customer records, two lots of data associated with the same customer. And they don’t talk to each other.
Paul Atkins: Yeah.
Kane Simms: And so it’s this kind of like — and that was a real big kind of hindrance for decades in government, central and local, which is just sporadic data in different systems all needed to be joined together. And we’re kind of, you know, we’re there again with knowledge now, you know. And I don’t know — I think given where data is stored in a business, it seems logical that you have this. For me, there’s this potential to have this kind of knowledge layer, which is everything vectorised across all of your systems, all in one place, all vectorised, and almost like a headless CMS type of thing where anything can get in and out of it. Above that, then you have like a layer which is your, I suppose, your AI layer, which is essentially your kind of business logic and the translation orchestration. Yeah, the orchestration and the business logic and the translation of the knowledge and policies and tools and stuff like that with the user front end. But I don’t know. I mean, the state of some businesses, I don’t know. It seems like — I think a lot of the AI platforms themselves probably view themselves perhaps as part of that, because a lot of the platforms are like, okay, just give us your data and we’ll just do it ourselves, you know? So is that part of that platform? Is it a separate platform? I don’t know. It’s a very interesting point, but it’s definitely — as you start to grow your use cases, it starts to become a very real problem.
Paul Atkins: Yeah, I think so. I think so. And therefore, if you offer it as part of the solution here at the front end, customer facing, almost backwards into your business data. What about if you build some — I don’t know, some OCR or like process orchestration and automation stuff over here that is accessing — but that might be from a different vendor.
Kane Simms: Yeah.
Paul Atkins: So then all of a sudden, oh okay, do we need another set of things over here?
Kane Simms: Yeah.
Paul Atkins: Do we need to build? Are we going to build and curate a different batch of resources and knowledge for that to access? Or is it actually — do we come from core data back out into a like a federated data set or a staging data set? And then actually then that has sync problems. And who’s got the latest version of what, you know? Like, even then in that sentence, I’ve gone off. You know, I’m all over the place introducing complexity into my business, right? Yeah. So I think you’re absolutely right. I think I’m not sure there’s a silver bullet. I think there will be some great products that tackle it in slightly different ways. And I think you’re absolutely spot on. There will be — AI will be involved in that for sure. Because as we were talking about, it used to be teams of people, right, that would keep things up to date. And even then you’ve got some like next best action functionality coming in and saying, right, this article, is this over? You know, this out of date, or users have rated this article poor. It’s not current knowledge. But the horse is gone by that time — the user’s telling you it’s not the right thing. You’ve failed. You’ve failed to give that user what they needed in the moment, right? Yeah, so actually having something that’s proactive and got some intelligence in there feels like the right way to go. But I’m really interested to see what, you know, what happens. Obviously, as we said, you know, LLM native wikis have been put out there as a potential way to tackle it. Federated data sets, all sorts of different solutions with kind of AI in between each data set that takes cascades, updates, pushes, you know, to that portion of your internal population and then reflects it in your customer-facing knowledge and that kind of stuff automatically. That feels like another way. So yeah, I’m really intrigued to see which way this goes and who comes out with what first. Right, because there’s already, you know, there’s already some stuff out there and there’s already a bit, some stuff being talked about. But yeah, I’m still a bit — I don’t know enough about it, right? Still yet to understand what the best thing is, for which organisation type and which level of maturity and all that kind of thing, right? Which will play a factor in what you decide to do ultimately.
Kane Simms: Yeah, yeah, absolutely. Oh, absolutely. I think that yeah, knowledge management and content kind of ownership is, I think, yeah, definitely the thing for 26, 27. What are you kind of, you know, either excited about or really looking forward to? What are the things that are kind of, yeah, really getting you excited around AI at the moment?
Paul Atkins: So I think, personally, in our organisation, we’re starting to have conversations now about AI being in our product environment. And by that I mean, like, in our app, right, where we’re navigating people to healthcare outcomes ultimately. And we’re just starting to do, like — we’ve grown a bit of a network now of healthcare providers that we have relationships with, that customers can access discounts and then obviously claim back from us on their policy, and we’re starting to kind of create that web of offerings for a customer when they arrive at our app. What I’m kind of excited about now is how we can concierge that journey and help the navigation of that in a more dynamic and conversational way. So instead of me going into an app, looking at a carousel of products or whatever, thinking about my back that’s really sore at the moment and going, well, what am I going to do? Do I need a GP appointment? And then is it that physio provider, and can I claim that back out of my pot, or is that discounted, or is that in my area? Being able to have a conversation about that in the product, in the app, and almost have a solution packaged for you, and you’re just clicking, yeah, that sounds great, perfect. And you’re able to get what you need in that way without having to do any of the like research and all that kind of stuff mentally, whilst you’re thinking about a healthcare issue that you’ve got. That’s really interesting to me. That’s kind of where we’re starting to build out our product and our offering. And obviously AI is definitely going to play a part in that. You know, where and how that fits in and all that kind of stuff. Oh, you remember the Microsoft paperclip? Oh, I see, you’re trying to do something here. Do you want to just have a chat with me and we’ll get it sorted. But that being AI-powered, that sort of really helpful — it’s like concierge, right? It’s like, let me just sort that for you type thing. So having that in the product, and almost how our AI capability that we’ve developed in service channels and in sales channels, you know, moves into actually kind of using that capability in a product sense. I think that’s really exciting, personally. And then I think I’m really excited to see how we get from agentic into proper Gen AI, because it doesn’t feel like anyone’s willing to go there. There are some — go there certainly in regulated industries. We’re still in agentic, taking action and keep the guardrails on. But I’m interested to see where that goes. And I think, as we’ve talked about before, it’s changed the nature of, certainly in customer services and customer contact, the nature of the roles you’ve got. You’ve almost got democratised technology here a little bit, right? In the sense that it’s not just the purview of your tech teams to build and run and support stuff. You’re actually starting to put that capability in the hands of people that have directly served customers and know the journeys inside out, and therefore are able to collaboratively, and enabled by tech and overseen by tech, right, with the right change release and the right safety and guidelines and expertise, we’re starting to see that really pick up pace as a thing that organisations are doing across the board. So I’m interested to see how that develops and then the impact it has on role types within an organisation, right? And new hybrid style roles and all that kind of stuff. So I think there are a few things that we’re keeping an eye on, and you and I have talked about as well. That’s quite interesting to see. We don’t know where it’s going, but certainly be an interesting one to watch.
Kane Simms: Absolutely. Couldn’t agree more. Yeah. And maybe, maybe in a few months’ time or years’ time or so, we’ll see where you’re at. Maybe we can do this again sometime.
Paul Atkins: Yeah, that’d be great. That’d be great. Really enjoyed it.
Kane Simms: Nice one. Appreciate it, Paul. Thanks so much for your time. And thank you all for tuning in. We’ll see you again on the next one. Cheers now. Bye.
