What does it take to design a good conversational AI?

Jordi Torras
Hello and welcome to a new episode of the Future of Customer Service podcast by Inbenta. This is Jordi Torras, and I have the honor today to be introducing Hans van Dam who is working on a very interesting project, a very interesting company called Conversational Design Institute. So I’m really happy and honored to welcome Hans over here. How are you Hans?

Hans van Dam
I’m wonderful. Thank you so much for having me.

Jordi Torras
Absolutely. Where are you located right now?

Hans van Dam
I’m in Amsterdam. Yeah, so the whole company is based in Amsterdam and then we have people sort of scattered around the world, but we’re based here.

Jordi Torras
Amazing. Amazing. Okay. I’m having this session from Santa Cruz, California. So it’s amazing in the 21st century to talk to each other as if you were around the corner. That’s amazing.

Hans van Dam
Yeah. And I’ve been to Santa Cruz once, and I just remember being stuck in traffic all day, so that’s …

Jordi Torras
Okay. That was a pre-pandemic world. Now, the good news about COVID-19 is that traffic jams are no more, and most of the folks work from home and that’s let’s say one of the few advantages of that, is being able to not to be in a traffic jam.
Hey, I have so many questions for you because we know you guys for a while at Inbenta. We know that you worked with some of our customers, helping them, understanding them a lot about their conversational design. So I can’t wait to start having, asking a few questions. Are you ready?

Hans van Dam
I’m ready. I was born ready. You bring it on. You bring it on.

Jordi Torras
Awesome. So Hans, what do you do at a Conversational Design Institute?

Hans van Dam
I guess that’s a good question. Right? So we help companies make their AI assistance more human-centric through design. My background is really as a copywriter, I started out, I always wanted to be a writer or like a novelist that I … Nobody was interested in that. So I became a copywriter, and I worked with tech companies. And then I ended up working in customer service for a bit. And then when conversational interfaces became a thing, I was like, “Hey, I can probably figure this out because I know how to write dialogues and I understand the technology from my time in an incubator, and I understand customer service,” because I’d worked in customer service. So I figured I’m going to figure out how to do this.
And that’s sort of where we discovered that new job, a new role called conversation design, where we saw that a lot of these conversational AI project were very much engineering focused and a lot of companies were not successful because they would just put engineers in a room to then automate the conversations they had with their customers. So that didn’t really work. So we were like, “Okay, we need to figure out how to actually design these interactions, how to write these dialogues.” And we started exploring that.
And our reasoning was that, if you have an artificial brain talk to a human brain, then technology, psychology, and language should all be equally important. We take a lot of expertise from behavior design and figuring out how people think and behave and interact. And we created an entire workflow around that. And that’s what sort of kicked off the company.
So what we now do is train and certify conversation designers mainly for enterprises. We have a consultancy that helps them as well. And what we see is companies are looking to automate conversations with people about 85%. A lot of them try to automate about 85% of the conversation that they have with people. And to do that, they use conversational AI technology. And we sort of see these enterprises built in-house teams around this, right? They need technology and they need people. And we try to figure out what do these people do? What’s the skill set that they need? How do they work together? And that’s where we really come in to help them.

Jordi Torras
Absolutely. Absolutely. That’s fascinating, right? Because there’s a lot of technology out there solving AI, machine learning, NLP, and whatnot. But at the end of the day, it’s like, “Okay, what is this chatbot going to say?” And this is what I am understanding now is that you guys help your customers to do that.

Hans van Dam
That’s … So the way we saw the market initially is that there were a lot of companies out there that were selling hammers and we’re the only ones training carpenters.

Jordi Torras
Wow. Wow. That’s … I love that. I love that sentence. That makes a lot of sense. Everybody’s trying to sell hammers, small or big, and say, “Okay, what I’m going to do with it?” What is the kind of problem that your customers perceive when they call you? Why do customers call you?

Hans van Dam
Yeah. So a lot of times you have an enterprise that has bought into that promise of conversational AI technology. So they’ve bought it, they’ve invested in it. And at the end of the day, it’s just not delivering. People aren’t talking to it, or they’re not enjoying talking to it, or they drop out of conversations immediately. And then they discover that there’s something that is new, that they need to figure this out. And that’s when they start to reach out.
If you look at, going from graphical interface to mobile and now to AI assistance, from a design point of view, you used to have graphical interface design, then UX design became a thing thanks to mobile experiences. And now, all of a sudden, it’s conversation design. It’s an entirely new field and an entirely new craft that people are exploring.
So what enterprises have, they’ve got the technology, they have people, but then they discover that these people completely lack the skill sets because they don’t understand how to create a human-centric conversational experience. They don’t know the dynamics. Language is very complex. I mean, you’re on the other end of the spectrum, figuring out what people say, but we’re on the other side trying to make sure that people also feel understood. And that’s really, that’s where we come in. That’s the problem that we solve for these customers.

