Andy Kingston joins us to talk us through the ways that HSBC is using conversational AI to improve customer experience.
Presented by Deepgram
HSBC using conversational AI
Learn how one of the world’s largest global banks is using conversational AI in its call centre to improve customer experience. Andy Kingston, Head of UK Customer Service Strategy and Transformation at HSBC, joins us to share key insights and learnings from HSBC’s journey with conversational AI.
00:00 Intro and welcome Deepgram
00:57 Introduction to Andy Kingston
03:34 Where did the need from conversational AI come from?
07:37 Meeting customer expectations
12:30 Different pros and cons of voice and chat
15:07 Building language models
16:25 Starting with chat channels
21:25 Collaboration with internal teams
27:02 Future of all channels having a AI assistant
31:02 Current supported use cases
36:12 Other customer needs that aren’t available yet
39:05 Measuring the value
45:30 Containment rate and conversation success
49:15 Judging quality from agents and assistants
51:30 Can some bots be too quick?
54:00 Levelling up communication
Got it. That should be about it. We’ll hang fire and hopefully you should appear on LinkedIn in a moment. But in the meanwhile, Andy Kingston Welcome to VUX world, my friend.
Thank you. Great to be here.
It’s a pleasure to have you alongside. Definitely it’s a bit of a different set of the usual we’ve we’re broadcasting through zoom. So yeah deepgram Apologies your logo isn’t on the screen, but I will give a shout out to deepgram before we kick off for presenting VUX, deepgram is industry leading speech recognition. And if you are in the market for speech recognition, to build voice assistants, or anytime any kind of transcription service that you need in your call center in your products wherever it might be. Deepgram is the place to go for that deepgraham.com/vuxworld is where you can go to find more information that is deep gram.com/vuxworld. And there we have it. So yeah, thank you for joining us, Andy. Appreciate your time. Whereabouts are you in the country again?
I mean, Yorkshire,
Yorkshire. Likewise. Likewise, we are a couple of Yeah, Yorkshire goers right now. It’s weather’s Alright, as well. I’ve been gripping a little bit recently because I don’t think that Harrogate has been overly kind to me in terms of sunshine, but the last few days seem to be alright, so it’s on the up now.
Yeah, it’s looking nice. It’s been sunny here today. Just cloudy.
Yeah. Yeah. Nice. Nice. So Andy, you’re the head of UK customer service, strategy and transformation. at HSBC. A lot of people tuning in now will definitely be aware of HSBC, huge bank, global global operation. And where members will start we’re gonna get into the conversation about how you, your team and the organization has been using conversational AI. But most of all, you kick off members a bit about yourself might be interesting in terms of your background, and you know, how you started, HSBC and me a little bit about your role as well be interesting.
Yeah. So 22 years, nearly, with HSBC now, started on the management Trainee Scheme, mostly in and around the branch networks, it’s been 15 years, working in branches across the UK, short stints in head office, before the last five years being part of UK contact centers, rather more recently renamed UK customer service team. My role is to look at how we deliver great customer service through the adoption of technology, but blend in with our people. And I’m sure you know, we’ll come into some examples of what that looks and feels like in this conversation. And in the last three, nearly three and a half years now has been about conversational banking, conversational AI, how we adopt some of those technologies and some of the ways of working with our people kind of sat in and around it to turbo charge it for our customers. And it’s going pretty well.
Nice, nice. And so I’m aware of HSBC, exploring channels, like voice assistants and stuff like that, I won’t name the person there, but had plenty of conversations around how HSBC has been exploring some of those emerging channels, you know, voice enabled devices, voice assistant platform, stuff like that. In your kind of area, looking at customer contact, contact center operations, various other digital channels. Where did the need come from? Like, why why? Why bother with this stuff? Like where did it all start?
Yeah, go back four years, we were predominantly a call center, named a contact centre, but really a call centre that also handled a bit of email. But what we saw is customers were more and more utilising our digital channels. So public website, internet banking, mobile banking. And we knew we needed to move the service closer to those interactions.
What we see today is a huge growth in digital Logan’s customers accessing the app nearly every day now on average. And we know we knew at the time that we needed to move, move the customer service part closer to where those customers were, rather than asking them to come to the contact centre. So the last three, three and four years has been about creating new channels, creating new contact points for customers to get to be able to get access to help. And I was I guess what we’ve seen is the Panasonic kind of shone a light on him on anything, but it was already there was the customers were starting to adopt this kind of on demand lifestyle, in their banking. So I’ve talked passionately about on demand banking for some time now, because we see it we will customers want the same control in their banking as they do with television, online shopping, food delivery, it’s the same, it’s the same needs you’re trying to meet, just the stakes are slightly higher, you know, if you’re, if you’re on demand pizza ordering goes wrong, you might have some upset children sat around the dinner table.
If you’re on demand banking goes wrong, there can be some pretty big ramifications. So having the humans sat alongside that technology is, I think, is absolutely vital in in financial services, in particular. And therefore, I’d say, you know, we’re a team of about 4000 people serving customers every day. The the idea was, how do we get them closer to our customers, and that was digital. So we set up chat channels, through internet banking, through mobile banking. And we’ve, over the last few years worked on how we grow the use of that channel, not only for contact centres, but the wider kind of retail part of HSBC business, in particular, because customers tend to come to the contact centre as a starting point. But they can often often head off in lots of different directions, depending on what their inquires. That’s, that’s the kind of telephony based approach that works. Well, why would you not have that in chat as well. So we’ve been working on, like fraud support teams and underwriting teams. And so the chat services is pretty seamless for our customers. And that’s, that’s been a big part of the plan for the last few years.
