Jeroen Das, Product Owner, ABN AMRO bank, gives us an insight into how ABN AMRO is scaling its conversational AI solutions.
Presented by Deepgram
Scaling Conversational AI
ABN AMRO Bank is the market-leading bank in The Netherlands, and Jeroen Das is leading the conversational automation charge. We chat with Jeroen about use cases, channels, business drivers and outcomes, as well as tips and tricks for scaling your conversational AI from nothing to over 1.5 million conversations per year.
00:00 Intro and Welcome to Deepgram
02:22 Introduction to Jereon Das
06:08 Conversational AI POC
11:07 Why create a digital avatar?
13:35 Measuring the success of a proof of concept
15:14 Strategies for getting something into production
19:30 Digital assistants: where to start
25:37 Senior ownership vs grassroots movement
29:45 How to use different types of data
32:21 Key skills teams need to mature
37:18 Misconceptions of senior leadership
41:18 Today’s chatbot use cases at ABN AMRO
48:35 How to educate stakeholders
54:26 Future: digital humans and AI call centres
Yo yo yo, hello hello hello ladies and gentlemen, boys and girls are welcome to vux world. I’m your host as always Kane Simms having a bit of trouble with the old musical intro there but live and kickin here we go. I’d like to give a shout out to deepgram before we do kick off this conversation which is going to be an epic conversation with Jereon Das, who is the product owner, for conversational AI at Amro bank, we’re gonna be talking all about Amro bank. It’s one of the Netherlands largest banks handling 1.5 million conversations per year. And we’re gonna be talking all about conversational AI maturity, as you begin to scale solutions across the enterprise. We’ll get into that in just one moment.
But first shout out to deepgram if you are looking for speech recognition services, to power your voice bots or even use cases beyond voice assistants things like call recordings, things like meeting transcripts, all that kind of stuff. Speech recognition has a huge role to play across the enterprise. So if you are looking for speech recognition solutions, do check out deepgram.com/vuxworld they’ve got immense accuracy, they’ve got incredibly fast speed and they are incredibly cost effective. One of the things that deepgram believe which I also believe as well is that there is no such thing as a generic speech recognition model, every speech recognition model should be tailored for the use case that it’s used for. And that means you’re going to get the best accuracy. If you do that. If you have custom acoustic models and you have retrained speech recognition models, based on your customer, dialect, language, vocabulary and all that kind of stuff, you’re going to increase accuracy, which means you’re feeding more accurate data into your NLU which means better performing assistance and better customer experience. So go to deepgram.com/vuxworld if you want to learn more thats deepgram.com/vuxworld.
Alright, now without further ado, let’s welcome today’s guests. Jereon Das, welcome to VUX world.
Thankyou Kane. And thanks for having me.
No problem. Thank you for joining us with a lot of interesting things happening in the conversational AI space in the Netherlands. So you must be enjoying things over there.
Yeah, definitely. It’s like it’s one big kindergarten in terms of enjoying experimenting and doing new stuff. That’s yeah, that’s what me and my team is really like. So that’s really good. Nice. Nice.
So I first came across you when it was only a few months back when you were speaking at the Chatbot summit in Edinburgh. And it was a European chatbot. I think it was and you were hot. You were on a panel with a bunch of other kinds of banks like Monzo was on there, a couple of others. And what was really resonating, which I think is a topical conversation, certainly for me at the moment was this concept of maturity and developing a mature AI team developing organizational maturity as far as leadership knowledge and education and support and all that kind of stuff is concerned. You so it’s a real sense about that kind of stuff. So I can’t wait to get into that conversation. But first, before we do, please do tell us a little bit about yourself, Jereon and what you do at Amro bank.
Sure, yeah, I’m Jereon. I’m 35 years old. I have worked at ABN AMRO for almost 10 years already. But since 2017 We’re working I work as a product on a conversational AI and that was also back in the days right when we started so we we made a small POC we put it live see how how it works and also our customers are dealing with a chatbot get the first learnings from that and also start learning Okay, does this technology has the potential to grow into something big, but that since 2017, and overtime together with a great team, we grew quite significantly so for my channel POC to the first year built on specific topics over and over again to an enterprise wide conversational AI platform from which we currently operate multiple chatbots for customers as well as for our colleagues. And yeah, we have a great team of over 40 people working on this.
Wow. 40 people, that’s quite a lot. When did you start did you say? when did you start the initiative? When was the PLC?
Late 2017, Yes, second half. So September October, something like that we started
interesting. So you got to 40 people in just less than five years maybe?
Yeah, that’s true. But if you look at the adaptation of the technology, technology and what the need is for digital context or context online it’s quite natural growth as well. We always grew up against the demand side not so much by pushing so that’s quite interesting.
Yeah, interesting. So what is your sort of like your interest in this stuff come from them Where was your what got you interested initially in conversational AI?
Well in general are like really like topics with regards to basically changing? Changing digital offerings we do to our customers is really turning something upside down and changing your organization or changing the way we offer our products or requests. okay with customers, that kind of disruptive technologies always drew my interest and yeah, from basically via via via colleagues who are interested in technology. Basically, my interest grew in that technology as well. And yeah, it was a fascinating opportunity we were raised to actually take, take the position of product owner over this technology. So that was really exciting, of course, and also to evolve. People have the same eagerness in that sense. What I did, so that’s, yeah,
nice. Nice. And so talk to us about this PRC, then what did the POC sort of entail? And what was it that you were trying to learn from it originally, back in the day?
