Maurits
Hello and welcome to Customer Friendship Conversations, the show where we bring you the latest trends, tools and insights into delivering customer experience as it’s meant to be. I’m your host. My name is Maurits Pieper and I work at Dixa and AI is one of the hottest topics around right now. It’s an exciting new field that has the potential to completely revolutionize how we approach customer experience and really do it from the ground up. That’s where today’s guest comes in, Daniel Bunton. And he’s the head of customer support at Cleo AI, and he knows exactly how to create game-changing customer experience with AI. And that’s what we’re going to be discussing today. So welcome to the listeners of today’s session of Customer Friendship Conversations. We have a very special guest here today, but before we get into that, I’d like to introduce myself. My name is Maurits Piefer. I lead partnerships here at Dixa. Now, today’s topic we’re going to be calling. It’s creating game changing customer experiences with AI. Reason why I’m specifically excited about today’s sessions. The guest AI is not just kind of an add on that they’re testing or that they’re kind of new to it is very much part of their core business model. So I’d like to introduce Daniel Bunton from Cleo AI. Cleo is an assistant to essentially personal finance and he leads up the customer support team there. But I’m not going to do the introduction. Daniel, you could introduce yourself, but perhaps also how you ended up with Cleo.
Daniel
Hey, Maurits, great to see you again. Thank you so much for having me. Yeah. So I’m the head of customer support at Cleo AI, which I want to be very clear from the outset here. I am non technical, so I understand AI, but I do not propose to be an expert at all. I’ll be back. Cleo first. Cleo is a financial AI, so it gives financial assistance and advice to millions of people in the US. And yeah, our company mission is to fight for the world’s financial health. So we really want to take financial stress away from people and that is such a major cause of stress. I went to drama school, actually, I studied to become an actor. And like most actors, I became a waiter.
So I found my way into customer support through that journey and then found my way into startups eventually. And part of. Part of what I wanted to do was recreate a kind of fine dining restaurant experience in the digital world. But I am finding that very difficult and it’s very hard.
Maurits
Very cool. Tell me a little bit about, like, you know, when you start the week, when you start the day. Is there anything that particularly gets you excited about jumping out of bed and getting on the job and working with the team?
Daniel
Absolutely. People are what get me excited. So, yeah, being a head of a department, like you’d hope that’s the answer. Helping people develop their careers, pouring into people in the way that people have poured into me in the past, that’s what really gets me excited about starting my job every day.
Maurits
Let’s break into a little bit of the topic on AI. It is such a growing and an expanding topic. Probably lots of the listeners have read or heard so much about AI, but let’s focus just on you. Could you just share some of your initial opinions about AI in general?
Daniel
I’m hugely optimistic. I think that, yeah, AI is going to change our daily experience. I think about 18 months ago, when chat GPT officially went live, the world exploded with about six months of like, high intense curiosity. And I think that has kind of died down over the last twelve months or so, and we’re now experiencing like, okay, what does this actually mean for our day-to-day experience? I really think the sky’s the limit. So, yeah, I’m excited about it. You’d hope. I am working at an AI company, but yeah, I think there’s massive things ahead and I think we’re just scratching at the surface of what we’re going to experience.
Maurits
Have things changed personally? And how you use AI in your personal life, is that something that you’re also excited or opportunistic about?
Daniel
Yeah, definitely. So Cleo actually gives everyone a, like, enterprise-level subscription to chat GPT, so we can go to town on chat GPT and not hit any message caps. Whenever I’m facing a work problem, I will generally go to GPT and go, this is what I would, this is the problem I’m facing. This is what I want to achieve. How and where should I start? And so what I see it as is like a, a brainstorming partner, right? And often it will come back with like, really bad suggestions and I’m like, okay, great. But, like, it still gets, moves my brain forward a little bit and gets the ball rolling other ways in which I’ve used it. I can’t believe I’m admitting this publicly, but I use it to design a gift card for my girlfriend, and she’s now going to find out that I used AI, and so I might be in.
Maurits
Trouble for that fair example I like the variety and examples that you provided. Let’s talk a little bit more about the challenge and things that the listeners can start to apply because I’m sure what you’re about to share is very applicable to what they’re currently seeing. There’s AI and all the benefits that it can provide. But I think one of those initial steps is how do you go about implementing it? Some people don’t take the step, other people are very cautious, and some people just dive right in right away. Any tips or advice or how are you seeing potential users of AI start to implement it?
