Conversational AI and Customer Service
DCX Podcast #5
Hello, and welcome to the DCX podcast where I interview leaders in the customer experience space about how digital is changing the landscape and how you can leverage these changes for success in your business.
Today, I'm excited to be talking with Ted Mico. Ted is the co-founder and CEO of Thankful.ai. an AI customer service SaaS solution for customers and support agents. Founded in 2019, Thankful already powers services for brands who lead the world in customer experience, including Crate and Barrel, Ralph Lauren, Bombas, and on UnTuckit, among many others.
7 Key takeaways from the discussion:
Conversational AI is a subset of general customer service AI
Each one of us will spend 43 days dealing with customer care throughout our lifetime
Creating a conversational AI solution for e-commerce is very difficult
AI is not yet at a level of understanding that knows when to engage and when not to engage
Customers of conversational AI tools have a basic misunderstanding of the capabilities of current market offerings
Customer service experience is mostly edge cases
Automation and AI are not a panacea for businesses - there’s still a long way to go.
Ted, welcome. Thank you for joining.
Thank you. I'm honored to be called a leader in my space and that's fantastic.
Today I want to talk to you about conversational AI and about customer experience. I'm certainly interested in understanding how you got into this space. So maybe we start with the basics you know, for people who are unfamiliar with the term conversational AI, how would you define it?
Well, let me define general customer service AI and then, conversational AI, I guess is this is a subset of that. It's a channel for that. I mean, AI in general can be so, so many things to so many people. And it is now seen as a value addendum to any company. For Thankful AI just means it's the brains that govern all non-human communication between brands and customers across all written channels. So we use deep learning models to power our natural language understanding. So we basically understand what the customer wants and then we use machine learning as well to provide a better experience and be able to mold the replies to reply, let's say in a more personalized, individual way.
Maybe a little bit about yourself, where you grew up, studied, and your path. You know, how you actually got here. I'd love to hear the story behind what made you decide to bring Thankful.ai to life.
Yeah, it's a crooked tale. I grew up on a small island off the coast of France, called England. And I started my so-called professional life as a writer in England, as a music journalist for one of the, I guess it was the equivalent of Rolling Stone in Europe called the Melody Maker. It's a very, it was an esteemed publication, I was their Features Editor.
And then when I moved to America, my first startup, I guess, was working for Dave Goldberg, who tragically died like six years passed six years ago at Launch, and then sort of had like that was a fantastic experience.
Didn't that start in CD ROMs. Was that launch?
I joined after the CD like as it was, moving online as it was sort of trying to replicate that immersive music experience in the digital space. It was definitely before it's time to before broadband was pervasive, taught me a lot and taught me like that having a great idea. But being too early was not you know, it's not ideal. So, you know, after that, I did a whole bunch of digital media jobs. You know, I was part of a four man team that launched Beats with Jimmy Iovine. I was the COO of a computer vision company called MariaDB, which was up-ending advertising as we know, you know, an in-between sort of written books and you know, worked for the likes of Boeing and the stones and Radiohead, etc.
So it's a very checkered it's a very checkered past, not your standard computer engineer. You know, I did I did a little bit of that not your standard pop, I mean, if it seems completely random, it looks more like a Jackson Pollock, paint splotter rather you know, a well work career linear path. You know, I didn't wake up at the age of 22 and think customer service software, but it's my future. I guess. Yeah. 22 I barely woke up at all.
But there's, but two common threads, I guess. Honestly. The first thing is that you know, whether it was digital media or advertising or disrupting customer service, it seemed that it's in my genes to shake things shake up the status quo, and to be at the forefront of, you know, finding an industry that needs CPR and applying it. And then I guess the second thing is that, you know, whether it's writing or marketing or advertising, you know, or customer experience, all of which I've had a lot of experience with, the key to success is a guess the ability to tell a unique and compelling story. And in an attention economy, like storytelling is obviously the most, it's the most highly valued commodity. So I guess those are, you know, if you were picking threads through what looks like total chaos, those would be the ones
Well, I don't know anyone who's had a linear experience in life. So I totally appreciate the path you've taken and so tell me a little bit more about the background of how that got you to Thankful.ai.