Well, back in the day, the question was really out there is that, okay, we read a lot of what happens on the other side of the large ocean, right? Because of chatbots, or digital assistants, that was really taking place. And what we want to learn is one on one side. Hey, is the technology advanced enough to also deal with Dutch language? That was also a big question mark at the time. But on the other end, are customers okay with it, or can be over it in such a way that customers actually like interacting with a chipboards instead of instead of just calling us or, or well chatting while it didn’t hurt yet at the time, but evil are coming to the office and actually interact with with one of our advisors, and investors? Yeah, we started with a small POC, we’re like some off the shelf components or services we use. We put something together. On subjects we know customers were talking about. We publish online quite rule based, there was not really an MOU behind it or anything. And then we took it from there. We just started seeing what was going on. And what was what was in terms of technology, but also in terms of what the customer wants from us, and how do they like it.
Interesting, how will you? How did you kind of determine whether customers liked it? Because that’s always a sort of, you know, from the one side, the business cares about getting stuff done, or whatever deflecting calls in evident transactions where there might be the customer kind of preference is always a tricky one, you know, your NPS CSAT those various things? Is that what you’re using? Are you looking for something else to try and gauge whether people were actually liking this thing?
Yeah, well, these days, it goes a bit broader. So these days, it’s not only about measuring, well, it’s also about NPS. And that kind of scores rise, but it’s also about Well, I think what is the most one of the core benefits of conversational AI is that basically what you can, what you enable yourself is with your you can basically improve the boat, your boat data drift, and so you basically get real time insights in what’s going on in your boats.
The customers go through your flow completely, you can read the conversations themselves as well. And basically, you can learn from that. Okay, how are our customers experiencing the interactions with Anna? And also in terms of its enable and as a chatbot? For the listeners, by the way, that’s our main digital assistants. Do the customers want it? Are they able to get stuff done via enter or do at the end over a month? Or do they just seem to give up and drop out? All these kinds of metrics we are measuring within the bolts and then that’s basically how we can determine Okay, and it’s good, Joseph, for our customers are not or we need to change something. And there’s actually only improved what to just gather for the next brings basically and, and put it on there. Yeah, and but if you look from the start to it, I know when we started back in the days, and we talked about tondar to auscultation, right? So it was really small, really a small POC small scale and everything. And then you were going through the logs yourself. And then you immediately saw that if you want to do this, right, you should do it big.
Don’t keep it small, don’t go for straightforward technology to use it. Because the customer expectation for a digital assistant or a chatbot is huge. So you just can’t get away with a few simple flows, clickable flows. And because it’s experienced really quick, it’s something that you want to avoid contact with your customer basically, and that you put a simple but in front of it. So what we saw, yes, there is a lot of potential in implementing digital assistance. But you should go back if you want to do it correctly.
Interesting. Interesting. Why Anna? Out of interest? Why is it called Anna?
We had some customers. We did some customer interviews at the start of the journey. So I also put out some questionnaires. And that was that there was a wide range of different avatars, different names, combinations, how we feel in relation to the bank when it’s so so what’s the tone of voice with the bank is also of course an important one because you want to have over your channels. You should implement it. It’s important that Anna has a personality, but it should be relatable to the brand’s you are of course, so there are different aspects we tested on. And in combination with the well. The Avatar, we use our own personality behind it. It’s yeah, it came out to test the most favourable humans for it.
Interesting. Interesting. Was that part of the POC as well, the avatar or did that come later?
No, it was part of the POC as well already. Yeah, they do.
Yeah. Interesting. Interesting. It was kind of a conscious choice, or was that part of the platform that you decided to go with at the time had, you know, an avatar capability? Like was that was the avatar a conscious choice?
Yeah. Yeah. So we actually have a specialist for that. So did you all investigate and well, UX specialists who did all the investigation and the research to get rid of customers and stuff like that come up with it. Even the designs actually show we tested a whole wide range of avatars? So it can be human like a robot like? Well, we’re not even relatable to humans or robots? And yeah, that this, this? Yeah. And these were basically the results. So yeah, it was quite surprising. But people went for a familiar face sort of, so we’re, we’re human faces. Yeah.
This cartoonish, of course, but yeah, yeah. And
do you think you measured or have you done any research as far as the role of having an avatar player versus not having another time or not?
Well, it was quite straightforward already. That’s it. Yeah, I think it’s fair to say that we went for something recognizable also because we didn’t want to sort of to trick the customer or get the customer in confusion if he or she was talking with a realtor or not. So yeah, our avatars are human like, but it’s also pretty cartoony. So it’s pretty clear that it’s a chatbot. And not actually an agent. Talking with them. Yeah, it was quite, it’s quite, it’s quite a fast designer that wants to go for something recognizable?
Yeah. Interesting. So it sounds so you’re beginning, the beginning point sounds very much similar to how most organizations would progress if you begin with a proof of concept. You’ve got some ideas about, you know, is this actually going to do what it says it can do? Are customers going to actually like it? Or that kind of stuff? You deliver the proof of concept. And you met, obviously, you spoke about how you got kind of customer feedback? So was there any particular kind of business metrics that you were looking for? When did you initially launch that POC? Like, what were you using to measure the success of the POC?
Yeah, the success of the POC was really like, Okay, does the technology work? That kind of stuff. But I think from a broader perspective, I mean, where did most corporations start as well, I think that it was all about automating customer contact with the customer. And was basically driven from the inside that a lot of interactions we have with our customers are, well, the conversations might be complex, but the solutions for large are pretty, pretty straightforward. Pretty easy. So yeah, the whole organization started to work for starting to get behind the tables from basically a cost reducing perspective, and the standardizing perspective, also from agents happiness, because the philosophy was also a bit yeah, you can answer the same question 20 times a day? Or do you really want to dive into the more complex stuff, right? People get more MP from the more complex stuff.
So it doesn’t pay to take it from those two angles. But that evolved over the years never really talk about the visioning to 17 to 18, something like that. But along the way, you saw that transforming as long as we knew more, you’re available through more channels, we tried different concepts. And then the whole, the whole vision and the idea also from the organizational, especially from the organizational perspective, transformed as well, because you see more and more value. Identify more and more value along the way. So that’s really, that’s really Yeah, that was really that’s really a nice journey to take stressful but a nice journey.
