Understanding conversations – [FULL INTERVIEW]

0:01
So hi, everyone. I’m here with Johnny Drexler today who’s the head of business development for record shore. And record sure a contact and document compliance and, you know, analytics analytics company, really for, you know, a lot of regulated industries. So Johnny, thanks very much for joining me today really appreciate it.

0:20
Ya know, great, great to be here, good to get to see.

0:24
So, so obviously, you guys spend a lot of time. So suppose with speech analytics, document analytics, those kind of things. And I’d be good to sort of start off just by just just listing around, I suppose, how you’ve seen that evolve over the last sort of like, you know, 510 years or so? Because it’s been a sort of burgeoning kind of market really?

0:41
Yeah, sure. Look, I mean, I think, and the the reality is that some form of contact centre and document analytics have been around for 2020 20 or 30 years to some degree. And it started really with the kind of the birth of kind of large contact centre environments where you wanted to get a better sense of what’s going on across the space rather than randomly just dipping into the occasional conversation as as sort of indicative of what’s going on in the business. And so, really, it’s it’s, it’s been about the growth of a lot of sort of transcription technologies that allow you to then assess what is being said within that contact centre environment. And, and being able to look, look at larger data sets to help you understand key trends and issues. And I guess, historically, what we’ve seen and utilised is kind of transcription and kind of keyword searching type technologies in that in that contact centre environment where you’re kind of going, have they mentioned a certain product has a specific risk word been said? Or is a specific sort of angry word been said by a customer? But the reality then of those tool sets is they’re quite unreliable, I guess, right. So some insight, you can get some good word clouds that show you what your customers are sort of talking about the most and which words and phrases are appearing most often. But from a compliance perspective, in particular, they can be quite rife with false positives, and then can be quite hard to configure because in a regulated environment is complex conversations. And the quality assurance that you’re doing is quite complex. And so just looking for the word guarantee can be a little bit misleading, if there’s lots of not guarantees that’s being said alongside it, for example. And so the real real evolution has come with AI and machine learning, right, then the third, the third technology or industrial revolution that we’ve all seen, as being able to move beyond poor transcription and keyword analysis to really assessing the context of conversations, how things are said how they’ve been responded to, and being able to do a much more sophisticated assessment of what’s going on. And record your ultimately has built that to solve for the problem of doing small sample assessments of what’s going on in a business in the hope that you’re going to find a needle in a haystack, as well as a more systemic issue. And the ability to say, we can now do human levels of assessment with machines across 100% of your data, to really point you at the highest risk and write that that trend analysis.

3:49
I remember back to the early days of call recording. And we were sort of blown away the fact that if you had a dispute and recall, you could actually get like hear the call back to actually resolve the dispute. And just think about how far that’s come these days to say just like transcription, and now to almost like interpretation, I think is sort of where we seem to be getting now.

4:07
Yeah, no, no, 100% I mean, and you can’t underplay the value of just recording, I mean, a lot a lot, then of what record show is extended into is the ability to capture other channels. Just having an authoritative record is a much better position to be in then then, you know, in a he said, versus she said type environment. And then if you can build analytics across all channels and start to link data becomes a really kind of end to end view of your business and the customer journey.

4:39
And we were chatting a little bit before, I suppose contextual understanding of the conversation and even in terms of there’s real time versus almost like near real time and having to having to sort of get to get to the end of the conversation about the context and coming back. I mean, how, how, how much of an evolution is there around that and then the importance of it

5:00
Yeah, so it’s the real, the real time outcome is, is I think the Nirvana for this space, or it’s the is the next leap that we’re going to make in this space, it’s the ability to say, as the call is going on, really accurately, something is something has gone off course, or the customer is not understanding what you’re saying to them, or that or you haven’t followed the process in the way that we would expect to. And we’re starting to eke into that world, we’re starting there. But there’s still a ways to go. Because to your point, really, what your analytics model needs to be able to do is to understand the context of the conversation, to assess whether you’ve done and said the right thing, because everyone’s ambition now is to not live in a very structured and scripted scripted environment, because it’s just a poor customer experience, you want to be able to be a bit more agile, especially in more sophisticated areas like mortgages and wealth management and pensions, where it’s kind of an advice sales process. So where we’re at currently is that within a near real time context, if you give the conversation a chance to run its course, you can then do extremely accurate assessments of that conversation. In real time, you’re sort of taking 32nd bursts of audio data, and transcribing and assessing it in that small snippet of data that you have. Which means you’re limited in the way that you can flag a real time prompt or intervention. And so what it’s good at, I would suggest is saying things like, just slow down, you’re keen over your customer and your or tonally, there’s a suggestion of dissatisfaction going on here. What it’s going to struggle with is to really assess things like vulnerability, right, which have a lot of parameters that sit around them to determine whether a customer is distressed, or in financial difficulty you need, you need more of the conversation to assess that stuff,