Daniel
I think first and foremost we should acknowledge that everyone listening to this is going to be coming at it from a very different perspective. So, for example, the experience of someone who works at an AI company is going to be very different to the experience of someone who’s working for an e-commerce company. So I think there are two main options that most of our listeners if I imagine, would be facing, and the choice is whether to build something themselves with AI. Working with a large language model like ChatGPT, you would need massive resource to be able to do that. So access to like developers who can build a pipeline with OpenAI for you or any other LLMs. I imagine the majority of our listeners are going to be in the situation where they’re looking to buy a product that uses AI to help solve some of their problems. So yeah, I think they’re the two main solutions and I imagine our listeners are looking to deflect more conversations. Essentially, they want to provide more instantaneous answers and yeah, just generally provide a better customer experience.
Maurits
Could you expand on, you know, what are some of those challenges that people typically run into when they’re being asked, perhaps by their exec team, perhaps by your competitors, they’re seeing their competitors do it. What are those challenges when it comes to actually implementing it?
Daniel
Yeah, yeah, it’s massive. So I imagine a lot of CEO’s out there, saw an article that Klarna put out. I think I’m remembering off the top of my head here, but I think they were claiming that they were achieving 70% success rate in deflections with AI. So I imagine a lot of Head of CX experiences are like facing that question from the CEO of what are we doing in this space? I think it’s a great question, right, because a. I think you should use that opportunity to ask for the budget to go and do stuff in that space because there’s an appetite. And so often cs and customer experience teams are put down at the bottom of the totem pole. Last team to get investment. So I think there’s a really great opportunity here to provide better outcomes for our users.
I would immediately pivot the conversation, though, to overall success rate, like very wide range success rate, into very specific quality metrics. How satisfied are our users with this experience? If you are not measuring that, then I think it’s very dangerous. I’m not going to name names here, but like, we did one test, we did one test where we tested an AI bot and we just let it run wild for a day and a half and it got trained on our knowledge base. So our FAQ articles basically, and we let it take about a thousand conversations. I personally read through 100 of those conversations and we think it did a good job. 33% of the time it was claiming success 60% of the time. So it’s really hard to measure success here.
And I would urge caution to any CEO that is like a bit gung ho about this and thinks that this is going to revolutionise their team and, like, costs are going to drop overnight. So I think you need to manage that conversation really carefully and focus on success and overall user satisfaction, which is going to lead to better retention, more increased customer engagement. Like all those things that they really care about, you got to keep them.
Maurits
Focused on that stuff in terms of managing expectations. Like, you hear a lot of buzzwords being thrown out of deflection, resolution, satisfaction. How do you navigate those conversations? Because you see those skyrocketing numbers of 70, 80% numbers of deflection. But like, you just mentioned yourself. Well, what about the experience? Like, what about it actually being a successful interaction with AI? Like, how do you manage deflection levels versus resolution levels?
Daniel
So what, like, what I would do in an ideal world, if I had some resource, I would a B test, right? So if I was to implement an AI bot, I would make sure that we kept 50% of our traffic on the old system. And, you know, the percentages are up to you here, whether you go 90, 10 or 50. I just want to monitor retention. So for us, we’re a subscriber-based company, so we want to measure the retention of a user. We also want to monitor customer satisfaction in all of this. So if we’re not measuring and if we don’t see a massive difference between those two customer retention and customer satisfaction between the two sides of the AB test, then I think you’ve got a pretty good case that the AI bot is doing a good job.
Maurits
Sounds like setting expectations is a massive piece of that.
Daniel
Massive key. CEOs are under a lot of pressure to reduce cost, right? Like the economic climate is really hard and so that is going to be really attractive to them and you need to help them through that conversation and be like, look, we could do that. Yes, we might be able to write off a bunch of cost in CS. However, think about the overall business, right? Like if revenue dropped because retention drops, that will also put us in a bad situation. So you want to monitor that as well.
Maurits
I like how we’ve covered how you manage expectations from the exec team to stakeholders like yourself or your colleagues. But of course, then there’s on the actual end customer. So how do you, if you’ve done it personally or if you have specific examples, but at Cleo, how do you empower your customer base with AI? Like in the sense of how do you continue to drive adoption, drive further trust, I would imagine. And you can definitely educate me on this. If you’re trying to bring in new product features that Cleo would be able to do for your end customers, how do you make sure that you have a sufficient level of trust and empowerment in your customer base, that they’ll use your AI for those new features and use cases?
