Teaching bots to have empathy to improve business performance, with Peter Dorrington, Anthrolytics.

Hello, and welcome to this episode of The Unnoticed Show. Today, I'm delighted to have Peter Dorrington, who is the co-founder and chief strategy officer at Anthrolytics. Now, Peter, this sounds really smart. Tell us, what does Anthrolytics do for the unnoticed?

Thanks very much, Jim. It's a pleasure to be here. So what Anthrolytics is interested in doing is uncovering the motivations behind human behavior. So why people do the things that they do and what they're likely to do next. But one of the techniques we use that is very rare is we look at emotional motivation and we add that to the other kind of motivators. So you can say that, why does somebody buy a particular product? Well, we're aware of things like price and convenience and its fitness for purpose. But what about the way they feel about the products, and how does that work over the lifespan of them being a customer or an employee? So Anthrolytics, we combine data science with behavioral science to answer that question: why do people do what they do? But also to help businesses then make better experience decisions, particularly on digital channels, the ones that there isn't a human at one end of the conversation. So yeah, motivation into action.

Well, it's a big topic you've chosen there, I can see. And then it's also to some degree more and more important, isn't it? Because companies are getting more and more separated from their consumers, B2B and in retail. How are you helping companies then to kind of bridge this digital divide as we're all really now dealing with companies and service providers through bots and through online chat and not really seeing anybody anymore. Take us through how you're helping companies to, to understand what's really going on the other side of the chatbot.

Absolutely. So what happened over the last 14, 15 months with COVID of course, is the dash to digital. So many more businesses moved to a digital operating model, and that means that they often don't have a human talking to another human being. So many of us as consumers have learned to self-serve. We use bots. And for most organizations, that's a very satisfactory, competent job. You know, it does what we want them to do. You do what I pay you to do. But what's happened is that disintermediation, that, that break apart of the relationship, has really affected the experience economy. Now that came about as a discussion when we've had customer experience as a named discipline for about 25 years. It's not new, but we've done all the easy stuff. We've done listening to the customer programs, we've done customer journey mapping, but what customers are saying they want is empathy. They want compassion. They want to be more than a customer number. They want to feel that they're in a relationship with you. And when they do, all the research shows that all the business metrics go off the chart. So customer satisfaction leaps up. Well, that's interesting. But so does loyalty and economic activity. They spend more money, they spend it more frequently, and they enjoy doing it, so they tell their friends. So what we have focused on is how do you replicate or produce empathic experiences in digital channels, because you can. So you might liken it to teaching a bot how humans feel, because it's important. And the difference is that if I'm hungry, then that's a rational need. I need to eat, I'm hungry. But if I want a big juicy burger and fries, that's more emotional, you know, and what it's going to do is make me feel happy. But happiness isn't the only emotion, of course. I want to feel less afraid or more certain in these uncertain times. So at the heart of human-centered design, which is where a number of businesses are moving into, which is really genuinely put the customer at the heart of the business, you need empathy. Or to take that competence thing further, you need compassion with the competence. And when you do, the business results are outstanding.

But you talk about teaching bots to be compassionate. Normally, you know, a human-to-human interaction, there are clues given out. Somebody smiles, another person mirrors that behavior, for example. How are you helping the bot, which may not see the person, certainly won't be able to pick up on the normal human clues? How on earth are you doing that, Peter, from practical terms?

Okay, so it's a 2-step process, and the first step is to use natural language processing or natural language understanding technology to analyze what customers say. Now, there's a specific kind of question that we would typically ask, which is to ask the customer to describe something. So describe this experience or describe this podcast to a friend or a stranger. And when customers do that, they tell you what's front of mind about their experience. Oh yes, I remember Jim was really nice, I really gelled with him. And when we tell those stories, we leak emotions. So the first stage is to analyze lots of those and figure out what are the things that people remember and talk about. And also, what do they feel about what they talk about? Now, when we put that into a decision matrix where we think about all of the other things like the price or star ratings, what other people think, what you can do is associate an event like listening to the podcast to an outcome, and you can make decisions about what outcome do you want that's best for both parties. So the bot has a decision-making algorithm that says there's a logical part of this, but also if this is a customer who is likely to be feeling anxious or uncertain or perhaps unclear about what they do, I choose a different tone of voice or perhaps a different product for that customer which is better suited to what they want as well as what they need. So you can teach bots to do this, you can teach this in any form of automation where there is choice. And this is where the only predicate for doing this is that there has to be a decision that's based around a choice. Now, one of the reasons this can be complicated is because lots of the previous models over the last few years only worked when the customer had a completely free choice. So behavioral economics and nudge talks about free choices, but very few of us have free choices. I would love to buy that Italian supercar, you know, no matter how much I want it and I love the idea, realistically I can't afford it. So it's not a free choice. So the decision-making, the operational part of that, is using those insights to let the system make better decisions on behalf of both the customer and the business. And when it does, you get something that feels a lot more humane. Now the real trick is to be able to do that proactively. So before somebody starts to exhibit fear or anxiety, anticipate that's where Jim might be right now, and therefore we could reach out and offer you reassurance before you have to dial in saying, I'm worried, what do I do? And that is an incredibly human feeling experience, one that everybody enjoys.