7:16
it might be multiple sentences within within a paragraph that and determine whether they’re actually they’re vulnerable or not, it

7:22
might be multiple data points in a one hour conversation, you know that over the course of an hour, you’re able to assess a specific risk point, but on their own, they mean very little. And so again, the real time can be can be useful, or it could prompt a supervisor to kind of go, you might want to dial into this because there’s some indication that this could go in the wrong direction, or a customer can be unsatisfied. But from a real sort of conduct and automated assurance perspective, it’s still a very long way from being able to say, we’re going to assess this and give you comfort that the conversation all the things that needed to happen in this conversation did. And so that’s that’s the journey we’re on at the minute from near real time into starting to look at real time as a potential sort of assurance tool.

8:20
And do you think like things like vulnerability and also like flags for vulnerability, and using, I suppose what’s being said, or what’s being documented for sort of flag potential vulnerability, I suppose is, is where we’ll get to

8:33
100%. I mean, it’s, it’s, it’s the most probably one of the first three things any bank or building society or insurance company is, is asking for at the minute, because it’s such a hot topic, the economic environment is only going to drive more vulnerable customers specifically around financial hardship and difficulty in the current environment. And it is such a focal point for the regulator. So and everyone is looking at how can you better assess and determine that stuff. And there’s some, there’s some elements of vulnerability that are more straightforward to assess. And there are some that are much more subjective. And they’re much less overt, because people don’t want to openly discuss their vulnerabilities in certain instances, right. And so again, being able to assess the entire context of a conversation to make those assessments is a lot easier. And machine learning, of course, learns through just feeding it examples of data and saying this, this is what vulnerability looks and sounds like an instance and then it can identify that moving forwards. And so the more of your that data you’re able to feed in and start to train machines and models on the better they will get at being able to kind of identify that stuff. But what we’ve what again What we see most effective in that space is to say, let’s help you ensure that when you do have a vulnerable customer, you are following due process and you’re doing the right things by that customer within your given environment or framework. And then over time, what you will be able to better do as you train on that data is to identify vulnerability further upstream. But again, long way from being able to kind of do that at a level of confidence that is going to kind of give a bank real assurance that you know, that that is the platform to find vulnerable customers.

10:37
And what about these shows? Have you got false positives, but less was more concerning ticking, the vulnerable or the financially vulnerable spaces, false negatives, as well. So you said, you know, your your false II, like determining that they’re not vulnerable? Or they’re not? They’re not in financial difficulties, when in fact, they are? I mean, what’s, what’s the best kind of approach for that? Yeah,

10:57
I mean, I think the bait, you know, on it for exactly the reasons that we’ve just discussed, the this, this type of tooling shouldn’t be used as a silver bullet, right? This is not, this is not the platform to identify all of your vulnerable customers, it is a platform to give you a better chance of spotting potentially vulnerable customers. So it’s about providing indicators, rather than assessments, that the outcome being delivered is the right one. So really, it’s about augmenting your human processes. It’s about leveraging the tool to do the assessment where it can be relied upon. But it’s more than about augmenting humans to say, if you’re going to point your humans or anything, point them here, because there’s indication of risk. And we’re going to show you the part of the conversation that’s indicating that risk to go and manually review, rather than this type of tooling being used to determine a suitable outcome because it understands enough about the called variable to do that, so becomes

12:06
like the almost like the the indicators to help guide the conversation, rather than the automated flying of the compensation. If you say to me,

12:15
yeah, it’s kind of a bit of both, right? I mean, there’s kind of more everyday stuff, you know, the right disclosures been read out in the right context, as the customer responded in the way we wanted it to, or them to, and then there’s the more subjective elements of it that are just informed. And it’s, it’s about just moving the bar forward, really, because what do you do at the minute, you probably have a random sampling approach. And there’s probably something that’s informing that you split it across all your advisors, you focus a bit more on the newer advisors, there’s certain parameters that will drive what falls into the sample, we’re just giving you a much more informed risk based sample because we’re doing 100% assessment of all your conversations, and building an automated rule framework to say, if this, then, you know, higher risk.