Interesting. So what were the next steps beyond the PRC? Then how did you because you mentioned that you realize in the PRC that if you want to do it properly, you’ve got to pretty much go big? What was the sort of strategy from the PRC to get something into production? What was that kind of journey like?
Well, I think grades early identifies that customers are not really well from a generic perspective. If the customers hit the Chatbot and start to chat it can. I won’t, I won’t disclose any surprising stuff here but they talk about anything related to the bank, anything. And if you have a very surreal A simple rule based head bolts, then that’s not going to help if you have just a couple of subjects in. And even then, even worse, for some sort of perspective, the customer can jump between subjects as well. So it’s quite a complex solution behind it and a scalable solution behind it to get stuff done.
Also, I’m not sure but it’s quite normal, a bank has a pretty wide audience, basically, everyone needs a bank account. So if you’re a company, if you’re rich, if you’re poor, whoever if you are living somewhere backward, or living in downtown New York, you need a bank account, you need to have a relationship with the bank. So the audience and the target groups of the banks are really wide spread. That also means that on one side, you and if you want to build a virtual assistant, which is at least usable for the main accounts of your audiences, then you have to go really, really wide. And that’s also when he saw, okay, so from a technology perspective, you need to go for the more advanced and new kinds of capabilities.
But also from an organizational perspective, you really need a lot of people from the organization tapping into your technology, or start talking with them. And together with them building your virtual assistants to actually make your future assistants really applicable for all these different target groups that talk about all these different subjects. So that was, that’s quite the that was actually a journey of over. Well, it’s still a journey, but it’s, it’s, it’s already been a journey for over four years now. And yeah, that’s that’s, that’s, that’s quite interesting. And yeah, to give you one insight, I think, I think the biggest mistake you can make as a starting team is to think about what you can build yourself up like, like, if you’re like, like, honestly, you start on the small scale, right? So your world is pretty small, pretty straightforward.
You have a solution, you build this, you have them to stakeholders, you take in their, you take into account what they want, you take it to college customers want and you expense reports, credit, really. But what you really see is that if you’re there is so much knowledge needed to include in an efficient digital assistant for a bank, is that you also need to think about okay, how do you get all those people along from all those different product groups, you have to talk about mortgages, investments, your bank accounts, name them, insurances, name them, and you have it, that it becomes a really widespread, widespread initiative. And also, when I think that because the banks in general are so large, it also means that the level of knowledge is also available in the organization. Right? So are we as developers, are we as data engineers? Are we as UX experts in having a conversation with a customer? Or those people in the contact centers or in the branches, etc? And? Yeah, a lot of stuff to consider?
Absolutely, yeah. So if people are listening to this thinking, go big Gore, kinda like, not necessarily go big or go home, but the learners and the PRC was, was, it’s big, you know, you’ve got lots of different types of customers with lots of different varying needs. And, you know, a lot of chat bots, that became a one use case, you could argue that potentially they disappoint a lot of people. Because if you’re only covering for 20% of your use cases, then 80%, potentially the customers are being left out. But at the same time thinking about, Okay, well, we’ve tried this technology, it’s keyword based, or rules based, rather, we don’t have any natural language understanding at the same time, one, we need to get our head around natural language understanding capabilities, and then to, we also need to go across the whole organization, and begin to then compile content, format, content, rewrite content, get it into accessible place, get it all signed off, across all those different divisions that you mentioned, there might seem a little bit daunting. What advice would you give others who are hearing this and thinking that sounds massive? Like where do you start with that?
Well, I think you can. I think you can start, I mean, during POC during the first year, we start with a single team, right? Because what you should identify for yourself are the stakeholders within the organization, maybe even the people you’re talking to you to you’re trying to sell your you’re trying to sell your digital assistant to you need to get them on your side as well. And I think in my experience, if you look to the stakeholders you talk to that are rarely their contact center, maybe people from the branches whatsoever.
They have so much knowledge on how to deal with the actual customer and how to deal or to talk with the customer. Basically, get them on board, also get the knowledge and maybe even get them working on the conversations of your digital assistant themselves. Because the experts in your organization who know How to Deal with a customer or your people be it telephones behind the desk in the branches and that kind of stuff. So my advice would also be that if you build a digital assistant first place to support your, your contact center, for example, is that also to get those people in and loop with them to the information you get in your chat both? In terms of okay, what is the customer needs? What are they asking and anything? I start developing the dialogues together with them, and also guide them and train them and how to do it and stuff like that. And grow together with them. So don’t do it alone?
Absolutely. How would you define maturity, as far as kind of conversational AI is concerned? You started with a very small team. Similarly in the PRC, you’ve grown quite organically, as you’ve covered more content, more use cases, across different areas of the bank, there’s now 40 People working within that team. What are some of the elements that you think constitute maturity? As far as this is concerned? What are some of the things that you do now that you think are kind of representative of a mature team that perhaps weren’t there before?
Um, well, what do I think of representing a mature team? First of all, I think it’s very important. One of the main constraints, I think, in the current market is that there is a lot of knowledge on how to do this, right? But the knowledge is with a very select group of people.
If everyone wants to do with a very, very, very database, or upscale to team size, it’s very hard to get those people out of the markets. And while I think it’s, yeah, you also need people out to market you need specialists in but you also need specialists really from reading your organization itself. So first of all, I think you should get a healthy mix of people who know everything about the bank and about the customer, together with really the people who know everything about technology and how to deal with it. And you sort of have to get them together and also have to get them to work together. And that can be a major, major challenge, because well, not speaking, what I sometimes hear in the market as well put some, some back Enders, data engineers and conversational designers in one room and get them to speak the same language and get the and also that we know from each other, okay?