Interesting. So it can then out of the rising customer expectations. You’ve mentioned pizza ordering and stuff like that. And everybody obviously is kind of a digital native. And I suppose the difference between a product and a service these days is blurring. Is Uber a product or a service like Spotify, a product or a service? You know, it’s kind of like that lands getting really great Greer, the more that people do online, the way other people do on their phones and stuff like that, the more they expect, and therefore, I’ve honored I’ve actually moved banks a number of times because of poor experiences elsewhere.
And what mentioned the brands where it’s not HSBC, but it’s, I’ve had poor experiences, and I’ve moved banks because of it. And so it was that the kind of the main driving factor was the fact that really meeting customer expectations was at the heart of it, or was there any other drivers that were going on? You mentioned the pandemic, I imagine, you know, all businesses were hit kind of fairly hard during a pandemic, was there any other drivers or was it predominantly to try and really kind of get ahead as far as meeting customer expectations are concerned
that was front and centre was a valuable customer customer service, we needed to evolve, and we need to evolve and get closer to where our customers are? And where we, where we believe they would continue to be and say the growth in digital channels is yes, it’s accelerated because of the pandemic but it was already there, it was already happening. Is it more efficient? Yes. So you know, there’s a commercial benefit to doing this but that wasn’t the driver the driver was customer experience, you know, that you’ve touched on moving your banking Cane the, the relationship the customer has with their bank is still one of trust. And it’s one of trust that works when the bank fixes something or reassures you or gives you support in a certain way. And that’s when things go wrong is generally where that’s fallen down. And that you can blame processes you can blame technology but often it’s the people in the middle of it, the contact centre, workers themselves are often stuck in the middle of it and the customers so it’s, it’s a people on people issues still every day.
So part of the strategy here was not only about creating new channels for customers to contact us, and getting our people closer to where those customers are, but also then being able to utilize the technology but with the people at the right times. So you know, no, I don’t think anybody wants to run a business where they’re constantly having to employ more and more and more people and driving up the cost base that just doesn’t make any commercial sense.
So you know, having customers in a place where you can deploy some automation deploying some, you know, some chat bots to provide that kind of on demand banking experience that I touched on, but always having the mindset that customers can always drop out to an agent at any point anytime you know, with a dead end they they always have to be an exit point. With central because it is a customer service play This is the strategy is about offering great service. And we see it in our NPS scores and our NPS scores on, on our chat channels are twice what they are invoice. So customers are, are voting for it to say they really like it.
Interesting. That’s interesting. You always, I always think that the voice channel feels more immediate. But the chat channel I suppose is more kind of contextual and easier to adopt. Because you don’t need to switch channels almost if you’re already in the app, you’re already on the website, whatever like that. Movies, chats more natural, more, it’s more conversational,
as in you can, there’s the way we structured it, there’s no distinct start and stop to that conversation. So our chat conversations, customers can leave it and come back to at any point. And yes, we try to be there as quick as we can, you know, the certain interactions where customers just need an incident response. But there’s certain interactions where the wait a little bit of time, or they want to go at their own pace.
When I think about how you know how we were in our lives today, you know, that you fit in and pick up the kids from school, you’re fitting in different work priorities, banking isn’t, you know, has to be it has to be able to fit around all of that all around life. I think it’s an important theory to this. And that providing a chat service, where it is a continuous conversation at the customer at the pace they want to go out, I think is a really important part. Whereas when they think about phone calls, if there is a distinct start and stuff, you often have to care about time, you know, we’ve all I think we’ve probably all had the experience where somebody in a contact centre of any you know, of any organisation has promised to call you back and they haven’t any in the frustrations of that. Whereas the chat conversation and the way we structured it, where you can pick it up whenever you want. The customer is in control of that. Not not not the organisation?
Yeah, there’s less pressure as well, when it comes to the automation side of things. You know, the problem sometimes with voice user interfaces, for fairly complex use cases is that sometimes it requires a bit of thinking time to think about what our response is going to be and and when to speak it in real time and can’t correct while speak as I speak a lot of NLU systems and Asr systems struggle with that. And so you were the chat channel, I suppose you’ve also, as you mentioned, they’re going up to users Pierce, the user got more time to think more time to consider my time to kind of, you know, think through what it is that they’re seeing sort of thing, which is not a disparage the voice chat, because I think that in the moment, voice user interfaces and voice assistants are absolutely ideal. It’s just that some channels have different kinds of, you know, pros and cons, isn’t it?
Yeah, we’ve you know, don’t get me wrong, we’ve invested heavily in our voice technology. We replaced the IVR, input conversational IVR, in a couple of years ago, so we that was a sizable undertaking to deliver that, and that has unlocked the opportunity to improve service forever, you know, the, the amount of agents agent transfers jobs, because we, you know, we better understood what the customer was actually calling us for, as well as you know, offering out enhanced self serve capabilities. So, you know, created space in there in the telephone agent population to handle the slightly more complex or slightly more emotionally driven, event driven challenges that customers customers face.