13:14
Yeah, so as part of your, your risk weighted assessment, and where to look first, right. And, and one of the big themes has been over the last few years, as we’ll be in almost like this end to end customer journey, kind of monitoring as well, which has always been a bit problematic in terms of the key, the old days of call recording, to go back and find all the calls and all the linkages and those kinds of things. And I suppose that kind of links in terms of, you know, the the analytics or the contact analytics in terms of like, that’s all context as well, because the context might not be just being in the next sentence with the next paragraph, even the same conversation. Yeah, could be related to what happened, you know, in a previous conversation, I would think

13:47
100%. So look at those more complex regulated journeys, in particular, mortgages, pensions, Wealth Management, debt collection is another great, great example of that, where there are, there are multiple customer touch points before either a recommendation is made, or a sale is complete, or, or a resolution is achieved complaints is another good example, right? And so, before you even start stitching that data together, just being able to ingest the data, but supplemented with the metadata that allows you to search and navigate your records more effectively, is a huge benefit for our clients, right? Because the minute they delve into kind of all often quite old or legacy contact centre platforms, and they can’t even find the call or the type of call that they wanted to go in and do that human review on in the first place. So they just waste a lot of time looking for information. And what what we’re able to do is to kind of pull out the calls the metadata, supplemented with data from other line of business systems because maybe you don’t have the customer reference number or the call type in your contact centre platform, but you do have it in your CRM platform. And then we’re also able to use our analytics to assess calls and go, what type of call was this, what personal details sit within this call, so we can quickly identify who the customer was all that type of stuff. And just being able to do that drives a lot of insight and efficiency, because you’ve just got more data about your calls before you even get to the clever analytics. And then once you’re able to do that, with your calls, you’re able to, you’re able to kind of supplement that with other channels, right. So in a COVID, environment, video conferencing has shot up, right, so everyone’s on teams, they’re on Zoom, they’re on WebEx, let’s make sure we can do the same thing with the audio from those channels, we actually do a lot of face to face, recording and branch and build environments, that we’re able to also then bake into the customer journey, then you’ve got things like chat, and I am. And again, what record show is focused on doing is being able to say it doesn’t matter, the channel will either record it where it’s not, or ingest it where it is. And then we’ll will unify all of that data in a single umbrella in a single repository. Because of the way that we ingest and analyse that data, we can understand who is the customer, all an advisor certain coals, all the team certain coals, and then you can look at assessing different groups of calls. Your point now I can look at a customer journey over the course of a collections and recoveries process that took nine or 10 months where a mortgage died, all that took three interactions over six weeks to complete, link that data and then say, I’m not just going to assess the individual, the individual interaction really quickly, so that I can go and intervene and that risk point that I’ve seen, but I can I can I can assess the entire customer journey to take, did we do all the things that we should be doing as optimally as we should be doing them? And is there any sort of journey or process interventions we might want to wait make, because we’re seeing a lot of regular challenge around the way that a certain process be run by an individual if you want to do targeted feedback to an individual, or across like an entire department, right?

17:42
And I suppose so that the hindsight is always 2020, isn’t it? And that’s all we need to say was like, so you look at something that’s gone wrong here. But then it’s like, Well, what was the genesis of that that happened? Six months ago. And it might actually be that it didn’t look right in everything looked fine until that point. But now you actually find what actually was the thing that six months ago that that’s what the that’s what happened? And then you need to pull out well, then what how many of those other events happened across the rest of the base to then go back and see if you can actually prevent something happening? Exactly. But you go back to then go forward sort of thing? Yeah, exactly. Right. What about omni channel, you talked about different types of channels that are sort of developing, we’re sort of gradually getting into this omni channel world, especially with COVID. I mean, chats now more popular. I would say we’re all we’re all on teams now. You know, I mean, how how, how does that impact I suppose what would have been speech analytics is now seen as becoming contact analytics, but there has to be knitted together. And is that tricky? Or?