Jordi Torras
That makes a lot of sense. And how do you know the job is done? What is the objective? What are you trying to accomplish when you’ve engaged with a customer on one of your projects?

Hans van Dam
That’s a good question. So a lot of times, so our belief is that they will eventually have their own team. And people ask, “How many do I need?” I often ask, “How many UX designers do you have?” “Well, we have about 40.” “Well, then you’re probably going to have about 40 people working on your chatbot five years from now.” So that’s why we want to get them.
A lot of times when we have these enterprise engagements, we show up with our own consultants and we first do the work for them. We kind of show them like, we make sure that the top journeys are handled well, that your tone of voice is set, that your key journeys are done in a proper way. At the same time, we start educating their team. We find the right people in your organization. Then we have an online academy with video lectures. And we do workshops to slowly skill up that team and take them by the hand. And we kind of do it together. And then we kind of go more and more to the background.
We’ll be on call when you need us. But over time, the goal is to let an enterprise have their own conversational AI team with people that have a structured workflow, that know what to do, that make sure that there’s empathy, and make sure that customers get their problems solved and their questions answered.

Jordi Torras
That makes a lot of sense. So a question. Let’s say I am a customer of yours, right? Then I say, “Hey, I need my team to learn all these things.” What pre-existing knowledge or skills should my people have? In other words, what should they know already before talking to you, or when you feel more comfortable with working with teams?

Hans van Dam
Yes. What we see, so we have three certificates now and they sort of emerged over the work that we’ve done and we defined three roles. There’s the AI trainer that really turns data into understanding. Part of that is training a language model. Your customers won’t have to do that because that’s handled. But they also go for the analytics and try to find improvements, etc. Those people often, overqualified customer service people, they’re often very used to working in complex systems and talking to customers. So they’re human-centric. They’re not afraid of some software and technology. So a lot of times you’ll see overqualified customer service reps, or very light data science sort of going into those roles.
If you look at the conversation designer, what they do is to really think about the structure of the conversation. For example, how do we, in a natural human-centric way, create a flow for a conversation. A lot of times these are your content managers, content writers, also some customer service people that are more language savvy or UX designers, sort of that background end up in those roles.
And then there’s the conversational copywriter that really thinks about the tone of voice, the real wording, that thinks more about behavior. Like how do I say things in a way that I control what people do? These are your expert copywriters. These are also people with a background in screenwriting and creative writing, and psychology. So people from marketing, people from communication. That’s often sort of every role has their own little group of people with their own unique backgrounds.

Jordi Torras
Got it. Got it. That is super interesting because we are in the same space. That idea of an overqualified contact center agent, that’s a really fascinating concept because at some point you know that call centers have a high turnover rate, and folks working in call centers, they might already be looking for another job. It is a temporary job. But here what you are telling me is like, hey, actually, you learn a lot of your company, your business, your customer surveys, and then you can use that knowledge to empower your entire organization with a right to training, right?

Hans van Dam
Yeah, exactly. And I think a lot of enterprises struggle with that career path for they have smart people in a service center, and a lot of them joined the service center during their studies, during college, university, and they sort of do customer service work on the side. And then either they graduate and they leave or a lot of them don’t … There’s always a few that never graduated and sort of stuck in that role for longer, whether actually very smart, they’re creative people. Then you take them by the hand and say, there’s an opportunity for you. Usually, they would go from being an agent to maybe a supervisor. And if they’re lucky become like contact center manager in a way, right? But now you can actually go from service agents to AI trainers, to conversation designers to copywriters or to product owners. All of a sudden there’s an entire path into the world of conversational AI.

Jordi Torras
Wow. Wow. So there are two benefits here, right? On one hand, the company may retain that talent through this professional career and path. And then, at the end of the day, the company also benefits from better AI and better automation. Right?

Hans van Dam
Yeah. And I think that’s really key. And I think a lot of people sort of don’t realize what kind of value they have in terms of talent in that organization already.
One of the things that we’re also working on is using an assessment to sort of, if you have 500 service agents, we will be able to tell you very quickly with one psychological test who could potentially join your chatbot team. So we identify those people. Then we skill them up, we guide them, and we make sure that you end up with that conversational AI.