And, you know, it’s playing out today is more and more customers will be lost on anybody that the cost of living challenges in the UK right, that, you know, they’re out there, aren’t they? And customers are coming to the bank to help, how can I, you know, I don’t need people processing payments or transactional banking on the phone, really, I need people helping our customers when they really need that support. So we’ve invested in telephony, infrastructure and technology as much as open up the channels. So customers now have a choice. And we’ve got, again, the same principles, the same blend of in the IVR, you can always drop out to, you know, always drop out. So human, as always, there’s always a human voice to help you behind that. You can, you know, this week, we’re launching the ability to even cross from the IVR into the chat channels to continue conversation there. So have that ability to continue that, that interaction rather than the distinct start and stop a voice.
interesting. Yeah. Yeah, that’s good. I mean, the routing is a really good use case because you want to solve a real customer and Agent problem, but you also gather such a wealth of data, because you’re not only getting people saying transfer me to To, you know, my complaints are transformative in balance, inquiries, whatever. It’s often very specific things that they’ll say, if you could build a language model around that you’re able to gather data that was previously unattainable or just going straight into the heads of the agents. And,
yeah, it was, it was a really interesting programme to actually even start putting the conversational IVR in because the amount of utterances is just mind blowing, if it’s just one use case, moving house, you know, there’s probably 100 different ways you could articulate you want to change your address. And trying to capture those, and then, you know, we’ve referenced it quite a bit because it’s not getting away from, you know, the pandemic flu in new words, new new phrases that we hadn’t even come across. Before, you know, that we weren’t in commonplace language and didn’t exist, that we’ve had to adopt, you know, adapt to and find ways of advertising that see, right, the language models being and then continuous tuning of that language model is, is absolutely critical.
Absolutely. Yeah. furlough. I didn’t even know
probably one of the most Google words when it first came out. Yeah, it really looking at what it meant.
Yeah. It sounds like a horse wrist and it for for for furloughs to go. Yeah, it’s mad. So. So did you start with Chatlin? If you go back to you know, you mentioned the other key kind of drivers work? Was it the chat channels that you started with first, right?
Yeah. So we were voice dominated bit of email. And really, it was difficult to get chat up and running. That’s where our customers are in the digital arena, we’re going to class. So if I remember correctly, I think it was to chat on a public website, then enter banking, then into the mobile app. And integrations that came there, that I think that was the biggest blocker, you know, really getting into the mobile app, we created this ability to move from kind of, we almost operated with very synchronous li in on public website and internet banking, though, it was almost just the replication of voice, which was never really where I wanted to take it. I want to take it into a genuine conversation.
So mobile, banking, and mobile chat has really been that game changer. And that’s where we see a lot of the customer contact today is in that mobile chat space. And we spend a bit of time operating without a bot, we decided not to do a bot on day one didn’t think it was the right thing to do. We wanted to learn what our customers wanted to chat about. And in to your point about data streams. So data rich, being able to interrogate chat transcripts and put into analytics tools to to draw out themes and the type of language. So we gave it some time, before we launched our first spark, or even before we started to try and launch a festival, I mean, frankly, came our first attempt wasn’t very successful, I think, probably. To me, they’re the same scars of their first, you know,
you always build it twice is Yeah.
So we were some time before we will probably have nine months post launch of chat before we get going. And even then it has grown from strength to strength and what why I’m really proud of our bot solutionist for was crosses all the challenges, you know, the same bot working across mobile, internet banking public website is that it’s not built by our technology colleagues. So yes, we need it yes, we need digital teams to help us but actually the bottom the bot flows and the conversations are designed by frontline contact centre agents. So we’ve set up a capability where about 5550 ish people that change on a regular basis, but working building a chatbot flows, and these guys are used to serving customers.
So you know, what we said earlier about the language how we need to explain things, how we capture that and then you know, got slightly more technical capability in terms of you know, the NLU and how we build the flows, links to that cake. So that capability, but it is, you know, genuinely is a bot built by the people who said the customers and I’m not saying actually unique and other organisations that are doing that, but that was a very, very simple learn early on. First iteration of a bot didn’t really work. It was very technology driven. Now I’ve got some super talented IT colleagues, but actually what it shone a light on is that it’s the people who are serving the customers that know the customer the best.
And no matter how much data you could throw at it, that human experience to make your bots more human. You know, we don’t hide the fact you’re talking to a bot, we’re not trying to disguise it, you know, we’re very clear that you’re chatting to a bot, but the people building it are the ones that serve the customers as well. And that’s made a, you know, that’s made a really big difference. What it also meant is that we are a little buzzword, but we are really agile. So you know, something happens in the industry, something happens in, you know, within HSBC that we need to, we need to get bought out, we can get it out in 24 hours. Because you know, what, it’s not a tech delivery, it’s a frontline delivery. So where we’re building knowledge content for the contact centre agents or content to go on a public website for customers, you know, for issues happened, we can build a bot exactly the same pace, it can all get launched at the same time.
That’s good, that’s a good way to do it in terms of having some integration with the rest of the business. So that content changes over here, and then it’s updated in the bot or new content is published over here. And it’s also put into the bot is quite a mature way of operating. Usually, it’s the bot team or the assistant team running around like headless chickens, trying to figure out what’s happening in the business and trying to work out what’s changed and trying to get it from somewhere. And it’s like, so would you describe the relationship between the team and the rest of the organisation is a fairly kind of tight knit one, then what’s the kind of collaboration like between internal teams?
Yeah, this is I think it’s really healthy ketamine, it helps this kind of vision this direction, we’ve had the best exec sponsorship, you could hope for, you know, our CEOs, things that sees the value. You know, so he’s been really keen for us to do this and really supportive. There, the ability for us to scale comes from that sponsorship, or there’s one part of it, you know, this isn’t a contact centre capability anymore. This is an organizational capability.
So, you know, it’s touched on earlier, that the contact centre might be the front door, but we have specialist teams now working in this area as well, you know, and we’re just just working on right now, for our financial support team for those customers who are facing some financial challenges we’re working on, actually, how does how does chat, how does a bot actually help in that scenario, because, again, put the customer lens on it, some people will, you know, be embarrassed or may not be in an environment where they can openly talk about their finances.