Daniel
I think my current role at my current company, we’re playing in a very different field than I think, most kind of Head of CS out there. So anyone who downloads Cleo has downloaded an app called Cleo AI, right? So they are buying into AI, right from the moment they start engaging with that product. If you’re running an e-commerce site, for example, it’s a different situation. And so when they are having a problem, that might be the first time that you introduce AI to their experience. So when they start reaching out to customer support, they’re suddenly engaging with an AI. Bottom. You have to be prepared that at least 20% of your customer base really hates AI and does not trust it, does not want to be involved in it, and just you’ve got to give them an option to get out of it. So you always want an option which goes, hey, if you’re struggling with this, just click this button and you’ll go to a human. So I would always want to provide that option for them. Then I think you’ve got the other 80%. And this 80- 20 thing, I’m just really pulling out a thin air. This is not, this is not research. Like I say, my company is. Everyone’s bought into AI who downloads the product. So I think with that other 80% they’re willing to try it, they’re willing to like give it a test and you’ve got to impress them in the first 3 seconds, I think, right? Like if they know they’re talking to an AI bot, they’re going to be like, what can this actually do? Like I’ve been through bots before, I’ve been through useless bots, I’ve been through bots that suck.
I think any gain we can make there is going to impress people pretty quickly and they’re going to actually start to change their perspective and like, oh, maybe this bot can help me. But if in the first 2nd, you say something like where’s my order? And the bot says, “I didn’t understand that, could you rephrase it?” Like, if that happens, you’ve lost, right? Like the trust level very quickly. Very quickly. And it’s like, cool. I’m not dealing with something that is advanced here, but if we can get AI to the point where it understands what users want and can perform actions for the user on the backend, that is next level. There’s a lot of dangers around that, which I’m sure we’ll get into.
But yeah, if you can actually talk to an AI bot that A: understands you, and B: can do things for you, amazing. What we’re at right now is bots can regurgitate information. Next step: can they do things for you?
Maurits
My gut feeling you may know better than I do, but that doesn’t seem to be too far away. If not, we’re already kind of touching those actions.
Daniel
Yeah, exactly.
Maurits
I think one of the next pieces I’d like to get your insights on is your product. Your company is performing a lot of very interesting and beneficial use cases for personal finance to end customers. And you want to further develop your product, but you also want to, let’s say, manage how your current customers, future prospects or future customers are adopting their expectations and their attitudes towards AI. Do you think just year over year, the population that is interesting for your company are just going to be more open to having AI do more in their lives?
Daniel
Yeah, I think the population that AI can do something for is going to radically increase over the coming years. I truly believe that. So for us in particular, right, like financial information, we are very constricted by banking systems, right? So everyone’s financial status is controlled by banking systems. Their financial history, their transaction history, which is what the key thing that we use to kind of provide insights for users, it’s all tied into their bank. Every country has a different banking system built on different APIs, built on different tech, built on different levels of like, legacy technology. That’s why neo banks are moving incredibly fast comparatively, because they don’t have entire decades worth of user history on legacy systems that they want to maintain. They can build things from scratch in a brand new way. That’s what’s really limiting our population growth for our product right now. Right. Connecting to banking systems in every country is really different.
Maurits
Are there any things on the roadmap that you can share that you’re excited to come out and start beta testing or just sharing with your customers in early access new features that you think are really going to take your product to another level?
Daniel
We’ve been doing this for like seven years already, so the LLM, but like, explosion which happened 18 months ago, really accelerated our growth. But like, we’re already kind of doing most of the stuff. It definitely lifted our gain though. Like, it made us better. One thing we’re really excited, like, one thing we love doing is helping people save. So, like, we use AI to look at a user’s transaction history and say, okay, this month we predict that you can afford to save $1.78. So if you click yes below, we’ll take that money from your bank account and we’ll put it into a little savings wallet for you. So things like that, ways in which we can like, analyze past spending and predict how much they can actually afford to save that stuff we’re really excited about.