Now, Peter, one of the aspects to that is you have an interaction with that company there and then. Our syntax may change, for example, but we've got all the legacy conversations and things that have happened to us, for example, in that day or that week or that month. Are you incorporating what's been going on for that person through digital data, or is it only at that interaction with that particular company? Because they have a very, if you like, rich long tail of behavior, don't they, before they get to a company?

That is such a good question, Jim, and it's one that's bedeviled researchers in this field for so long. The reason being that if you try to model that at the individual personal level, like me, you wanted to understand what my history did to put me here, and you use traditional techniques, it's simply undoable. There's not enough data and it takes far too much compute power. Nonetheless, our history informs our habits and our opinions, and that was what I took 2 years to figure out how to do. And in doing so— and I'm not going to talk too much about how I do that because that's in the black box— what we cracked was how we could extend that across an entire customer base of tens of millions of customers and update it every single day based upon what has happened for that customer. So rather than try and do regression, which is to build a really big complicated model or throw AI at it, what we came up with was a much more straightforward technique which was incredibly efficient, needs relatively little computing power, And although it's never going to be 100% accurate, and I don't think any behavioral model ever will because there are too many other things going on in our lives, it is way better than anything else that was out there. So to give you an example of what that means, you probably get offers from a marketing department where they've used a predictive model and they put you into a segment, and you are customer segment A1 or whatever they want to call you. And that's great, and there's a lot of people in your segment. But what we would say is that angry customers within segment A1 will respond very differently from happy customers in segment A1, but if you treat them both the same, you'll probably irritate both of them. But if you could tell the difference, then you could treat those customers differently according to their needs, and that's where the slight differentiation of the golden rule comes around. which is we're much more focused on saying treat customers the way that they want to be treated, not the way that you want to be treated. After all, the customer should be at the center of your thinking.

So in that sense, you're using legacy data from that customer within your own organization rather than what they've done before they got to your organization, is that right?

Yes, so imagine everybody talks about customer journeys in customer experience, okay, so we plot the journey and there's a happy journey and then we're gonna Derivations of that, but actually what we do is we think about customer landscapes because our lives are messy. They're not sequential. They're not linear. We have overlapping journeys. We get detours. We get nudged off course. Something else happens. So when you take the approach we've taken, your history is a bit like dead reckoning navigation. It's where you are right now and what direction you're going in and how fast you're traveling. After all, knowing that, we can start to take those decisions. And there's something somebody said to me once: It doesn't. matter how fast you're going if you're going in the wrong direction. Well, this is the same when we're thinking about customers. It doesn't matter how fast you respond if you're taking the wrong treatment. So it's not necessarily that businesses make wrong decisions, they're not making optimal decisions, they could make a better decision. Now, from the human point of view, this is not about reducing choice, it's not about saying certain people only get certain offers and others don't, it's about offering everybody the best possible choice. that meets their needs and the business needs. So everybody walks away as happy as they could be given the circumstances.

Right, and Peter, when you talk about combining data science and behavioral science, are you also moving into this whole area of sort of personalization and personalization at scale? Because if you've got everyone's data, you've got how they behave, you're able to really personalize not only the offer but how you deliver the offer. Is that something That you're really then Anthrolytics helping companies to deliver?

It is, yes. And the most talked about version of that in the digital context is what's called hyper-personalization. Now what hyper-personalization does is it takes personalization, what it is that we know about you, and it adds the real-time context. So it says, what is Jim, a customer about which we already know some things, trying to do? Now what we add to that is And what's the emotional empathic context of that? So how's Jim likely to be feeling? Because your decisions will be influenced by that much more so than people recognize. The estimates are between 98 and 99% of all of the decisions we make in any given day are in our subconscious, and they're really influenced by our emotions and our habits, and they don't actually follow logical rules. And this is one of the things I think many businesses get wrong, in that when they are talking to their customers, they think giving them more facts helps. And some customers it does, but quite a few customers, if they're confused, the last thing they need is more facts. What they want is help and guidance, and perhaps something that's more of a narrative that says, actually, this may not be right for you, but we think we've got something that will. And how many of us have made purchases on the basis of it was too complex to understand intellectually, but it felt right, it felt like I could trust that organization, or it felt like this would give me what I needed. Now we have the tools and the techniques to understand that and replay that back, not to do the dark psychology, which is to influence people to do things which is not in their best interest, but to actually say, if we treat customers like human beings and if we address all of their needs, both emotional and rational, It's a better experience for them, and it's a better business relationship for us. And it's one that is both stronger but also nearly impossible for our competition to reverse engineer because it's based upon a relationship, and it's the relationship that's the unique, and your competition can't replicate your relationship with your customer.