18:39
Yeah, yeah. I mean, look, I mean, I think it’s getting easier because the world, the world, and banks think about an ecosystem rather than a platform now, right. So they are either updating or purchasing platforms that should be easy to integrate into API and to suck information out of and somewhere else. And therefore, it becomes easier for us to kind of unify datasets, then when you get to the analytics, it’s quite interesting, because the way that I speak to in a video conference or in person is even quite different to the way that I speak to on the phone. Because there’s something I can see you nodding and understanding what I’m saying, or I can see you kind of your face is completely uninterested in what I’m saying versus on a phone, you get a lot more sort of vocal sort of check ins and assessment. And then on chat, you can kind of go even more shorthand, but you don’t have the challenge with voices always Can I get an accurate transcript, because a lot of my analytics is gonna come off the words and the language that was said, whereas with chat, you don’t have the worry of is the transcription accurate because people have written what they say to the communication style just differs a little bit so Are the voice analytics models that we have are slightly adapted but very versatile, whether it’s face to face, video telephony or chat, to be able to thematically assess the topics and themes that are being discussed, and then allow you to build the rules that kind of go, did the process occur in the way that we wanted it to? Or is there any kind of language that gives us cause for concern. And then, as you say, any customer journey, really, it’s not just about voice, it’s not just about chat, you’ve then got email, you’ve then got documentation, you’ve then got sort of the customer case file that sits alongside it. And so what we’ve developed as these ecosystems have developed is, okay, we’ve actually tackled voice, which is probably the hardest bit to do, because it’s the most unstructured data. How can we point those models that chat and I am and sort of instant messenger. And then we’ve built a complementary product, which is all about document Analytics, which uses some of the smarts from voice, but it’s actually about just assessing document types information within the documents, allowing you to build similar types of rulesets. But it can also suck in emails, attachments, to emails and all the bits that sit around that customer sales process, to then give you that, as you say, you know, Omni channel, doesn’t matter what format it’s in or where it’s come from, we can do the same assessment on it across that customer journey.

21:38
I mean, I think the the pandemic’s been quite interesting, because there’s definitely been a push towards digital. And so chat as an example, tools is become quite popular. But it’s, you know, there’s definitely been always been a bit of a bias to say, Well, look, we need to talk with someone, right. And we’re now getting away from that a little bit, saying, well, actually, I want to do stuff on stuff, stuff, stuff on email, I want to do stuff on chat. But I think you’re right, in terms of like, the way we interact with the channel will actually change the outcome as well. I mean, do you think that the models have to be how much do you think models differently than the models have to be? And what can you say there’s what’s written, there’s also what’s not written as well. And then there’s nonverbal kind of communication as well. That also happens.

22:21
Yeah, so so it? Yeah, I mean, there is there is some, some nuance to it, right? So again, an email isn’t a conversation, right? An email is a, an explicit sort of instruction or a desire, it’s not a back and forth that you can kind of assess in the moment. I mean, it becomes that over time, but it’s not, it’s not there to do, chat is, is live and it’s kind of a bit more, you get a bit more of a sense of where the person is at and what their motivation is, and what outcome they’re seeking to achieve as a whole. And, and then those other channels, you just, you’re actually just getting more and more information as you move towards those kind of voice and video and face to face channels. And so the level of assessment you’re able to do becomes a bit more sophisticated. So in all of them, you can assess the language, what was said and done, and what was explicitly said, really, as you move into more chat, and then vocal interactions, you can start to look at things like tone and pace. And what was the intonation when they said this word? Yeah. But even in those channels, you’re somewhat limited. When you get into tonal analysis, it’s a really grey area, right? You can probably say, is that person having a positive or negative experience? And but actually, there’s some really interesting research around whether you’re really really excited, or really, really angry, for example, tonally, if you took the language out of it, the indicators are quite similar. And being over overlaying that on to language can help you understand things like dis dissatisfaction, and sort of engagement. And but there’s still there’s still, there’s still more false positives associated with that, versus just assessing the language, what exactly was said, Where was it said, and how was it said in the conversation and again, from, from a compliance and assurance perspective, in particular, that is what you want to focus on. And then you can move more into kind of customer experience and sales training, supplementing some of those other data.

24:38
And I suppose you’ve got the written stuff that you can sort of rely on and sort of documented right then that’s quite and that’s, that’s good from a compliance point of view. Then you’ve got the nonverbal stuff. That’s is that there’s actually quite a lot of information. I mean, do you think it’s going to go that direction? I mean, we’re chatting on a video call. I’m thinking actually, if you even meet in person, think about someone like micro expressions people have and this is how We, we might we’re not even aware of it, but his hands interact, right? And and some of those things we might not be aware of, but you think we’re gonna, we think will eventually evolve there for video calls and live live calls? I

25:11
suppose. So. So what what record show is really focused on is the voice right? What is being said in the conversation and how there’s people doing some really cool stuff around micro special expressions and tells and using that in kind of fraud environments and being able to, you know, is someone lying? Are they telling the truth? Are they happy? Are they sad, we can do that with the language, you can do that with the tone to some degree. And you can do that with kind of video assessment and picture assessment of some people’s facial features as well. So it’s all there. And I think ultimately, it will all operate in a single ecosystem where you can do your turn on all the bits you want for any given type of sort of assessment. And, and there’s, there’s someone focusing on on every aspect of the customer journey from that perspective. Absolutely.