So they may, they might want to chat because it’s more private, or it’s, you know, it’s probably potentially slightly less embarrassing to chat with somebody than then tell them if, you know, if you’re facing a bit of trouble, so and then, you know, what we’re trying to work through there is then Okay, so we still want a human to deal with acts really progress and emotional event that many humans do, but actually can still use a bit bit of AI a bit of a bot in there maybe to do the data collection, because if you’re going to help somebody build some kind of financial plan or repayment plan or you know, whatever they need, we do need the data, we need the information, sometimes you got to go out from a customer. So you know, we’ll work through how we build like a tango bot that can come in as almost like an assistant of the agent. So yeah, the business is bought in, which I think is really, really important.
I’m at the point now, which I really proud of is that I get requests asking for my team’s expertise into other parts of business to help them do it. So it feels like we’ve got momentum, it feels like we’ve got lots of people that really interested in buying in to what we’re trying to do. And that, again, is making a big difference. Because if we look at the number of customers now utilising our chat channels, and using this technology and crossing from voice to chat, and it’s, it’s growing massively. And I definitely think that’s because, as a business, customers don’t differentiate, being either dealing with HSBC, they’re not dealing with HSBC customer services or fraud, you know, it’s just we’re a brand and we’re a bank. And, frankly, customers don’t care which department or just one serving. So the more parts of the organisation we can have working in this way utilising this technology, but in the same ethos, and the same theory can only be good for the customers. And they are definitely voting with their feet because the numbers are growing hugely.
Definitely. That’s when it starts to become a strategic capability. You know, when you think about a good example it uses if you look at some of, I suppose your competitors over the pond cover or one, for example, where they’re kind of assistant, as you said, it’s the same, it’s exactly the same when it comes to digital service delivery. In general, if you’ve got an app, customers go to the app, they don’t expect to have to go into the Fraud Division and deal with the fraud departments architecture in and systems and processes, and then go into the customer service conversations have to deal with the customer service systems and processes. It’s all just here, HSBC, it’s all just the brand.
And when you speak to the brand about anything, you expect consistency across all of those conversations, and therefore having a front facing assistant that’s there as your first point of contact for only all of those issues and across all of those channels. Seems to me to be the way that things are heading. I think one of the really good examples is Capital One where their assistant is available as an Alexa developer in a call centre, it’s in the app. But not only is it a conversational assistant, it’s also a genuine digital assistant, because it’s there and email, it will send you proactive emails, and it will come from email or not from Capital One, they’ve got like browser extensions to facilitate secure payments on websites that sort of that all branded as email. And so my question always is, what point do customers stop thinking that they’re interacting with Capital One the brand, at what point does ainol become its own brand. And at that point, it is well and truly not only an internal kind of capability, but it is definitely a strategic differentiator.
It sounds as though from what you’re explaining from internal support and buy-in at a senior level plus, internal departments lean in on your team for expertise plus channel expansion, multi channel use cases. And imagine we haven’t gotten to use cases yet. But I imagine you’re doing some kind of transactional use cases, as well as the kind of low hanging fruit on the FAQ side. So it sounds as though you’re developing that capability for this assistant to be the front facing thing across most channels, the front kind of first touch point across most channels, is that where you see the future of customer experience head on? Is every channel having an AI mediated assistant on that.
I absolutely do, Kane and I think where we are, we potentially could end up a few years down the line. Yeah. But I think where we could get to is that we have a very active runway of functionality we’re trying to build inside digital channels, all based on customer demand, all based on what our customers are asking us for. That will get to a point where you know, you, you will never fill in that. But can you make an app that busy with that much functionality that becomes hard to navigate, because we’ve seen that with websites.
Now and I think we’ll get to a point where a virtual assistant and a genuine virtual assistant, not just a bot in the chat channel will be almost be like the concierge it’d be at the front of of an app and it will, it will direct you around it will take you to the parts a bit like an IVR dots connected to what based on what you need is even if that’s just to flick across pages or carousels in an app, I think will we will as a consumer base, we will almost self serve the on navigate the app less it will get done for us. And and that opens up different different capabilities. Because you know, as you’ve touched on this, you know, voice assistants, whether it’s Alexa or you know, other brands are available, obviously.
The question I was asked last week, I think, I don’t know the answer to yet. But was, you know, where does voice notes fit in our strategy? Where, you know, we see it in Asia quite a bit with customers talking to their devices, sending voice notes, see it in the kind of younger generations using voice notes to interact more. So you know, this four or five years ago, people said to me, my voice is bad enough, then it’s going to move to chat. Well, no voice isn’t dead. Voice is resurgent, but probably in a different structure. And it’s probably invoice notes rather than phone calls. And we got to work out what the strategy is for that. Because it will be slightly different. It’s got lots of similarities to kind of an asynchronous chat experience but it will again will be slightly different and how we make it conversational and how the brand it how do you make it feel? As humans you want to make it feel?
They’re all interesting dilemmas to face. Yeah, absolutely. I think we’re we’re not there yet in terms of creatinga service where customers recognise it as anything other than HSBC. Capital One was clearly operating in a different space there. But do I think it’s something that over time customers will, almost default as their contact point? Yes. And it’s already happening now. We’re seeing it with millions of customers now, naturally going to chat and actually using a bot using the technology and not defaulting to the phone. And that hasn’t taken that long for that to change.
Interesting. Interesting. Yeah. I mean, it doesn’t even have to be branded, you know that, you know, what, they made a very conscious decision to make it a nine to make it its own kind of thing. But, you know, it doesn’t even need to be that, to be honest, as long as it’s delivering value for customers and delivering value for the business. And that’s all it really matters. What kind of, I imagine when you started, you’re starting with some kind of low hanging fruit as I was kind of alluding to FAQs and stuff like that. What kind of range of use cases that you could run. Now, you mentioned routing in the IVR, I think you alluded to some self service in the IVR, as well, I wonder if you’d just give us a lowdown on on voice and chat, the nature of the use cases that that you’ll cover in a minute.