Jordi Torras
So you have all of those tests ready?

Hans van Dam
Yeah.

Jordi Torras
In the spur of the … That’s amazing. That’s a very sophisticated process.

Hans van Dam
Yeah. And so what we could eventually do is we can do an audit on your chatbot and say, our philosophy is kind of like the more conversations you have, the better they should be. If you look at volume versus experienced design, the more you talk, the better it should be.
So we can look at how many interactions do you have, what should the quality be in terms of experience. And then if you’re below that, why is that? Maybe you don’t have the right people. So let’s go out and find people in your organization that can get you to that next level potentially. Everybody has people that are very strong writers, and sort of write in their evenings and write stories or write blog posts. Well, those people are perfect. Now you just explain to them a little bit about technology and then they can join the team.
I think it’s easier to teach creative people about how a chatbot works than to teach an engineer how to write very creatively. So all of a sudden you have these career opportunities in AI for people that have no background in technology.

Jordi Torras
Absolutely. Wow. That makes a lot of sense. So let’s talk objectives. Say, okay, we engage with you guys. We have this project. What are the tangible goals and the tangible objectives that you will help create?

Hans van Dam
Yeah. So I mean, one of the most simple things is the containment of a conversation. This means making sure people get to the end of a conversation. That’s the completion rate. So making sure that they get, people get help so that they don’t have to go to the agent and you just lowered the volume in your service center. That is usually really the one that is the main objective, what we come in for.
So it’s increasing the success of the conversation that you have. And a lot of times it’s always also with intent-based models or machine learning models, making sure the recognition of questions goes up as well. Right? So there’s understanding better, improving cognition, and also improving the completion rate of these conversations while maintaining NPS. But they go actually hand in hand.

Jordi Torras
Absolutely. So you can actually measure the return on investment, right? You say, okay-

Hans van Dam
Yeah.

Jordi Torras
We can get that team. Now it is trained and the products are delivering, and the benefit is something tangible that we can measure every day, and not only today but every day after that because these new skilled people will create content that will stay forever. So that return on investment is something that the companies will enjoy. Not only immediately after these people will have the skill, the new skills, but also they will stay there for a long time.

Hans van Dam
Yeah, that’s definitely true. And what we’re actually doing now with one client is running tests where we can actually see in different countries and across markets if we change this little phrase in this message, what’s the impact really in return on investment? We change three words and at the end of the year, it is going to save them two thousand bucks. And we can do that for each message.

Jordi Torras
Wow.

Hans van Dam
And some of those things are like the key in conversation design, sort of the workflow that we teach, that’s kind of what our teams do. If we really dumb it down in just a few steps, we mentalize the context of the conversation. So we don’t just say like, “Oh, somebody lost their bank card, and therefore they need a new one.” It’s like, “Oh, wait. What’s actually going on? Someone’s insecure. They’re upset. They lost a bank card. They can’t pay. Maybe they’re worried even about paying groceries tomorrow and they have a kid to feed.” Like what’s actually the dynamics of a conversation. So we want to understand that.
And then what we do is we have the conversation. We role-play the conversation figuring out what goes on. And we sort of do improvisational theater almost on there, and I say something to you, you’re going to respond to it and you’re going to … Any concern or thought that you might have, it comes out. And then it allows me to sort of capture everything and create a first draft of the conversation. And then we validate that with other people. So we sort of test that real quick. We create the first draft, we test out the conversation flows the way we want to. And then we optimize that for behavior.
And what that means is that we’ll sort of apply a layer of psychology to make sure that the success rate of a conversation is much better. So if I say, “Ask me a question,” you can ask me any question. If I use some smart framing and I say, “Ask me a question about a glass of water, ask me a question about water,” you’re probably going to ask me questions about water, right? So already there by using behavioral psychology and specific wording, I can already guide my user to resolution.
What I do through that role-playing exercise is to create a lot of empathy. And if I talk to you in a way that you feel understood, you have a lot of confidence in that I’m going to get you to the resolution as well. Right?

Jordi Torras
Absolutely.

Hans van Dam
So if I show empathy, you’re coming on a journey with me. And then I might identify a few issues where maybe I need to ask you five questions instead of saying, “Here’s question one, here’s question two, here’s question three.” If I say, “I’m going to ask you five questions. For most people, this only takes a minute.” What I’m doing now is I’m using social proof and I’m using expectation management, and this removes stress on your end. So you feel understood. So you know I’m going to get you to the resolution. You might feel some barriers. So I’m using social proof and expectation management to guide you to the end of the conversation. And that’s really the skill set that we teach and that we have applied to these chatbots to increase their success rate.