So this hyper-personalization sounds like we're kind of going full cycle back to where you used to have a personal relationship with your client— excuse me— understand them entirely. And they would share, as we used to do after work, go for a drink with the client, they would share the backstory, and that would build some of the relationship about what you'd be doing next, right?

Yes, it is. And obviously it's more difficult to do that in digital environments because it's not a human-to-human relationship. But people ask, well, how can you do that? So let's just talk about the 2 bits of empathy that are important. The emotional stuff, you know, what people feel about people, we'll put to one side because that's being treated quite well with emotional intelligence training. There's cognitive empathy, and that's very easy to do in our world. That is actually understanding why— what people are feeling and why they're feeling it. Okay, so I recognize you're angry and I know why you're angry. The compassionate bit comes in taking action as a result of that. Because you're angry and because of the circumstances, this is what we're going to do about it. When you link the cognitive empathy With the compassionate, the action empathy, that's when you can display it. And you can do that in a machine. You don't need to be a human being to do that. It just means make the right decision that reflects what that customer values.

Now, Peter, this all sounds amazing, but what about for your average company? How do we play out this hyper-personalization at scale as an owner-operator?

Big businesses, it's easy. The endgame is obviously fully automate and operationalize this and do it for millions of customers every day. But for smaller businesses, there are a number of things you can do. Almost all businesses nowadays have some form of customer feedback or voice of the customer program, and the first thing you need to do is ask the right question. And the right question is not asking people how they feel. If you want to know how somebody feels, don't ask them how they feel. Nobody can answer that properly. But to ask them to describe something, ask them to describe their experience, or ask them to describe this podcast, and when they do, what they will do is tell you What was the most important feature of that experience? Yeah, Jim was really pleasant and nice, and I really liked Jim. And when they tell you that story, they're going to leak emotions. Now, if you have no infrastructure, you sit down with those little verbatims and you look at them and you identify, you highlight, this is what this conversation was about and these are the emotional words. But if you've got a little bit of infrastructure, you can start to use some of the AI tools that are out there for natural language processing, and they will begin to strip out from those. This is a common theme. I'm sure you've seen word clouds, you know, where the most frequent word is the biggest word in the word cloud. Imagine, if you will, that that word associated with it are a range of emotions. This word is typically associated with anger and fear. Armed with that as a business, you can start to make better decisions with that. So at small scale, You can do it manually. As you get into a little bit larger, then you can use some of the online AI tools. We don't make those, but you know, there's certainly a load of them out there. But if you're a big business, you need to actually turn that into something where it doesn't have so much human effort involved. So natural language processing or understanding is automated, the stripping out of that what is important and how you use it is automated, and that feeds directly into things like your marketing system or your customer service system. And it's amazing how important all the interconnected bits are, because when something goes wrong, the customer doesn't blame your delivery driver or doesn't blame your supplier for letting you down. They blame you, the brand. You are responsible for any failure, and no matter how unfair that might feel from our point of view. So really understand what do your customers genuinely care about and why do they care about it, and then design your experience around that.

And presumably one can even be as simple as building out different templates for follow-up after a customer visit, for example. If you know it's gone well, you could have one. If you're not so sure, you can have another and have a decision tree of templates, right?

Yes, and that's exactly what we do with things like conversation design for bots. So if you, you know, imagine that you're phoning in with a complaint to a call center and you get routed through to some bright young thing who's very happy clapping, oh, hi Jim, it's wonderful to talk to you, and you're angry. That's the last thing you want to hear, right? So what you want is somebody who's perhaps a little bit more somber, you know, who's going to reflect back to you, yeah, we're taking this seriously, this is important. So imagine you're doing that with a bot and saying, well, you don't be happy-clappy all the time, you know, sometimes you need to have perhaps the script, if you like, is slightly different, you know, we don't use certain kinds of words or we take a certain tone of voice. And I like a very short cliché, you know, if you mess up, fess up and fix it. And I think that when you do that right, it doesn't have to cost the business more money. If you actually do it in the right way, you can save money, because how many times have you had people make repeat calls to complain about something? I phoned last Tuesday, I still haven't heard from you, I'm really angry. That's a failure of process, but it's also a failure of thought.

Okay, and really just, it sounds that we need to really start by listening, and whether it's as an individual or technology listening to what the customer is feeling before you respond. Now, Peter Dorrington of Anthrolytics, if people want to find out more about you and to work with you, how can they do that?

So the first thing that I would suggest they do is go to our website. One of the things we're very conscious about doing is spread the word. So if you go to anthrolytics.io, we've put lots of resources on the website that people can freely download that talks about the issues. If you want to find out more about whether this is the right thing for your organization, we do offer a free discovery workshop.

Peter Dorrington, Chief Strategy Officer, and I think chief analyst and cliché maker of The Analytics Company. Thank you so much for joining me on The Unnoticed Show today. It's been fascinating.

Thank you, Jim. It's been my pleasure.