Yeah. So on the voice side, we do a bit of a bit of transactional, a bit of colour information provision, not necessarily FAQs, or such, but it may be simple, simple interactions, like playing out a list of transactions, through to the ability to cancel, cancel cards, replace cards, etc.
I think there’s more opportunity in that space. I think it predominantly still revolves around transacting, you know, changes and addressing those type of things. But really focused on the high frequency activity that I think that’s where an IVR does, its heavy lifting. But we do you know, we do use it to try and help educate and steer and direct as well. So, you know, we’ll, we’ll, we spend a lot of time looking at the messaging in the IVR. And how we infuse data in the IVR, to make sure we play messages that are relevant to our customers. In a chatbot space, you’re much more FAQ focused. We have got some transactional bots running today, which are mostly based around card usage and card inquiries interact, you know, cat interactions, which was partly use cases, partly because of some regulatory change that we wanted to support. And we knew that, from a retailer perspective, they had a bit of an adoption curve to go through three commerce transactions. But for our customers, we did as well. And we wanted to make sure that both the bottom and the IVR could support that and support customers. I think when you have a payment declined online, for whatever you’re trying to do. It you know, it’s hard not to default to the worst possible thought first, you know, it’s on the cards been declined, I’ve got, you know, as many as my bank account empty or something gone wrong, actually, you know, in those scenarios, it could be something as simple as the transaction hasn’t been quite processed in the right way. You know, we want to be able to very quickly reassure our customers that, actually, there’s nothing wrong, you just just ask the retailer to reprocess it, they’ll send it down the right route, and it’ll all be everything will be fine.
So, you know, having a bot that’s just got the rate of intelligence having an IVR. So that the intelligence to recognise that to be able to differentiate the different decline codes, all of those types of things that might happen in the background, was our first step really into a fulfilment bar transaction bot. And I’ve got a huge long list of things that I would like to do. I think the most important bit when this came in was what I’m seeing is that I don’t think there’s a huge value in trying to get a bot to do what customers can do themselves in the digital space.
And particularly now you can make a payment in three clicks on our app. Why? Why would I want a bot to replicate that? Well, what I’d like a bot to do is to enhance that experience, you know, so, for example, paying credit card bill couldn’t get any easier than doing it in the app, simply saying, well, but if you want to chat to the bar, about your credit card bill, and actually you ask it to make the payment for you, and then it provides you access to information that you can’t self serve. have, you know interest rate your arm, if you pay if you’re not making the full payment every month, how long it will take to pay off or, you know, and present some offer up value adding information that, you know, 20 years ago, 30 years years ago, you walk in a branch that might have told you that information, you know, can a bot do that value add? I think you can, and I think it should. So if you’re gonna get the bot to transact, not just replicating a digital interaction, it needs to be more value added. And that’s the bit that we’re now working on. So the right long list of use cases. But what’s all what’s all the subcategories off that that would actually add genuine value to that transaction?
That’s really sensible. Yeah, I suppose there’s, yeah, if something’s working, well, why fix it? I suppose. The other thing that an assistant will give you in certain contexts is new, not necessarily new use cases, but certainly new queries that you might not have anticipated. Just from the digital channel alone. You mentioned they’re paying your credit card bill. People can go into the app, enter a credit card bill, and it’s done and dusted transfer money from one account to the other. It’s done and dusted. What you don’t get is, what else do people need? That you’re not yet serving but you don’t know about it, And I think that’s an interesting space where assistants can play like, are people asking what their interest rate is? Are people asking what the next month’s payment is? Are people asking about limit and all that kind of stuff? Like? What would be the needs that arise? As you see the subcategory to the broad use case? And is there anything in there that you’re able to uncover through virtue of having an assistant and therefore, you’re not you’re not exposing more needs, what you’re doing is essentially capturing and delivering against needs that were there, but being previously unmet. Does that make sense? Yeah.
100%. And, and this is where having a, you know, I’m really lucky, I’ve got a great team with me who are looking at alternatives looking at failures in the in the NFL yet where it’s just not detecting what the customers actually asking for. And that’s, that’s the goal to strike. That’s what we’re fun on the IVR. And on the Chatbot side, that’s what’s building those use cases. You know, those, I call them the plus one conversations, you know, you’ve got a primary intent, but what’s the what’s the plus one?
That’s what we find out? And that’s where you can do it the other way? I think it’s because this year we digitised the service, I think what we get is we actually look at what are the inquiries you still get in the human channels, which you haven’t digitised yet. But but the quickest way to get to it is probably you know, he’s he’s looking at you’re looking at your transcripts, look at the unrecognised intents and saying, right, that isn’t just a change in language, that’s an inquiry that we just are not set up to handle. And, you know, commercially, it’s probably going to be more cost effective to build a bot to handle 3456 of those together than it is to build an interface where the customer can self-serve it.
You still use your mass interactions, you probably still want to be digital teams to build the interface to still build on your app or your website. But actually, for some of us the fringe cases, the smaller those plus ones are certainly I think if you can access a simple API and draw that information, and then have five, six different journey flows running off the back of that information, that’s going to be more cost effective quicker to bring it to your customers and start to add real value to that interaction, which is that’s where you’re going from a transaction to a proper interaction.
Yeah, absolutely. Yeah. You mentioned value there. How? And at the beginning, at the top of the conversation you were talking about, the real driver behind this was to get into the channels where customers are, extend the availability of the service, all that kind of stuff. How are you measuring the value that these assistants are creating now? Like if someone was listening to this and saying, Oh, well, I don’t know whether we should do this or not? Like, what are some of the examples of the value that you think is important in doing this?