Yeah, this is the story that watered down story is that it was born out of frustration, you know? Like, the more accurate story was that bad customer service from a number of different parties drove me from my usual state of like tranquil human being in love with the universe and being in the flow to a homicidal maniac that threw my phone across the room, I have hardwood floors and exploded which was pretty much an apt metaphor for my brain at the time. I, unfortunately, had already used up all my Apple customer care, my Apple Care. So it the culmination of bad service cost me like off the top cost me $1,000 in the new phone.
But I just could not understand how technology had enhanced and enriched the customer experience so much, up until I hit the buy button. And then, and then optimized it and make it better for me. And then as soon as I hit the buy button, it was almost exactly the opposite. It was like it was being used as a barrier to stop me getting where I needed to go. And I just, it was baffling to me how you know like that the same brains that were involved with sort of the the on the acquisition side, the same sort of firepower and mental agility and ability to solve complex problems. None of that was being applied to the retention side.
And I felt, you know, with a, with a lethal cocktail of ignorance and arrogance, I guess, I felt with my co founder, Evan that we were the ones we were going to be the ones that solve this problem. And I will tell you that if I had known how difficult the problem really was to solve, we've picked email to begin with because it was the most popular channel and because we were sort of, Thankful was born out of our frustration with customer service. So we went at this from the the customer's point of view with what makes us different to everybody else. We wanted to solve the problems for us.
I read a stat for this before but I've read I read a stat in like I think it was Time and it was copied in Forbes or Fortune. Which said that in our average lifetime, each one of us will spend 43 days dealing with customer care, oh, over six weeks of my life somewhere burning and roiling in brimstone and hell. I just like it just like like so the idea of morals, you know, didn't appeal to me. So I got I kind of figured you know, if there was a if there was a moral imperative anywhere it was to give everybody at least three weeks of their life back.
So, so that that was kind of and we call the company Thankful because that is exactly the opposite word that most people would associate with customer care. If we were being accurate, we probably should have called the company hateful. But because that's again, how people walk into a customer experience, they've had so many bad experiences, especially with technology that they think you know, like it's going to be bad. And most of the time they're their expectations are met in that.
And I am it seemingly I am not alone in that I just made maybe my my patience was tried just because I had this confluence of bad things happen at the same time. So it was more exaggerated. But everybody that I spoke to when we were thinking about starting this company had exactly the same experience.
And weirdly enough, and this is sort of an odd thing. A bad technology experience made me feel like a failure. For some reason. It was like I couldn't I couldn't short circuit the system. There was no star you know, star zero there was no like I didn't know I didn't have the keys to working through a chat like I felt like somehow I did this other people had the roadmap and I didn't. And it turns out there is no roadmap. It was just bad technology. But I felt like, I kept my failure secret. Because I didn't really want anybody else to know that I was incapable of circumventing let's say a bad chatbot and it turns out, you know, everybody has everybody said the same bad experience.
Yeah, there's a there's a very large contingent of people today who are pushing for, myself included, more digital experience, more self serve, for customers so that they can manage their challenges or experience 24/7 from wherever they are. And there's a vision for where these experiences will go, where it can be much like a human agent. It understands me. It knows where I've been. It's, it's empathetic, and it gives me the right answer.
Right. It understand first of all, it understands me and secondly it delivers the right answer. Exactly. Yeah, yeah. So.
So where are we in that process? And where do you think like, how far are we from that type of experience? And is it even is it even possible to get to that place?