What can be the value from each other about what we each bring to the table that can be very complex. So if you have a combined team, which from those different specialists who can actually who actually understand each other and pay time to each other, and actually can suggest maybe even improvements towards each other, because they understand each other’s worlds, I think that are some really good signals that you’re really having with your team. So it’s all about a mix of knowledge and a mix of experience bringing it together and getting it to work together.
Interesting, it’s really, it’s really interested in that concept of having people that are part of the business versus, or as well as people that know about the technology, because there’s been times in the past, in the distant past where I’ve been, I’ve been working on things without really the business context, or you know, you’ll, you’ll hit a, if you’re starting from scratch, you will, you won’t be able to start. And then at some point, you’ll hit a wall where you’re not really sure whether what you’re actually seeing as part of the conversation is even true. And then you’ve got them a long time before you can get it in front of someone for them to have a look at it. And then depending on how big the system is, it can take quite a while for people to actually get through various stages of the experience to get to the point that you need to pay attention to and then how you document feedback and deliver feedback and stuff. So I can see how having a team that’s made up of people with the knowledge of the business versus as well as the knowledge of the technology, working collaboratively together can really expedite kind of timescales, I’m sure.
Yeah, absolutely. And it’s essential to be on the same boat right about not only from the team itself, but also from an organizational perspective. I mean, if your ecommerce departments or your contact center departments or any other maybe some kind of product team or department is not on board doesn’t share the purpose and doesn’t share the goals you try to achieve it then it becomes very hard to, to, to build a sustainable system for your organization. And I think that is also that is also key that there is a certain understanding that everyone needs to everyone needs to contribute to the digital assistance in order to make it work. And yeah, indeed, like someone like a, like a copywriter from ecommerce reads customer feedback differently than someone from the contact center or data engineer shuts off and that’s Yeah, and you need to get all those insights you need to get them together and sort of come up with a coherent plan in how to improve digital assistance. So yeah, it’s quite important that it’s ain’t sure we might have been. Yeah,
yeah. And that is that it is that driven from a kind of real senior level or more of a grassroots movement I’ve been in organizations before where the collaboration between teams was, it’s almost like a bargaining is such where this team is over here, they’re doing their thing. And they’re really busy. And under pressure, this team’s over here doing their thing, they’re really busy under pressure. And then the transformation team as it as it was then called just kind of in the middle, trying to buy a bit of time from here trying to get a bit of bad time from there and trying to kind of do things in and around business as usual, without it being like a dedicated project team where people can come out of the business to work on something. And so I’ve worked in both instances where, in that case, it’s kind of grassroots driven. It’s all about building relationships internally and trying to get support internally. But also, I’ve been on the reverse of it, which has actually been driven by the senior stakeholders and board members and whatnot, who kinda like all the VPs, who have been like, this team and that team need to work together on this project for this period, because this is really important. In your experience, what have you found works more effective? Or is it a combination
of those absolutely combinations. If you only have your grassroots movements, what do you call these, but you don’t have support for senior management, or they don’t see how it contributes to the vision they have for customers or for the organization, it’s not going to fly. And because it is, let’s be honest, it can be quite an expensive project to deploy as well, right? It’s not something you can do out of change money from your annual budgets or something. So be really dedicated to it, and understand why we are doing why we’re doing such things. But on the other hand, yeah, without collaboration in teams,you basically run into the problems you just mentioned, right?
So teams with different specialists have different goals. People might do it on the site instead of dedicated and this kind of stuff. Yeah, and then you get the they don’t get the potential out of the technology you want to get from it. So yeah, and especially the customer because the customer is rightfully quite easy in declining a specific channel, if it doesn’t help them directly. So yeah, it is nice to come from both ways. And I think well, the strong point of conversational AI, in that sense, is that the customer talks with you in plain language, and you can capture that language and you can analyze it a lot and also on a larger scale. So you know, very specifically where you need to improve and where you need to change. And I think that’s that’s, that’s, that’s for compared to other digital capabilities companies regularly offer online, I think that’s one of the main strength points, customers can talk to you in plain language, what they think is good, what they think is bad, or you can easily subtract it from the conversation, and use this very valuable input for your assistance, or maybe also the other capabilities.
Interesting. One of the things that’s also a kind of, I suppose a signifier of maturity, both from the kinda like, technology capability perspective, and also from the culture, teamwork perspective, is taking that spoken language that you mentioned, being able to analyze that at scale, as you mentioned, but also taking relevant kind of insights, and being able to offer that to other parts of the business. So there’s a really good example of Commonwealth, which is a Canadian telco. I use this story all the time, because it’s really interesting, where they had customers call in their IVR. And the bot that was in the IVR is intended just to raise support tickets, if your modem or router goes down, or you’ve lost your Internet, whatever, raise a support ticket, someone will get in touch with you later. But what I’ve noticed is that from this very specific place in Canada, one town, we’re getting loads of reports of outages.
And that because they had the setup internally was able to be flagged to the Support Division to the business, then actually found out that the internet was down in a part of the in a certain part of town. So the Insight coming from the bot was able to inform the business of an issue that it didn’t know about, then went and resolved. So I’m just wondering whether that is the situation at Amro bank and how else do you use that data over and above to improve models? If anywhere?
Yeah, well. Yeah, so maybe not really directly linked to the example you put on but for example, it can also mean that you communicate in a lot of different ways with your, with your customer, right? You can even send letters, we have a website, you have the app, you have your main contact channels, that kind of stuff. And sometimes, it can be the case that sometimes you see that you get a request from a specific place on the web of the app a lot of times about specific products, which is relatively quiet. In our example, we had three years.
So, for example, our customers were asking a lot. In the Netherlands we have a payment system, which is called ideal and a lot of questions came in about it. Can I retrieve it? I’m not sure what the English words are but can I retrieve the money I just sent my idea myself. And of course, as a product manager or as a copywriter, you write a lot of stuff down about specific products or about specific topics which product actually can do or which Yeah, which product can actually mean for a customer, but not really what you can’t right. But that doesn’t mean that there are no questions about it, but at the customer side. So in that sense, sort of observing and seeing what is going on in the customer’s mind in terms of okay, what questions do they have about a product can also help you basically to optimize content or information elsewhere? And I think that is, that’s, that’s a generic example, how you can use it.