Other things we’re really excited about is helping people improve their credit score. So we have a credit builder card on the market and if people use it, their credit score will go up. And for users in the US, credit score is an impact on your daily life. For example, in the UK, like, credit score comes up, but nowhere near as much as it comes up for Americans. And our chief credit officer has a saying. He says it’s incredibly expensive to be poor. Poor people are paying by far higher interest rates than rich people. It’s so broken. It’s not how it should work. They are not the people who can spare the funds. There’s a genuine cost of living crisis happening right now, right? Like, we have richer people struggling to afford, going to the pub the way that they used to and having a fun night out the way that they used to, whereas at the other end of the spectrum, we have poor people not being able to feed their families. Improving people’s credit score is like the – the one thing that we think could be a massive game changer in the system. So that’s really what we want to do, we want to improve people’s financial life and decrease their stress.
Maurits
That’s probably what I love about your business model, is you’re taking a more cost efficient way to provide personal finance advice and support to those that might not be able to financially get that, which might be to the higher end of those on the salary curve. So if you’re helping them with their credit score, if you’re helping them with savings, with personal financing and doing it in an interactive and personalized way, there’s probably no wonder why your business is booming in the way it is. Could you give me an example of how it could be a funny one, it could be a weird one. How are you providing personalized experience to your end customers?
Daniel
Cleo’s tone of voice is really unique. Essentially, our whole company started because our CEO designed an algorithm, like, to spot when he was about to go below zero, right, about to go into his overdraft, and he rigged it up so that he would receive a text message a few days before that, and the text message would say, hey, idiot, stop spending so much money, you’re going to go below zero. And we built an app around that idea, right? Like, people, like, getting told off by a robot is funny, right? And when it’s accurate, it’s funnier. So, Maurits, I don’t know what your eating habits are like, but one of the first things we do for everyone who downloads the app is roast them, right? So we will analyse their transaction history.
Daniel
And if we see McDonald’s is like a regular occurrence on someone’s thing, we might say to them, listen, Maurits, if you keep doing this with McDonald’s, I’m going to start listing it as a recurring weekly expense. So that kind of like humour and coming at people from a different angle really raises their awareness. And it’s different, right? Anyone can look at a graph. Anyone. People get bored by graphs and they don’t want to look at them and they don’t want to understand their finances. But if you can get roasted by a robot for spending too much money on Amazon, like, honestly, we set up a whole different thing just for Amazon. And we set up a whole stream, which is like, analyze someone’s Amazon spending. The numbers that we see is like wild.
Daniel
So we’ve got users who are like, you know, hey, guess how much you’ve spent at Amazon over the past twelve months. And we show them three numbers which are wildly different. One of them is like three grand one of them is eight grand, one of them is twelve grand. And the users will always pick the lowest one. And no surprise, it’s always the highest one. And so getting people to engage with their finances like that in a way that they. And not hearing it from a person is very important, right? Because a person telling you’re an idiot and should stop spending is very different to a robot telling you. But yeah, finding that line is really difficult. And actually, tone of voice has got us in some trouble in the past. The way in which we prompt the large language models is really important. Right? So, for example, we tell. We tell our large language models that we use, we want you to sound like a kind of big sister with an attitude. So, like, kind of loving, but also not afraid to speak the truth. And we’ll give it to you straight kind of thing. Yep, that has been good and bad. A user said, I’m expecting a baby and I don’t know what to do. And the AI responded saying, wow, congratulations. That’s such great news. Let’s talk about how we can get your finances in order for the baby. And so you’re like, wow, beautiful. That’s incredible. What a great response. That feels emotional. On the other hand, we had a user say, give me a refund right now. And the AI knew what the user’s name was and said, calm your tits, crystal.
Daniel
Which charge do you want to talk about getting refunded for? And so in that situation, we’re like, oh, God, right. And so, for example, in that situation, we go, okay, AI, you are not allowed to say these words. So, yeah, we have a whole list of words that our AI is not allowed to use. Unfortunately, tits is now one of them.
Maurits
Thank you for the example. I think that’s definitely a unique experience, but a good learning opportunity to trend AI. Like, how do you measure the success? Do people give feedback if they don’t like that interaction? If they use it very frequently and that they engage with your product, how do you know which of these product features are gaining traction and which ones are not?
Daniel
Yeah, so we measure, like, daily engagement with chat and daily engagement with other product areas. One thing we really do analyse is NPs. So we’re really into NPs. We look at App Store reviews, we read all this stuff. App Store reviews is where we hear a lot of praise for Cleo’s tone of voice and, like, people love it. Like getting told off by a robot. And then, yeah, for our other products around the AI, that’s we’re also looking at NPs and how much retention we’ve got there.