Yeah, we could spend three hours trying to talk. I’m trying to give you the quick version, because this is where we’re probably on. Version 2025 Maybe have a calculated train where they are still because this is not easy. What will I say at the outset and it links to something we said earlier about kind of the senior sponsorship on this type of stuff. When you are in this world, you can start to see the value once it’s live once once it’s just an idea.
It takes a bit of a leap of faith. So you know you can calculate as many numbers on a spreadsheet as you want, but until you go up and run with it. Are you never 100% Sure he’s going to add the value think he’s going to add. And when asked whether, whether that’s the chat channel itself, whether it’s investing in IVR experience in that space, you know, proving that a business case is, is invaluable. But there are a few models that, you know, can tell you this type of stuff. So I think, you know, what I would work on, especially in the chatbot space is here, when when we first started, there was a lots of lots of focus on containment, right? I’m not convinced that’s right. Because, you know, anybody who’s been in and around contact centres for any period of time knows, you can basically gamify any, any data, any single KPI you want to, and it’s the same with containment, right? You know, you could just get your bought, turn out your phone number on every inquiry glove on because I’m concerned, right? Just not what you want. So number one is always NPS, we measure NPS religiously.
And that, ultimately is the most important value, because although it is very, very hard to then spin into a commercial bottom line. Benefit, you know, that age old, keep your customers happy, everything else will follow, you still still tune you know, that hasn’t changed. So, you know, the best measure that we’ve got is his MPs. So we measure MPs religiously on what we do with the bar, what we do with chat, what we do with the IVR, what we do with the service as a whole. But in that in the box base, we’ll also look at a bit of containment rate, but we’re looking at it through a lens of what is it we’re trying to contain versus what what do we where do we want it to hand off to human. And when you start to get complicated when you build things like Tango bots, where you’re dropping them in and out, because that throws out your numbers all over the place, that can become a bit tricky. We’ll also look at the number of interactions per conversation.
So although I said earlier that, especially in the chat space, the conversation never starts and stops measuring purposes will normally measure over a 12 hour window. So you can start to run some of your kind of data off the back of it the performance data. And we’ll look at the number of interactions, almost like a customer effort measure to say how many times does the customer have to interact in that conversation and you know, clearly on each conversation will be different, because the need will be different. But when you scale it up over three quarters of a million chats a month, then you can kind of get an average of it for messages and customer senators in that conversation is it five or six he started customer effort, then you can flip it the other way in terms of the bot doing better than his human doing. So you can start to build out your efficiency view there. In the in the IVR space is slightly different, but not significantly.
So NPS still number one, still look at that what I want to understand the customer experience, I do look at containment rate in the IVR probably slightly more than I do on the Chatbot side. And I want to make sure that the IVR is containing in the right way. So that’s why the NPS side comes in the what we’ve what I do focus on specifically in the IVR probably because there’s more exit points into the organisation, our chat, our chat population, yes, we have some specialist teams. But the kind of chat bots basically means to route when a handoff to an agent just needs to route to a group of agents and IVR needs to consider much more specialisms in the routing. So you touched on the transfer rate. I do look at that as a almost like a commercial outcome because you can spin that into a you know, put the dollar sign to that one and work out actually how much does reducing the transfers really cost.
So we’ll look at transfer rates and accuracy of the IVR. And how well it’s detecting the intent does it need to use disambiguation questions? We look at that as well. So if it’s using just some figuration, is it actually enhancing the routing capability? If not, it’s disfiguration certainly isn’t enhancing the customer journey. So you’ve got to make that trade off and say, well, fun activities are big raising questions, and it still doesn’t get routed any better. Why am I even doing it? This is because hopefully it’s okay. It’s very customer focused. That’s where, you know, that was a reason for doing all of this and this is why I stay trying to stay true to that today. Yes, it can measure the commercials and all kinds of spreadsheets and calculators to do that. But ultimately, knowing how well your customers you interacts with the technology and the humans and being able to have almost like the human MPs, the technology MPs and then a blended one for the interaction. That’s the underpinning one. That’s That’s what dictates whether we make changes or when we keep it the same ultimately.
Yeah, that makes sense. It’s good. I think NPS is a good one because at least it’s a Northstar. The problem I think I’ve seen often is that there’s so much data that you can get from this stuff, how many intend to recapture? What was the accuracy? What was our norm matches, and the ones that rate like, and then what our speech recognition rate, and then all this kind of technical data that you can get to and then it’s like, well, in the past, you know, we’ve kind of implemented things where it’d be like in the FAQ side of things, as you’ve talked about contemporary, I concur with you that containment as a phrase is generally a bad kind of phrase to use in this space, because the end of the assistance is not really to contain somebody and like, trap them there.
The aim is to serve them and get them what they need to get done. And if speaking to a human is the best outcome for that use case, and that’s the point of the conversation. So it’s very difficult to use containment railfan. But measuring whether this conversation was successful beyond NPS, like did the user get what, because you’re gonna have a conversation with a chatbot feel it’s a pleasant experience, you know, give it a decent NPS rate? And because there’s nothing really wrong with it, but you might not have actually got the actual answer that to your question.