Yes, the short answer is yes, it is. I guess in the if you look at the customer service, technology spectrum, you have the dreams of every customer and every brand is that you'll suddenly come up with Jarvis and you know like Iron Man's assistant and it will be able to understand you and say ‘Have a nice day’ and ‘the weather is going to be so and so’ and provide you with the perfect answer that you'd never have thought of. Are we there? No. Will we get there? Maybe maybe not like I'm not sure. That is a long long long way away from the kind of Neanderthal chatbot experience where like one grunt for yes, two grunts for no, that we usually get now. In the in these evolutionary ladder that we aren't we are we at Thankful at least are far closer to Jarvis than we are to you know, the primordial ooze of a you know, we're not clubbing people of the head the woolly mammoth bone, put it that way.
One thing I will say is that just because chatbots have been a toxic wasteland that have served no one because they just don't understand the customer and it feels robotic or you feel like you're being slung a very cheap piece of tech. And the experience is governed by cost rather than the desire to actually help me. But I don't think you can necessarily, to answer your question. I don't think you can use past experiences to dictate future outcomes. You know, I think you have to be at least optimistic that technology can solve one of these problems, if there is a willingness to do so. And one of the the issues I guess, that I had as a customer was that it felt like the needs of the business that was providing the customer service platform were prioritized over my needs. So like the needs of the platform was, get something done cheaply and quickly and be able to integrate quickly etc. And the needs of me as a customer is that I'm understood, that that the reply is accurate, that you know the hopefully that isn't the sort of excetera in our personalized to me because that, you know, all of those things were sort of ignored, whether deliberately or through the limitations of the technology to begin with. In the end, it didn't matter that the outcome was the same which was a poor experience.
I think it started with the idea that customers could do more. The companies wanted to do more, either they were building something in-house or they were sold a solution and it AI natural language or natural language processing is still in, relatively in its infancy. Setting up the infrastructure to understand and deliver through a conversation, if a customer says this, I say this, what's the next step? It's a lot more difficult than it’s being sold as.
Oh, absolutely. And like I said, If I had known exactly how difficult it was, I may not have chosen to do it. I really I seriously thought that I would have an MVP that we could produce some you know, some classic classifier models, etc. We would have an MVP in like nine months because, you know, Evan is the fantastic engineer. We had we had great engineering had like we had great vision, etc.
Three years later, you know, we barely had a product that like you know, it was it was so much more difficult than we thought. Just understanding the nuance of what a customer really wants, not what necessarily what they say is a real feat in itself. And then you know, just the ability then to you know, finding the answers is one thing but really understanding when somebody says, like ‘I need to, I need to return these for a size larger,’ they don't mean a return or ‘hey, you know, if if, if my stuff doesn't come by next Thursday, you know, like I'm gonna cancel,’ the last thing you want and what you get with a lot of chat bots is here's how to cancel your order. Like it but you want I'm not I'm not dissing on those people.
That's really tough stuff. You need more real deep learning models you know, you need to invest three years to actually do them because we wanted to start in email we had to because in email you have to get them you can't grope toward an answer using some bad digital IVR, you really have to understand the customer because a back and forth and email over like you know over hours it's not a good customer experience.
So tell me how that works. Tell me how that works then with you guys with the so email you're saying is your is your focus?
Every written channel is our focus. We, our focus is that we actually don't think that brands or businesses shouldn't necessarily dictate the terms of engagement at all that they should be wherever the customer wants them to be. So and obviously, it's for us, that's easier to do than everybody else because we started with the most difficult thing to begin with. And so you know, a chat our version of chat using both buttons and freeform text, whatever is so much, you know is much easier than somebody that's trying to migrate from a sort of a logic tree to forming the kind of you know, NLU or natural language understanding and systems that we already had. So, I guess that customer-first focus that we had, as challenging as it was to begin with, and it was a let's just say it was a dubious business decision to start with. Turns out to have a you know, it's had some great consequences down the line.
Are there better parts of a journey right now that are suited to AI?