Interesting. So, you know, for those kinds of teams that have the resources to be able to do things like that regularly, which you should do, absolutely, if you’ve got any trouble living, this should be a thing that you do routinely. But you’ve got a lot of different skills that are needed to do this, your PRC, if you look at the two teams that you have now versus the PRC, I imagine you’ve got roles that exist in your team currently, that probably might not have even been thought about in the PLC. If you’re working with lots of different divisions across the bank, and they’ve got content and data in different places, you need somebody in your team to be able to bring that data together, create relationships between it and make it accessible to be called by a chatbot. And all that kind of stuff, you’ve got to do quite like the data science requirement there.
We’ve got someone who understands NLU systems, if the POC was kind of like rule based and the recent edition has natural language understanding, you need someone to be able to know what to look for. In those transcripts, you need someone to be able to maintain the integrity of your NLU models, as you kind of retrain it for specific use cases. So wondering if you can talk about that stuff to give us the you know, role for all your team makers. But some of the key skills that teams as they begin to mature will absolutely need in their team that they might not have thought about when they were just beginning.
Yeah, I think, look, I think I think machine learning and other applications in general that are quite generic applications, right? I mean, you could do a whole range of stuff with it. And I think just building the experience, experience and gaining the experience how to utilize it right in terms of conversation in terms of implementing the actual flows, notes, basically, and LinkedIn notes and make jumps in that cetera like that. And that’s a professional itself. And I don’t think I don’t think it’s something you learn on your own, the university basically, or not, necessarily.
So that’s really something you have to gain by experience. And I really think that’s something to find those people and authorized to find those people who are interested to learn this. That is, I think, the key question there. People pick it up quite rapidly, I noticed if you deal with the right, if you have the right systems in place, of course. So I think that’s quite important. But what about another major insight loss is that designing conversations is actually professional itself. So actually, you can’t you can’t pull people from the contact center, drop them behind the laptop and having them write conversations on one side and other other sides, get a few content specialists who write for 1015 years already write content for a website, or, or whatever, or, and do the same with them. It’s really professional in itself, because the dynamic of conversation is completely different than than it is if you just publish content on the website or in an email, or whatsoever and understanding that dynamic and also realizing what the problem is behind the questions the customer also, you can offer a solution more rapidly than maybe then instead of asking a few additional questions in your flow as well, that is really key and getting debts into getting the knowledge from those two sides, getting it combined.
So know how to write content, know how to have a conversation with the customer, combined with synergy that really are combining that knowledge that really got together to add that to our realization that, yeah, a conversation designer is a professional itself. And it’s not something it’s not something you can do on the site or something like that. So yeah, that was quite a lengthy journey, right? Because, yeah, you have to gain the knowledge as well how to do it as well, by yourself also with some external support, but that’s something which has to grow within your organization. And if you just mentioned, I just mentioned, for the people working on lands while I’ve been with counsel, the conversation designers themselves in that. So we have quite a bunch of conversation designers in the bank, which you really have to advocate to bring on the level. You want to have them to make them really effective. really valuable for Anna in this case. And I think that’s the major insight in terms of creating new professions. I think the conversation designer is something really something new Yeah,
Definitely, definitely. Yeah. It’s, it’s really interesting because it’s, it’s almost got, it’s gotten to the point where it’s becoming a recognised role. I imagine now, if you, you know, now compared to 2017, certainly, it’s more recognised now as a profession. But it’s really interesting conversation design, because people come into conversation design from all kinds of different backgrounds, which is like, you know, yes, some people might come from customer service, or content writing or whatever it might be. But some people come from all kinds of different places, you know, it’s absolutely crazy, the varied backgrounds that people have in it, because it’s a real, it’s a real discipline, you know, like, being able to, you know, craft dialogue, understand the nature and the flaws of conversations, twists and turns, you know, grounded and error recovery. And that kind of stuff is, it’s not trivial, necessarily, but it’s something that I think a lot of people overlook.
What do you think are some of the things that people overlook when it comes to senior leadership support? So in order to get from a POC to production, you need to be at a demonstrate that it’s got potential to meet business value, obviously, in order to go from production for for one thing, and to scale that across the whole organization, all departments for mortgages, to insurance to lawns, the whole kind of bank accounts, the whole nine yards. Your it requires, definitely, as you mentioned, senior leadership plus some maybe some grassroots stuff, but looking back now, what what would you say are some of the things some of their like, maybe it’s common misconceptions, maybes that leadership have around this technology, and ways in which you kind of approached educating and gathering kind of buy in from senior leaders. But it was first, before we get into How was what some of the events, not just that Amro bank, but generally, from your observations? What are some of the kinds of misconceptions that you think senior leadership has around this stuff? Sometimes?
Yeah, I think if you’re, if you’re not that closely involved in working with this technology, like, Look, if you’re coming from an age where you have a lot of software as a service kind of approaches, right, you deploy a piece of software in your organization and his works. And while people don’t realize good enough, what you really have to bring to their attention is that well, the analogy used within the organization is that we’re building these two assistants, you’re basically raising, raising a kid. So it’s not something the kid is there, and then the kid knows everything, it’s really about the match, the amount of time and effort it takes actually for the team to understand the customer really well.
So to come with the right dialogues, it is also that it’s not that you deploy in No, you you build a nice interface, you build a nice avatar, you call it that and you put it live, and they’re added technology does it all so or the AI does it all, basically, because it’s AI, right. So it’s really intelligent design for intelligence, it knows how to deal with customers, right? So bringing them along that whole journey from starting at scratch and to basically expanding dialogue for dialogue topic for topic, scale, to scale, to upscale with to a boat, which on the one hand, is still learning and on the other hand, is actually already delivering value. Because that’s also one of the aspects right with conversational AI, it can only really learn once it’s live to customers. That’s the big way out of this. That’s the way to go.