So the gap that I’ve seen a lot of is, it’s very difficult to measure when a conversation is successful, versus when it’s not, especially with those kind of like quickfire FAQ style, things like it’s quite a quantitative kind of qualitative rather, thing to measure. But I think that Northstar NPS, and everything else flowing from that at least gives you one it’s a valuable metric. And then everything else should come from that everything else is aimed at improving that MB NPS or keeping it where it is, which is it’s good. Yeah. Because a lot of people can get distracted. And it can be quite confusing to figure out where to start with honest stuff, you know? Yeah. And,
you know, we’ve utilised first contact resolution as a measure. But, you know, most contact centre people will be well, I’ve failed with that. We’ve utilised that as a measure of the challenge with with fcr, as we’ve seen more and more customers move to chat. And this kind of asynchronous, this never ending conversation, as it were, is made that really blurry, you know, when your ser is ultimately just going to measure, did the customer come back in a set period of time to ask you the same thing? Well, if I’ve started a conversation, then other than pick the kids up from school, and I come back and I want to pick it up again. No, how is it? How do you go back into ser measures?
So that’s, that’s the kind of historic kind of contact centre what I would have done some. And we still look, we still do look at that. But I don’t think there’s. I think what I’m learning throughout this is there is no one single data point that is going to give you a true view, it’s going to be a blend of them. One thing that I’ve wanted to experiment with a bit more is that you know, we have a quality model for our people, you know, and the quality model should all lead to good customer outcomes and ultimately equal then good MPs. How do we build the same quality model for our power systems? Our virtual agents? Now, does that look and feel the same now? Yeah, the obvious bit there is that the machine is built to work in the way the machine is built. So we probably do it the wrong way around. But could the design of your system lead to good quality? That’s maybe a way of doing it. Because I think you’re spot on a is was it a successful outcome? Unless you ask the customer for what would be a successful outcome today? How are you? Are we really going to know?
it’s, it’s the mosaic affections or can you piece together the important bits of the data to say, that now makes a picture? And I’m happy with that, or I’m not happy with it?
Yeah, absolutely. Yeah. Yeah. It’s interesting. I’ve noticed that a lot of companies actually having the same kind of parity as far as how you judge quality for agents and virtual assistants. And almost there’s actually been deployments where the exact kind of metrics that are used to kind of judge agent performance, are the exact same use for the bot and the bot basically treated exactly the same. Which is interesting, but I also think that in some cases it again, it’s horses for courses, isn’t it because your bot might be doing very specific things that are different to the agent, like your live chat session. This is the case on top of for example, your live chat sessions might expire after a period of time because the person is speaking to other people or whatever. Whereas you know a bot or a messaging use case it might be synchronous and the conversation might last three days. So long as the customer got what they needed from it, the effort on the bots part is negligible. Whereas if it was spending a human five minutes each time to respond to 20 messages over four days, it’s a slightly different level of effort than it is for a bot kind of thing. So there’s still, yeah, this thing hasn’t been solved yet. I don’t think which is really interesting.
No, and that’s the downfall there is you’re trying to measure efficiency and effort on the machine? Why? You know, my, my simple view is being customer obsessive, what’s the effort of the customer? Yeah, does it go at the pace that the customer wanted to go out? You know, as we’ve learned, especially with chatbots, as we learn over the last few years, you can make really subtle little tweaks that make a big difference to the experience in other times where we’ve had the bot, it’s been, it responded to quickly. And it’s almost like seeing your face, it’s almost like this passive aggressive box over here. That’s not very helpful. And actually, that can put some customers off. Whereas, you know, we’ve also experimented in not the other way in for quite big delays in to the point where people like, if my mobile phone signal is still there, I’m still connected to it, because the thinking something’s gone wrong. So even though really little tweaks can make a big difference, you just got to, you got to test it, you got to play around with it and, and work out what works.
Yeah, yeah, it can feel a little bit too, obviously robotic, when the response times are too quick. Like, there’s a book called Mitsu coup. And for years, it won what’s called the Loebner prize, which is like, for a while, I think maybe still today the measure of how good a chatbot can get. And it was created by Steve Wozniak, we’ve had him on a podcast in the past. But if you try out the Mutsuki, I don’t know what it’s like today. But the second you press return, the response is given to you. And it’s just so quick, you’re like, it’s, it’s then obvious that it’s not, it’s obviously it’s definitely pre programmed. Whereas creating a bit of space, at least enough for it to be more human, like, like, you won’t respond to me like that, you’ll need a few seconds, a millisecond or so a couple of milliseconds to try and understand first and then put your response together. And having it too quick makes it feel like it’s obviously robotic, and pre deterministic, rather than natural and conversational kind of thing.
Where I’d love to get to a point where I can almost mirror the customer’s pace. Yeah, so the bot can automatically detect and you can do sentiment analysis today. So why can’t we detect that a customer’s potentially pausing or, you know, being very thoughtful in the way he’s interacting? So where are they interacting? And can that can the bot kind of mirror it? Yeah. But then you’ve got some customers who just want to get on with it and want to get the answer could it speed up or slow down, that would be, that’d be a critical position to try and get bought into and get it work that way.
100%, there’s a company called behavioural signals, which, which just shows it does that but not the full extent. So it will listen to the caller, and how they’re speaking, when they’re talking about who they want to be put through to, and then it will match them with an agent that’s got the same conversation or speaking style. And it’s been proven actually to end up in higher quality conversations. However, they define it with a 10 PS, whatever it is. And so the technology exists to be able to understand the speaking style. But I’ll remember right, in years ago, I think probably in about 2018, about kind of almost what you’re seeing, but it’s like come AI to profile somebody’s personality type for one of a better word, and then be able to respond accordingly. So if you’re the kind of person that jumps in a car and says, give me directions to work, but you really want to understand detail, then it will say, Well, we’re going to have to redirect you because there’s been a traffic jam on the M1, and this is what’s happened, the delays are going to be this much. So I’m going to take you this way instead, and really explain everything. Whereas if you’re someone who doesn’t care about detail, you just want to get to the point or just say okay, redirecting you because of an accident, and that’s it. And so it’s kinda like being able to use the AI to better profile customers to then match it with an agent that is better suited to then level up the quality of the composition. It’s I think we’re a bit further away, but there’s signs of the technology being there that might enable that.