Yeah, that's a great question. So I guess that the mantra for us and should be for everybody is kind of let technology do what technology does best so your humans can do what humans do best. There are things that there were repetitive tasks that AI can handle much better than a human because they can handle policy decision making much quicker. They can get to things much quicker, they can be more accurate, more consistent. They can handle, AI can handle surges in demand. You know.
I read recently a report that says that, but the average in E commerce especially the average time for consumption is like between eight and nine o'clock at night, when most customers service people aren't even around right. AI is perfectly equipped to deal with about 40 or 50 different things. So there are things that AI does very well and things that but AI can again, not only understand the customer well, but play its part in making sure that customer journey and chat is much much faster that you're presenting the right button you're not presenting just a digital IVR experience which is just dreadful.
It's really understanding what I've done before. If I've only have one order, and it hasn't gone out yet, what are the what are the likelihood of that I want to track the order. Well, it hasn't gone out yet. So there are all sorts of machine learning that you can apply that puts even the selection in the chat in a more predictable form. And then obviously, AI can also look at your service history. It can see like oh it the fifth doesn't have goldfish memory. It has a really good memory. It can see oh, this is the fifth time you've asked about this. It should be you know, it can check your LTV from the CRM and see how you know how much of a discount do we now have to give you to keep you it can measure your sentiment?
The key, I think, really is to have a level of understanding that means that you know when to engage and when not to engage. Whatever AI you choose you everybody needs to make certain it like buyer beware that it's smart enough to know when when to engage and when to leave it for human beings to engage. I think that's the most important.
So when you're talking with customers, what what are they looking for? What are they trying to solve when they come to you?
That's a really great question. So so when we first started talking to customers, this is sort of vaguely interesting, I guess. When we first started talking to customers like four years ago, like we were building this thing. We asked the customer we asked our brands and they pretty much all said the same thing, which can you build a better deflection tool? We need to stop customers getting to our very expensive agents, because it's killing us. And we thought about that hard, and then we decided to ignore that advice and because that wasn't why we started the company, we started the company to fix it for us. So we went you know, we said like, well, we're not going to deflect tickets. We're going to try and resolve the problem. We're going to build integrations with everything so we can resolve but we did exactly what they didn't want.
Nowadays, everybody's thinking has definitely evolved on this on the service side, you know, they're looking for, I don't know, speed, expertise, consistency, reliability, control, you know, which again, you don't always have with humans, either personalization. What's the term resource elasticity, you know, the ability to handle surges in demand? Yeah, I mean, I'd say that's the that's what people are looking for now, and they are starting to look for a multi-channel experience. Again, there's been that sort of case where it's like well, chats, cheap and easy. Let's do chat, as opposed to what do customers want, right.
So when you start looking at what customers want, you know, they want it they want to be engaged, they want to engage across all channels, you know, and then certain times certain channels so why, when, in a world where the customer where the power base is now with the customer, not with the company, why would you possibly try and dictate the terms? You know, let the customer dictate the terms because that's empowering and again, makes the customer feel as though you value them.
These types of solutions require a partnership with the company, right? It's not just this outsourced solution. So what should a company know about their involvement, their investment, both people, resources?
The first thing I would say is, you know, not all AI is born equal. And to really check that level of understanding that is, is it nuanced enough to be able to understand my customers across all channels? I'd say the next thing is to see how often the company brings up the idea of partnership, because I think it is a partnership no matter how great the technology is, you still have to have, let's call it expert, there are AI expert, you know, automation experts, that have done this, you know, 50 times they've really understand Okay, with this set of circumstances and these policies and these integrations, what is the best possible outcome?
And that's so that's, I think, that's not a nice to have, that's a requirement. But yeah, there is effort, there's like because if you think about it, what automation is going to do is it's going to, it's going to iron out all the wrinkles that of your business that you've sort of inadvertently put into the business over sometimes many years. You know, and what worked when you were four people in a garage didn't necessarily work when you were a multibillion-dollar company. And humans can kind of let's say hide or obfuscate the problems, whereas machines are much more like cut and dried. It's like it either it's either going to be this or it's going to be this.