Jordi Torras
Wow. That makes a lot of sense. Let’s talk about hammers for a second. I love that symbolism, right? It’s like, okay, everybody is interested in selling hammers. We teach chiropractors. I love that. But my question is, when you have your customer, I’m assuming that at the end of the day, the company is already engaged working with some kind of technology, specific technology. And we know this, the classic subjects. There’s LUIS and Dialogflow and Watson and all these guys and Inbenta. Like everybody trying to be there selling their own hammer.
So my question is, once you have a team that is trained by you guys, and using as a background a specific technology, how difficult it is for that team to then move to another tool? So in other words, to what extent if a company trains their team in conversation design for one specific tool, if the tool gets replaced, how much of this training can be reused?

Hans van Dam
Yeah, that’s a nice one. We train people. What we preach is voice first. Our belief here is that if you could figure it out with voice, which is the minimum real estate that you can have in an interface, you can also use it in voice, but also in a chatbot or on a third party chatbot. And you can even use a button and a picture all of a sudden. But if you figure it out with voice, you can use it everywhere. So that’s the first thing in our entire workflow. Everything is voice first.
The next one is really technology agnostic. So we teach people how to create a good conversation. And we teach them, particularly with the AI training stuff. It’s also very generic. It’s really, how do people talk? How do we understand that there are different ways of saying things? How do we make sure that we have a model and the structure and how do we … Everything is technology agnostic. And if you teach people, if you train people in a technology-agnostic way, it becomes much easier for them to work with different hammers.
So we do see a lot of these out of the usual suspects, these tech companies, where they have their own course on their specific products. But it’s really like stepping into a car and there’s like, “Yeah, so here’s where the key is and here’s where the dial is,” but you still don’t know how to drive. Right? So you need to learn how to drive. And then if you have a different car, it doesn’t really matter. It’s like, “Oh, where’s the navigation? Where’s the air conditioning? How does this work here? Oh, wait. Oh, it’s the linguistic model. All right. Perfect. That’s … I don’t have to bother with that too much.” Right?
So you can spend like a day or maybe a day or two to familiarize yourself with the new technology. But let’s say you were first if you’re used to such an easy to use platform like Inbenta and, all of a sudden, they swap it for some reason, I don’t know, with IBM Watson. Half your team is going to go home with burnout or at least call in sick, or like, “Yeah, that’s too difficult for me.” Right? Whereas if you teach people how to craft a proper conversation, and how do you want to talk to people, and how do we make sure that we have the frameworks in place to create that conversation, you can deploy it with any technology and on any conversational interface.

Jordi Torras
Absolutely. That makes a lot of sense. Hans, can you tell us about some practical example, some specific situation, some use cases? Delight us with some examples.

Hans van Dam
Well, there are so many. So what we see now, a lot of it, we now see more and more internal chatbots, which are kind of fun. I think one of the things to point out there. How frequently your customers talk to the assistants has a lot to do with how you want to talk back to them. These are always interesting things to think about, right? So if I’m a customer and I only have a billing relationship with my company, and I only show up if there’s a problem, I have an entirely different tone of voice than when I’m an employee chatbot that’s helping out on a daily basis and there’s much more engagement and there’s a lot more fun.
In those internal chatbots where you want to have a lot of adoption, it helps create a lot of personality content, which is fun. So when you design a chatbot, we always create a persona. I don’t know if you’re familiar with that. It’s like the tone of voice and the backstory of the assistant, and you create something like that. But then there’s also something called the personality concept. So that’s chit-chat, little jokes, little Easter eggs to make it more fun and engaging. And what it does is actually makes people more forgiving, right? So if I only have a few use cases, but I do have an engaging personality, then there is more to do than just that one question. It becomes a very delightful experience all of a sudden.
So we had this on a cruise ship where there are smart speakers. And people would talk to them, but there’s too much stuff going on on a cruise ship. There are all these different buffets and entertainment things. So in terms of use cases and stuff they could do  it was limited, because it was too complex. So what we do is we actually add a lot of personality content. So it’s like, “Hey, when does the sushi bar open?” I’m not sure, but then actually have some reference to sushi and an engaging experience it makes people much more forgiving.
So those kinds of little things are always fun to think about. Like what’s the frequency of interactions that you have with customers? What do they expect of the assistant, that sort of influences your tone of voice and the content that you’re going to produce. There are so many things, I guess.