If you want to skew it on the lower side, that understanding and realizing that it’s not a software package you implement basically, that’s a dynamic, which is, which takes time for people to grasp. And even if they say, Yeah, I do understand that you need to put a lot of data and a lot of conversation in if you’re long lost before your assistance is, it is mature. I still experienced that. Even if people are saying it in there. Yeah, unconsciousness, they still approach the whole process and the name and then mixto as a software package you’re implementing? Yeah, yeah. That really takes time.
Yeah, interesting. It is really interesting. Because I mean, I’ve been involved in lots of conversations in the past, where the expectation is that it’s kind of this black box that you just turn on. And then it kind of works, because everyone uses Amazon, Alexa and google assistant and stuff like that. And it kind of just talks about how it works, doesn’t it? So why would this be any different? You know, it’s really, it’s a totally different mental model to get your head around, isn’t it?
Yeah, no, absolutely. Absolutely. And, yeah, of course, the kind of companies, they’re delivering great stuff, but they don’t necessarily show how many people are involved to get it as good as it is. Yeah. And understanding that. I mean, it’s not a silver package that has generic buttons. It really needs to be tailor made for your specific customer. You can’t just take out the data model on our sidewalk so I don’t know about it or something or maybe Vattenfall, a larger energy company. He implemented it there and it works fine. No, because it’s a different company in a different sector, with maybe the same customers, but who have different problems and are approaching it differently. So you need to adjust your model on that. So you basically yeah, it’s for a lot of companies, it’s really your building your system from scratch. And you basically mean that you basically buy the algorithm, then you’re, that’s step one, or step zero, basically.
Interesting. What? So now then, beyond the POC in production, four years down the line, big team support and you know, there’s the support and buy-in from senior leadership, you’ve got people involved from all over the different parts of the organization. What kind of use cases do you cover? Now, you mentioned the beginning, you kind of covered a small number of use cases, it kind of advanced to be a lot more broader. Are you? Is it kind of like a question and answer? Do you do any kind of transactional kind of stuff? Like what kind of use cases is it on today?
Well, we are also available with relevant content for basically our customers. So businesses, where our wealth management customers are regular retail customers. So basically, we cover it all. Even expense, so we have a sort of multilingual thing that is really nice. That’s what I really like, that it works as well.
But also in terms of Yeah, and a professor at some services, so she can change your address or check your address one before we send something to you. Yeah, there are multiple, there are multiple things Anna can do. So it’s really service oriented. It’s information oriented. And, and I can also base on you she also targets customers with specific messages. So we sort of have a sort of practice concept in production as well. For example, Yeah, well, the nice case was one of the first cases. It was actually one of the nice cases as well is that basically, and notifies customers, three minutes before their banking card expires, that their banking cards are going to expire. And we’d enter the customers to check if their address is still correct.
So we don’t set the banking card to the wrong address, once we send it to a new bank, that’s kind of stuff you know, so we actually tackle a problem before something comes up. So before we send the card to the wrong address, and the customer is at the shopping mall, trying to to pay for the goods, basically, and the car doesn’t work. That’s kind of problems, that kind of frictions, you can take all with sort of a proactive strategy in that. And we have multiple cases on that, which is really nice. Yeah. And it’s all about accessibility. Right? So multilingual, I think it’s really a part of being accessible for all your customers. But also things in terms of things in terms of the way you display yourself, or only the chat screen in the button. Do you like web interfaces? What kind of concepts? Yeah, we’re doing all kinds of stuff in that sense.
Schoolboy error there, I had myself on mute. Yeah, the proactive stuff is really interesting, because it’s where a lot of value is. And a lot of organization, I think that’s a good sign of maturity, as well as that you’re able to do productive things. Because if you think about the amount of resources that it would take to do that thing that you just explained there, which is to reach out to customers tell them that the car is going to expire, but once on the way, confirm that the address is the correct address. And then okay, to then send a new card, I imagine how much time and effort it would take to send that as a letter in the post, or have someone actually make a phone call to confirm that. Or even, you know, the time said, I’ve lost my train of thought, but yeah, basically, those two things would be very expensive to do. Whereas once you’ve built it in the chat bot, it exists forever. And there’s just nothing but value afterwards.
Exactly. And it’s also like building a digital assistant, there’s also building sort of a relation of trust with your customers, right? So if you can do this with your digital assistant, these kinds of cases, and the next time the customer comes to the website, because I don’t know what goes on.
They want a mortgage or there’s a specific problem, they want to talk about someone and they see Anna popping up who helps her greatly the last time even practically, then people also tend more to click on it and start talking with Anna instead of picking up the phone or asking for an employee directly. Right. So it’s also a huge trust in Gartner I think but also from a broader concept right. As the bank I think as the banking sector as a whole we are sort of also in the digital transformation right? So I think the same happens in the UK, small local branches are closing everything moving to the app or to the websites. Yeah. And then one of the big gains of having a bank in every town and then every shopping street basically He was always you would have a conversation sometimes with your customer, right?
Which is not only about specific problems or about specific products, where they were coming from, but also about, well, how are you doing in general? holiday plans? Or what are you saving for? Oh, you might think about moving places, that kind of stuff. So to bring that kind of the touch points and bring that kind of relationship, we also need to transfer it to our digital domains, right? And I think in that sense, it’s not only conversational AI, but I think conversational AI is an important component to this, but actually the customer comes in, and that’s basically and I can also determine, Okay, who’s the right person for you to help on?
And can she pick up those small hints that the customer wants more than just what they’re talking about? And connecting them to the right people within the bank? I think that’s, that’s, that’s gonna be a major driver of conversational AI, and in the banking sector as a whole? I think. So in that sense, I think, I think it’s really valuable to pursue this. Pursue personalization strategies online. But I think conversational AI is an essential part of this. And not only for banks, actually, I think for everyone, for all digital companies.