So sometimes, you know, you have to understand the parameters of a machine and be okay with that. We've had companies that literally like sometimes they think that every integration comes with a magic wand, and Shazam, you know, like, all their problems and all their legacy stuff and everything else magically goes away. You're going to have to eventually, you know, make your business more automatable no matter what if you want to scale.
So, you know, any kind of integration into AI is going to force that and I think you there has to be a willingness not only on the on the customer service department side, but actually on the company side to resource that okay, because we've had various instances where because customer service is usually the caboose in any business. That is it is always grasping for resources, that sales and marketing and product get, especially engineering resources, that has to change. You know, in an economy that really requires business, in business terms requires you to put service first. That also means you have to resource services.
And I think that again, if you if you think of that we are entering a a purely entering a service economy, then resourcing customer service is not again, a kind of Oh, that's a nice idea. I'll park that for a while. It's actually vital to every company's survival.
I will say I mean we're in the Econ. So I can speak specifically to econ because econ is kind of the canary in the customer experience coal mine. What's happening to econ and the rift valley between a customer's expectation for service all those things I listed before them, you know, I want to prompt I want to personally want like an any brand's ability to deliver against those expectations, the shearing force is getting wider and wider and wider. And the wider that gap becomes, the more revenue risk there is. So the only solution to that is some form of technology and automation. It's not the cure for everything. There still has to be a lot of people involved. But it is a driving force in meeting some of those customer expectations. So it's kind of what happens in E commerce I think will happen everywhere else.
So we're entering kind of this period of economic uncertainty. You know, how do you see AI helping companies drive savings and customer retention?
Well, look so as acquisition costs rise, you know, whether it's, you know, Apple's, no tracking, whether it's cookies and Google like, or just general, you know, market forces. More attention gets paid on the LTV side of the business. And the happy hunting ground of LTV the path to that is through service, there's no other. There's no workaround. It is always through service. So I think it plays a more, you know, customer service in general. And as I said, automation within that plays a vital part in not just a business thriving but actually surviving through a recession.
So go out a little bit further, go like five years or so what's the level of agility and opportunity with AI and NLP?
So I think that the level of NLP is purely based also on the nature of the problems, you know, for instance, there's a there's an electronics company that does in-home speakers and I've, had issues with them. And I ended up like always with a Tier 2 Seattle based person, you know, going through Wi Fi frequencies, etc. AI is not going to be doing that probably ever. There are not enough use cases to build enough data to be able to do that accurately.
So if your use cases are a spectrum of pretty intense technical support, for instance, you know, the AI will take you know, will take a little bit of that, but you're no you're still going to have to resort to the like, how do I do this, you know, like, changing seats on a plane. And, you know, like, like, good support can do that. Now. Either if you're self-serve, or you know, a little bit I can do it or it can be done for me or in an automated way.
But there are other things where it can't hey, I need to try to think of it I need a hotel that hasn't had a terrorist incident for the last 50 years. Like there were all sorts of weird travel, things like that. Again, those are edge cases. And one thing I will say is that what I didn't realize going in and now I'm sort of a become a, you know, an exponent of is that customer service isn't, it is just a world of edge cases. There is no there's no like 80/20 rule out we can do 80% Here 20 It's all pretty much all edge cases and I think you have to be at peace with that. And again find technology that can deal with that.
Anybody that says like oh were going to handle these things without a problem is either delusional or lying. Because it's it's difficult stuff. And understanding customer intent is not for it's not for the faint of heart. Or the challenged. Yeah.
Well, Ted, really wonderful to speak with you. Thank you for sharing your experience and the challenges and the opportunities in this space. I think it's extremely exciting. And wish you all the best.
Conversational AI and Customer Service