Jordi Torras
Totally.

Hans van Dam
Say it out loud, try it out, have people interact with it. And then where the conversation goes flat, if we do that role play and I ask you a question and you’re quiet for two seconds, probably means it was too difficult what I said and you’re processing it. So it’s a flag on the play and I have to rewrite that and figure out how I can then use behavioral psychology to solve that problem. A lot of times there are cognitive overloads. Our messages are too long. A very simple rule of thumb. If you can’t say it in one breath, it’s probably too long.

Jordi Torras
That’s another-

Hans van Dam
That’s like … Oh yeah. Okay. If we go practical, this is what we see a lot. The company has a long answer. It’s too long. You want to say the whole message out loud. Takes more than one breath. Okay, it’s too long. So what we then do is the Jenga technique. So you’re familiar with the game Jenga, right? The tower of …

Jordi Torras
Sort of.

Hans van Dam
Yeah, yeah. So you’ve got this tower of little blocks. You take out these blocks. And whoever makes the tower collapse loses the game. Well, we do that with writing. When you’ve written something, usually you can remove about 50% of the words that you’ve written. You can go back to. Everyone just please listen to this, if go through your emails, go through the blogs that you’ve written, go through your chatbot. A lot of times you can take out 50% of the words without the tower of meaning collapsing in your message.

Jordi Torras
Still in there.

Hans van Dam
Yeah. So it’s still there. So check how long it is. Take out about 50% of the words. You’ll discover that the message is still there. And then you rewrite that a little bit where what we always have is an acknowledgment. That’s kind of what you’re doing now. Like you’re nodding your head. What is that in a chatbot? It’s like an okay. Check. Sure. Makes sense. A little phrase like that tells the user that you’ve listened. Then you use a confirmation that signals to the user that you’ve understood what they said. And then you have a prompt where you ask a question to signal to the user that it’s their turn to speak again.
So look at your message. If it’s too long, take out words. Then apply acknowledgment, confirmation, and a prompt. Your chatbot is going to perform a lot better probably tomorrow if you do that.

Jordi Torras
Wow. And I believe it was Blaise Pascal, a French philosopher who at some point had a very long letter send to somebody and said, “Okay, sorry, this letter is very long because I didn’t have much time to write it.” Right? So it’s exactly what you say is like, well, if you don’t think about it, if you don’t work on the content, you end up with too long. So being long, doesn’t mean it’s well worked. It might be just the opposite, right? That’s-

Hans van Dam
It usually is the opposite. And it’s really those simple rules of thumb. If you make a checklist and have that there and figure out how to do that, life becomes much easier.

Jordi Torras
Absolutely. Absolutely. All right. Time flies talking to you. Really it’s amazing what you guys have accomplished and the kind of knowledge that you are providing. Unfortunately, we have this limitation on the podcast that we need to stick in. So how people can find you? How can contact you?

Hans van Dam
I’m not that hard to find. So conversationdesigninstitute.com is the easiest way. I’m on LinkedIn, Hans van Dam. Twitter, hvdam. Really, if you google Hans chatbot, you’ll probably find me.

Jordi Torras
All right. Okay. I’m going to do that right now. I’m going to do that: Hans-

Hans van Dam
Oh god, oh no, no-

Jordi Torras
… chatbot. And this is google. And here we go. Here you are.

Hans van Dam
Yeah, yeah. Okay, good, good, good.

Jordi Torras
That’s amazing. That’s amazing. All right. So thank you so much for being here. It’s really an honor to having you here and sharing all this knowledge with us. And I believe that our audience will really appreciate it because I think that there are many takeaways already for all of us from this conversation. So I want to thank you for being here.

Hans van Dam
Well, thank you so much for having me. I look forward to doing more work together.

Jordi Torras
Absolutely. Here we go. All right Hans, thank you. Take care.

Hans van Dam
All right. Bye.

Jordi Torras
Bye-bye.
 
Thanks so much for tuning in. This podcast was brought to you by Inbenta. Inbenta symbolic AI implements natural language processing that requires no training data with Inbenta’s extensive lexicon and patented algorithms. Check out this robust customer interaction platform for your AI needs, from chatbots to search to knowledge centers and messenger platforms. Just go to our website to request a demo at inbenta.com. That’s I-N-B-E-N-T-A.com and if you liked what you heard today, please be sure to subscribe to this podcast and leave us a review. Thank you.