No, yeah, definitely, definitely. I mean, as as companies begin to mature, you start to see, I mean, Capital One, I think, is a really good example in the US where their digital assistant II, nor is it in the chat channel, it’s on the app, it’s in the call center, it’s even as a shortcut on it in a web browser that can facilitate secure payments. It’s also an email. So the proactive stuff you were talking about there from Anna it kind of, it’s all email based, so you get an email about your card, whoever it might be. And so it’s really this kind of like, on true omni channel presence. And when you get to, not even that level, but everything that you’ve been explained so far, around Anna being proactive as far as anticipating customer needs, being facilitative, in terms of getting the customer to where it needs to where they need to be understanding the needs of the customer to be able to build and, you know, create services and models that are able to deal with that kind of customer queries, helping to relieve tension and stress from customer service agents and lighten the load.
So they can focus on more complex needs, having an accessibility angle as well, you know, there’s all of this stuff isn’t kinda like, tactical, you know, stuff, it’s, this is all real strategic value organizationally. And so the AI front end, in order to get a good AI for an end, first of all, we’ve talked about a bit of it, which is that you need to have your back end in order, you know, API availability for line of business systems delivering the right kind of shape and format and accessible for the for the for the bot. And so there’s a whole bunch of stuff that goes into it in order for it to get to that point where it is a strategic asset, I suppose. But I think a lot of organizations still, when they’re immature, believe that a chatbot is a point solution. And it’s a tactical implementation and nothing beyond that. So how do you go about educating stakeholders as well as as well as other people within the organization about the fact that this thing is a real strategic advantage and not simply a point solution for your specific customer service problem?
Oh, that’s, that’s. Yeah, that’s a long, long, long process. Yeah, I mean, it’s not so tough, in terms of the fact that they’re hard to get. But it’s, it’s the point is, I think, this technology can be transformational for the way we offer digital services to our customers. And that means that you have to basically go through a lot of facts set in stone, sort of, I wonder about basic principles of how the organization works. So before you have that kind of thinking, you are able, I mean, everyone can tell a lot of people my stories, right? A lot of people can make nice, flashy PowerPoint sheets with all kinds of facts, why it’s transformational, why there’s a benefit and that kind of stuff.
But actually to build a relationship with your organization and take them along step by step. Okay. For example, right now we have a POC on payments objects, we think it’s gonna be variable for mortgages, as well deal the step as hey, we have a mature digital assistant in the chat channel. But why is it driving the app as a whole, for example, that whole process there is like, there are like multiple steps in between them Take, take it slow. Don’t try to hit your homeruns straight away, right? So define for yourself the steps and each time define a new step in your vision, what you have what you think if it’s like driving your app, if it’s jet only, or if it’s, I don’t know what you see now, for Esports or for telephony, is it more and more coming up in the market? That kind of technology, take them along step by step proof.
It’s a small use case, okay, this can actually work well that’s, well that’s the value compared to the current system or the current interface. You’re currently offering them and all operational sites are there where you form the episode that’s more on the operational side where you really show the value step by step where it’s grown to. And also, to take them along, like, on a strategical note, like, hey, if your contact moves digitally, so first, that also means on a strategic level, people have to first realize, Okay, you go from analogue to digital context as well, if you are in a digital transformation, right? And if your stakeholders are there, okay, what can what can it mean more? Is it only about automating your contacts? Basically? Or is it right? Or does it for example, mean that in terms of zoom, doing your services online, frictionless or effortless? Or whatever term you want to give to it? Can conversation play a role there? And why can it play a role in there to to start using, it’s a process of of yours and also good understandable because things are working things are delivering value. So you’re not going to stop that? Because someone with a funny name walks into your room and claims conversational AI is the new way to go.
Right. Yeah, it’s a lengthy process. But yeah, and never give up, I guess. So. Be aware, be conscious the first 20 times or 10 times you probably get to know. And then yeah, step by step.
Interesting. Yeah. But as you said, you know, if it’s, if you’re constantly focused on business value, then it you know, it’s an actual gradual process, isn’t it whereby you demonstrate some value, there’s then there is, you know, the next step is identified, you deliver that bit, and you prove that this use case also delivers value. And so then it’s like, okay, well, if this is delivering value, then maybe this will as well. And so over time, you know, intuitively you kind of build the capabilities and build the knowledge and maturity.
Yeah. And as well, I think, also experiment with a lot of stuff, because don’t be afraid to kill specific concepts. So if something just doesn’t work, yeah, well then accept the fact very quickly, it just doesn’t work this way. Because sometimes you see people just keep pushing it through, well, maybe if you improve this or that, or whatever. But customers are just not adopting it. I think that’s quite open for Yeah, be open for failure, literally accept there’s effects and make sure you have a relationship with your stakeholders, which do not depend on that single experiment. Right, then you get a whole different kind of talk. I think that’s important as well.
Interesting. Well, that’s
what I wanted to say, sorry. Losing my mind for a bit of what I want to say. But also be very open to ideas from your organization. So if you don’t think you have you know it, all right. So talk with the contact center agent installed with copywriters from the website and the app, talk with people who, who are there on the front not working in the fields you’re working on, and try to set up something together and also to work concepts are based on their ideas and see if it works, right. I mean, people who build something, they love it. I mean, you love your kids as well, automatically, right? So if you get something to production from one of your stakeholders, and they see it working, it’s also easier to adopt it and to build on that right. You are open to all kinds of colleagues with ideas. Yeah,
absolutely fantastic. That’s really useful. Not necessarily thinking about Amro bank. You don’t need to give anyone any secrets. But in terms of you, when you observe things like technology developments, things like different use cases that you see existing, and you think about the future of AI powered customer experiences, I call it what are some of the things that are kind of exciting, new about, you know, you can see Google with the with the lambda and the large language models, you mentioned voice channels becoming pretty popular, you know, there’s a whole bunch of different things kind of emerging, what are some of the things that are exciting, new about the future of this technology and these use cases?