Maybe the choice of language for that intent is part of the key in the scenario of asking Alexa, what’s the weather today versus am I going to get wet today? Yeah, you know, am I going to get wet today? So simple, yes or no answer. So it kind of gives it away that the customer or the consumer just wants that. What wants to know, whereas somebody who’s asking about the weather could be you know, once it is reading out verbatim from the newspaper type of branch.
Yeah, yeah, it’s interesting. What was another corner of town just want to get your thoughts. for people listening in who might be I mean, it sounds from hearing you kind of speak, it sounds as though you’re definitely kind of on the more mature end of organisational maturity. As far as these conversational AI services are considered, you know, you’ve got, I don’t know what the size of the team is gonna get into that. But you’ve got a team that work on this stuff, you’ve got organisational support, at the very top, it’s obviously got budget assigned to where it’s performing well and delivering value, you’ve got internal stakeholders throughout the business coming to you for your expertise, probably in the process of having some kind of hub and spoke model set up where your Centre of Excellence can advise.
So you’re definitely kind of on that scale to definite maturity. If you take a step back, and you’re not people tuning in, who maybe it’s only got one chap or deployed in one channel, there is it’s not quite doing transactional stuff yet, or they’ve got maybe some IVR route in or kind of chat presence and maybe leadership and not really as support as they might be, you know, just generally think about yourself three years ago, whatever. What are the kinds of pieces of advice that you would share some of the critical things that you either did do or would have done differently in the past, if you’re going to advise someone trying to get to the level that you’re at right now?
Yeah, I think the biggest bit of advice I would give is, invest the time to share with your stakeholders, your eternal sponsors, your execs, invest the time to share what you believe with them, to get them to believe it as well. Because that just unlocks everything for you. So that, you know, the biggest change for us was really when we got all of the kind of component parts of the retail business to and we’re not still not there yet, is obviously a big organisation, but the majority of it, we invested heavily over probably 18 months to educate, educate, educate, and just get them to understand that understanding, not how the technology works, the you know, the exactly, that’s always the that’s a those who enjoy knowing that stuff. What does it mean for the customer? What does it mean for you people? It ultimately mean for your business, you know, those are the three things that we try to get across. And then I think the second thing is, growth doesn’t come by just wishing it to happen, you know, you’ve got to be very deliberately go out and get use cases. So you know, when I see people from across the organisation that has asked me for help, or they say, I want what you guys can do.
First thing I say is write to tell me the challenges you face, tell me the challenges that customers are facing that are interacting with your part of the business. And let me help solve the problem solve. Because a lot of this is about problem solving. It’s about getting the technology, the people closer to where the customer is. And so rather than the fact that I’ve had many people come and say, Oh, can I have a chat bot, please? Normally the contracts, I’m trying to do this, you know, like the financial support team example where, you know, the normal customers need help in a different way, right? We’re not to do that, let us let us in the door. And we will come and help you design that. So I think the most important bit would be get your business, get the business, get the organization bought. And if you’re leading it, you’re gonna be passionate about it, you’re gonna believe in it, that should be infectious. So go in, you’re going to get time in front of the people that you need to help sponsor because that will just make life so much easier.
I think the next bit is definitely like picking off your use cases, really can’t do it all. So pick off a heat each bit at a time, go out whichever the next bit, which is an expert, but then use the terminology now nostoc. And I’m really clear on where I’m trying to take this to what I’m trying to achieve. So, you know, these phases can’t be done all at once. So we said it’s probably going to take five plus years to get to where we’re at, I think we want to get to realistically, it’s probably going to take even longer. But know the phases you’re trying to get through. So if you’re just starting off, right, where do I want to be in 12 months time? And how, what is it going to take to get that bit right? What’s the next bit I need to build the next day I need to build? And then the final bit is far better advice is don’t be shy to tell your customers this is what you’re building for them. I think if I had my time again, I would have been slightly bolder and slightly more.
There with our customers. Initially, you know, we were quite conservative. We were quite sensitive. We know when we first put a mobile chat live library 10,000 customers. Well, I’ll just pause here and just see what they think. We kind of knew it was the right thing, but we weren’t quite sure whether we’ve got it Right. And then it wasn’t until we’d like to roll it out to all of our customers, you know, millions and millions of people have access to it, that we then started to tell our customers it existed. And it wasn’t until actually, you know, we got the first like, proper set of NPS data where we were and actually, this is, you know, customers are really loving that suddenly that we then really began telling them it existed. I think we should have probably been a bit braver and gone a bit earlier, because it’s kind of all interlinked, and so you get more customers who want to use it. Easy to grow, easier to do next steps.
Fantastic. Well, Andy, this has been an absolute pleasure. Thank you so much for joining us. And sharing that story. It’s absolutely fantastic. And thank you, everyone for tuning in. Whether you are tuning in live or on the podcast. Definitely join us tomorrow Cobus grayling of, Vodacom. We talk about all things conversational AI technology. We’ve brought him on the podcast, which is what we’ve talked about in this episode. We’ve talked about the value of data, and what it can be used for and how you can leverage it as a strategic asset. And that’s basically what bought him help you do. And then we have on June the ninth 6pm GMT, we’ve got a really good webinar with deepgram and we’re going to be talking about why big tech doesn’t innovate as fast as you would think and how you might find better performance in niches. And so please do join us for that. I’ll put all the links to that in the show notes. Andy. There’s been absolutely fantastic. Thank you so much for joining us. Nice one.