Maurits
You’ve been with the company now for what, almost four years?
Daniel
Yeah, four years. It was 30 people when I joined and I think we’re 350 now.
Maurits
Then it’s quite the journey. Are there any things that you would have loved to know from the start in terms of how to approach customer service or customer experience, like some significant learnings that you went, man, I wish I would have known this before I started at the job or just at those beginning. Three to six months.
Daniel
Yeah, it’s really hard. I think what you need to succeed in the early days is really different to what you need to succeed as the later you get in the company journey. So, for example, in customer experience in the early days, you want real problem solvers, you want people who are going to go, okay, what is happening for this user? How can we fix this? What’s here? What’s here? What’s here? What’s the data? How can I see this? But when you’re going through serious growth, suddenly at one point that completely changes and you actually don’t want people to leave a pathway too quickly. Right. Once you understand a user journey and how it can be resolved, you kind of want people who can follow a playbook and are happy doing that. I find it’s not the same people. Right.
Daniel
The people you need in a 50-person company for CS is very different to the people you need in a 400-person company for CS. I wish I’d have learned that earlier and found out ways to find out ways to serve both of those types of people right there. Ultimately, they’re equally good and they provide good value, but they’re made for different types of roles.
Maurits
Before we kind of wrap up today’s session, I know like you mentioned at the beginning, there’s probably going to be a wide scope of listeners on today’s session, but any last tips or small pieces of advice you could give to people that are just trying to get into AI just want to make the first step because an exec asks them to look into it because they’re getting beat up by competitors? Like, how does one take that first step to at least make progress in understanding it, potentially adopting it into their respective businesses?
Daniel
I’ll get a little bit technical here. There was a case with Air Canada just recently. Air Canada promised someone a refund when it didn’t actually align with the policy. And the user, like the customer, like, took it for granted. Like the AI bot told it would be fine. And I think the customer screenshot of the conversation, and the courts actually ruled in that customer’s favour, and Air Canada had to refund that user, even though it was against their policy. So what’s really important here is understanding intent. So when a user comes into a conversation, user or customer, they have an intent, they want something. They’re trying to understand something. Using AI at that point, to understand the intent is the best thing we have right now.
To understand that intent, where AI is not so good, is at providing the same answer every time to a user. So I would say you want AI at the top of the funnel to understand the intent of a user. You do not want AI getting involved in telling a user what your refund policy is. You want AI to send a user down a pre-written pathway that clearly outlines the refund policy. So understanding intent, and then if you’re using an advanced product, you want AI to listen to the rest of the conversation to understand if the user has a second problem, so their intent has changed during it.
Daniel
For someone out there who is hearing an executive saying, get into AI, just go do it, do it, be careful and really only use AI at the top of the funnel to put a user in the right flow. Do not let AI run rampant with your users, especially where potential refunds are concerned, and especially when users are really stressed. And always give users an option to get out right and get actually straight to a human. That’s my key advice for someone who’s under pressure to start using AI.
Maurits
I think those are all very fair. And so, just to summarize, the reason why I like today’s session so much, Daniel, is obviously we’ll learn more about Cleo, how you see the impact of AI in the customer experience and customer service space. I like the sections where you outlined, how does one go about actually implementing it? And if you do come across challenges, well then how do you overcome them? Again, like you mentioned, there’s so many potential listeners on this call. So I like the more general approach that you took, because we don’t want to give specific advice just for one industry or one type of company. And then, of course, those final thoughts where we’ve been discussing, how do you get your end customers to engage more with AI? Trust it more.
Maurits
Is it going to go up in terms of acceptance to the public year over? Year over year? So, overall, I’d like to say thank you for joining us today. I think it was really cool that you were able to join and get a better understanding of how you – Cleo – take a good look at customer experience, especially with AI. But then I’d also like to thank today’s listeners on the call of Customer Friendship Conversations. If you’ve enjoyed the show, make sure you’re following us on whichever podcast platform that you like. And if you’re interested to learn more about customer friendship and customer service, then check out dixa.com. But thank you very much for joining, and thanks again, Daniel, for speaking to us.
Daniel
Thanks, Maurits. It’s an absolute pleasure.