Well, I think my being broke might be an interesting development. I think, what I start to hear people talking about it’s not I haven’t seen that concrete yet, but what might be very interesting, if if the technology you use, so forget the fact about this assistance, but the technology use and how you apply it, it becomes part of a wider range of analytical and NLU driven products, which are sort of managing and driving your contact center as a whole. So, sort of intelligent contact center kind of concept, that kind of stuff. I think that’s, that’s, that’s, that’s a very interesting field to explore. So you take it further from basically you take you you approach it from a 360 angle, so your customer relationship right so not only focused on call on your, on your customers, but also actively support your agents, real time with specific insights or automate certain thoughts for BIM or gain insights for your workforce management.
So they know how to plan capacity, better, all that kind of difficult to apply AI over your contact center. White, for example, can be a very interesting topic. What I see in the first experiments, it’s not what we are doing, but I saw so first experiments in the healthcare field especially, is also I think, the concept of digital humans. I mean, it’s not mature yet, but like real time 3d avatars like, like we’re talking that you have sort of like an in screen as a third party or something like that. I think I’ve seen some digital humans with quite some decent emotional intelligence as well. So who can respond to your facial expressions? Where were I? Yeah, there might be a lot of potential in that as well. Maybe in the longer term, I think it’s not there yet. But I mean, most of the interaction, most of the interaction you have with the person is like, yeah, it’s physical. So it’s based on facial expressions, I’m sure of English terms for that. But if you can mimic that near real life, through a digital human or something, I think the feeling of a customer or feeling listened to and being looked at, you get a way better connection than just firing off some subtext lines to each other. Right. So I think from a few perspectives, from a customer relationship perspective, that can be a big, that could be a big gainer in the long term.
Yeah, yeah, absolutely. Yeah. Yeah, really interested in there, I think you can see some examples of the kinds of AI first contact centers becoming available. You know, if you looked at last year, when Microsoft announced its call center capability, alongside Dynamics 365, I think there were some like 15 features, or 14 features, and about nine or 10 of them were all natural language processing capabilities, from analytics to bot automations to all sorts. And then Google’s collaboration with uj. I don’t know if you’re familiar with you, jet.
But your jet has the best omni channel capabilities that I think I’ve seen. And we’ve been heavily embedded with Google, cci in terms of its integration with dialogue, flow, and whatnot. So the use cases that that offers are pretty interesting. But I think that if you look at it because the call center is basically all conversations, you can have, you know, the ultimate route in the IVR, that takes you to the right type of team, right, five skill sets, for the right use cases, automation in the IVR itself. And then when voice biometrics are there as well, which we don’t see much of at the moment. And the same thing can be said in chat as well, this exact same process can work in chat. And then you’ve got conversational intelligence running in the conversations themselves, something like a symbol of AI or an observation aspect that extracts, you know, from the live conversations that agents have into conversational intelligence from that. There’s things around agent assist, where you can be you know, ultimate in transactions are given edges, next best options and that kind of stuff. And so, the potential for NLP across all these customer service channels is absolutely huge. You know, it’s frightening.
Yeah. But that’s, that’s Yeah, yeah, I definitely, definitely see the potential to but on the other side, it’s not that strange. I mean, if you can understand the language of a person. I mean, yeah, that’s from an automated solution perspective, you could do a whole, you could, you could do a whole lot to interpret it in so many ways, and set out so many actions afterwards. I personally believe services outside the bank, I personally believe this is really, this can be a really transformational kind of technology, which is also easily deployable over different channels, right? I mean, not meaning only digital channels, but also maybe analogue channels or, yeah. So in that sense, I think, I think it can be really transformational. From the easy perspective, if you’ve seen 10 different services in your app, that you need to offer 10 different interfaces. And if you do the right conversation, you only have one and a customer just never searches for it, gets no screen, type something in the order, and this process right away. So only from that kind of perspective. I think it’s really a potential to do something big in the upcoming years. are already doing it actually. So
yeah, absolutely. Yeah. And the digital humans for those tuning in. If you’re interested in the digital human side of things, check out the podcast we did a couple of weeks back with Rob Cunningham, who was an innovation manager at ner, which is a railway service in the UK. They tried one in the train station and got some pretty good feedback, to be honest, the limiting factor, we’ve actually run out of production.
I mean, the vision is that you would have a digital human everywhere throughout the train station, so that at any point in time, you need any help at all. You’ve always got somebody close. It’s just the cost. Basically, it’s quite expensive to get a real good high fidelity, digital human that has all of the things you mentioned. You know, being able to recognise when someone’s gazing at you instead of activating in a way that I’ve been able to cut, you know, marry together the dialogue and the actions and all this kind of stuff. It gets quite complicated. But yeah, potential is definitely there. Yep. Cool. Well, you’re on this absolutely fantastic. Thank you so much for joining us. Really, really appreciate your time. It’s been a nice one. Thank you all for tuning in. Tomorrow we will be speaking to Patricia, who was the CEO of private AI. And we were talking all about privacy in AI and natural language processing and how private AI is helping further the cause, which I think is a really important topic as well. So we’ll give that some airtime. Shout out to a fake or Hollander who said, well said you’re in Dallas, I won’t bring up the whole comics. It’s quite large, but he’s quoted your phrase of transformational technology, strategic for an organization and not just a point solution. It’s a process of yours demonstrating the value iteratively don’t be afraid to experiment, be open to failure and be very open to ideas from your organization. Couldn’t have summarized the conversation better myself. Thank you Jereon. Appreciate that. And thank you, everyone, for joining us.
Yeah, thank you. Cheers.