26:04
I suppose we have to wind ourselves back from like, the year 2030. Now, yeah. But it’s,

26:11
but it’s all it’s all. Yeah, you know, all all directionally. There’s some really kind of clever stuff going on around all aspects of that customer interaction.

26:21
But even I suppose what what we have today, I suppose was just, I suppose is, you know, Text Analytics, voice analytics, those those kind of things, do you think? How should we think about the right context have the right channel at the right time versus when you need that extra information? So if you’re talking with a human, you actually, you can actually get some of that information. And it’s useful in terms of the the outcomes, right, or the sales or the complaint, those kind of things, but another another areas, it’s probably not important and chats, okay,

26:49
I have 100%. I mean, you know, all all the general insurers want to push as many people as digital as possible, right? It’s kind of, it’s a bit more transactional, it’s a little less advice based, and their their cost model benefits much, much better from from digital interactions. But and that’s, again, a big part of the journey that we help people with, right? If people are calling into your contact centre. With queries that could be solved online, let’s help you identify those at scale, and understand what you can do to signpost, those digital channels more effectively. But again, the reality becomes, especially in financial services, especially when it comes to kind of the more complex end of it. Mortgages, pension insurance, I would argue as well, debt collection. In particular, from a bank’s perspective. People want to hear from an expert, they want comfort that they’re doing the right thing and an automated chatbot that sometimes gives you weird answers or unhelpful answers to the question. Housing doesn’t do that, for me when I’m about to, you know, invest in a mortgage, which is probably the best investment I’ll make in my life, and one human assurance that I’m doing and saying the right thing. So a lot of banks call it moments of truth or key, key customer interaction points where the human contact will never come out of the equation, I don’t think we’re a long, long way from that being the case. And then it’s just about understanding what your customers want, where they sit in different sort of journeys, and how you offer them the best platform to be able to do that. And then, again, you learn a lot from your human interactions about what customers really want and what questions they ask and what response gives them the best outcome. So you’re able to again, utilise that data, once it’s digitised to inform, well, what should my chatbot look and feel like? Because this is what people are asking in the contact centre? And if I want them to have that question answered digitally, I need to be able to replicate that kind of q&a. Right. And so one informs the other and allows you to kind of get to that digital strategy or that kind of channel transformation strategy. Effectively, yeah,

29:20
I do like the idea of almost like, how do you assess a customer journey and make sure every single point on the customer journey you can assess, right? Well, actually, you do need to speak to someone now because there’s information here that we don’t have, or that you need to get just really to make sure right, so you have to make sure that that’s that’s the right thing. Because even even from a tax point, we might not be sure up till now. But but the you know, but these other journeys, I mean, they can just go through automatically and it’s just a query, just just go through and don’t don’t push them to person.

29:51
Exactly. And it’s the it’s the bank, having a regulatory obligation to push the customer to a human interaction because they don’t have the right information, or they, they need to have some level of engagement to ensure they’re doing the right thing. And then it’s giving the customer the option to talk to someone because they’re feeling uncomfortable or out of their depth, or they’re not getting the answer that they want. And that that’s that the balancing act that everyone needs to tread the line or that the minute because everyone is trying to drive digital, everyone expects younger and younger generations to want a digital and fast and quick experience. But that doesn’t mean it’s the best experience, right? In the in every scenario?

30:41
Well, Johnny is super interesting, I think it’s clearly a certain view this sort of evolving, and you just think about call recordings become standard speech analytics is quickly becoming a pretty much a standard in certain call centres. You just wonder if this sort of like analytic stuff is as well. So just, you know, in sort of, like, you know, two, three years time, I would think so it’s fascinating really is,

31:01
oh, yeah, look, thank you. I mean, we’re having a, we’re having a great time building it and delivering it. I mean, it, I think in the last few years, it has become the more of the norm. So some of the largest banks in the UK and insurance companies use this wall to wall across their contact centre and branch environment to do their conduct assessment. And now, even the regulator, it is able to sit up and say, the technology is out there to assess 100% of your customer interactions. And it’s affordable. So you don’t really have an excuse not to do it. Now they’re not, they’re not mandating it. But if you’re in hot water, or there’s suggestions of MIS selling or bad customer practice, they’re very quickly going to start asking why you’re not using this type of technology. And that’s where you really see the kind of the mindset change within the industry in the organisation and where it’s all going.

32:01
Yeah. Well, Jeremy, thanks so much. Really, really, really appreciate it. It’s been been great to chat with you. It’s exciting stuff.

32:08
Now, likewise. Appreciate it, Chris. Great chat.