Transcript:
Now come to the stage Hans Von Dam. Thank you very much. yes, let's see. yeah, that's working dusted. that's good. that's why you'll see me doing this a lot show of hands. Who knows who this is? A few people so this is me, Finn. She's a character on a tv show called Westworld and what sort of takes place in the near future. it's fantasy park and people go there to sort of interact with robots.
So, maybe there's actually a robot and she's a host in the park. and on one hand. you're very service oriented and cater to everybody's needs and they also tried to persuade people right which you'll see in the show is that there's a combination of artificial intelligence and people actually writing the dialogue if you now look at like the chat bot. if you know we're using chatbots for sales and persuasion and for service and it is that combination between ai and people actually writing the dialogue. yeah, this one's probably more familiar right this from the movie here. It's about a guy that buys an os and he sort of starts talking to it.
It's very awkward in the beginning but then he feels more comfortable around it and they sort of become friends. they become very good friends and they fall in love and I built this intimate relationship and all of a sudden, he sorts of discovers that she's talking to tens of thousands of other people and that's what he sorts of drops out of the experience. he ends up in the uncanny valley and that's sort of the end of it as ruins it for it, but it sorts of shows you that you know people and bots will be living together. we'll be working together. So, we really need to learn to communicate with each other. yeah, so generally we're going to have a new role calls conversation designer, right? so that's what we do.
We recognize, develop and promote the rule of the conversation designer. we trained and certified people and we helped branch directly next one. So, some stuff that we work on is like a cruise ship while you have a smart speaker and every cabin and you ask questions about the ship and you know, it's a sushi bar open is the karaoke. key the whole experience you can sort of do in a cruise ship or if ways experience that if you go to a restaurant and then you tell you tell google is like, hey remember that bill from yesterday. can you split it with my friend john then john gets a payment link through to google assistant kind of cool? but what we really do is work with big companies big tech companies to figure out what the department is going to look like every enterprise is going to have this conversation all AI department five years from now.
So, who's going to be working there? what are they going to be doing all day and that's what we develop curriculum for and train people for and yeah, let's go to the next one straight away. right? so if there's one slide that I’d like you to remember I guess it's this one if we're going to have a robot talk to a human then two things really stand out and one has an artificial brain and the other one has a human brain and they both have their own limitations and capabilities and they both have triggers that they need to function and what you see now is that the whole industry is only focused on natural language understanding right that turns, you know natural speech into structured data so that an ai brain can perform but in the other hand, we have a human right and they have completely different needs. they need empathy and helpfulness and guidance.
So, we need to design for that and if we're going to have conversations between people and bots than natural language creation should be equally important as natural language understanding next one. So, if we talk about conversation design, we should learn about technology psychology and language and also for you look at your departments that way that set up you want to have a similar balance and what you see now is that most companies have like for every writer to have 10 engineers, but you probably want to have a better balance between the two to create better and more natural conversational experiences.
So, there's a lot of techniques actually that you have to learn to become conversation designer and there's a lot of stuff that just comes from copywriting or from blogging or just traditional conversation design. So, I’ll just walk you through a couple of those. to get some idea of what a conversation designer actually does. oh, yeah, three mistakes. yeah, let's go with that first. let's go to the next one straight away. Yeah. So, what we see companies do a lot of time and everybody's making these mistakes and you're probably making them as well is to start with technology to focus on knowledge management and to lead with the business operation, right? so they go out and they buy the school conversational ai stack.
They start building the engineers are like a year into the project already and that's when they bring in the first writers. that's a bad idea or like some knowledge management think from the all the exceptions and try to give the most complete answer all the time that doesn't work either or sort of start with like the business process that you have and then you turn that into a flow chart and that you give it to the rider and say we need some words for this that doesn't work either because that's not how people talk right. So, this is really, you know, avoiding. these mistakes are going to save you a lot of money already.
Let's go to the next one straight away. yeah, so sort of illustrate how people communicate there's a really bad joke that I have to tell so there's two fish and they're swimming in the water. and then I see a third fish is like hey, good afternoon. good afternoon. how you doing? how you doing? and then this site, you know isn't the water lovely today and the two fish look at each other and were like what the hell is water? because fish don't know about the water because they're surrounded by water and the point I need to make that's what a silly joke is for. it's the same with people in communication, right? so we communicate all day long our entire lives and they're very good at it.
We have all these advanced techniques that applying that were not even aware of, right? so if we want to design better conversations, we first need to understand how people communicate deconstruct that and then we can apply it to conversation design. So, the cooperative principle some of you I was referenced yesterday as well. it's really that when people communicate so identify but paul price there's a couple things that we do we sort of provide truthful information. we provide the right amount everything we say is relevant to the conversation that we're having and we do it in the right tone of voice pretty much and this comes natural to us and is very complicated for an eye.
So, if you know we tend to skip steps you can pay using PayPal credit card. do you want to pay I’ll use Mastercard, right? so don't expect a yes or no answer to a yes or no question. that's not how people talk. it's not going to work in a chatbot. you have to design for that or this one. how many rooms are in your house? well, let's see. there's the living room two bedrooms children's room kitchen bathroom. that's how people answer that question and that obviously never works. So, how most companies solve this problem. it's you know, we'll make like a list of all the different rooms that we have will create synonyms for that and then if we hear bedroom will do plus 1 right and then kitchen will do plus 1 and that takes us about a sprint or two and then we get it to work. but if you got conversation designers on your team go to the next one, you can just say, you know, sorry how many rooms does that and you see a lot of companies actually, you know turning this into an engineering problem, but you can actually save a lot of time and money by seeing it as a design problem and this is just called a rapid re prompt and it's takes five minutes to write or five seconds.
If you know what you're doing and then you solve this issue. but also, we infer meaning right. So, are you all set for the meeting for today? what time does it start again? please answer? yes or no? that's not how you want to communicate right? but this is how most chatbots do that? yay, so you would want to say oh, you still have three hours and it's only a 10-minute drive to get there. this is what conversations are between people that it's very difficult for bots. yay, so, let's go to the process that we have. yay, so we put together this design process because what we see is that since there are so many engineers and so few designers is that it's very difficult for them to get and going they don't really, you know, they show up with the best intentions and to write the best way they can but they don't really have the skill set or to process.
So, that's what we've given them. you take your use case and then we have, you know to bot needs in the user needs. I’ll show you that in a second and then you do something called sample dialogue which is like improvisational theater pretty much and then you turn that into a flow chart so you don't even start with that. they do some quick wizard of oz testing and the expert rewrites and what we see is that at having this process really allows your writers and designers to build more confidence as well because they now have a structured way of working. they have a way to you know, they have supporting arguments for whatever it is they design so it builds a better team in relation to the engineer's so we have a canvas for the user needs and the bot needs and this is like a like an overview pretty much it's more detailed. but the way we sort of see it is that we're going to have a conversation and the human has both rational needs and emotional needs.
It has certain expectations. it has certain experiences. it has certain anxieties and motivations. So, you want to map that out on the canvas and it's the same for the bottom on the other hands, right? it has certain capabilities certain superpowers that can look stuff up real. wake or certain limitations information that it has to ask information that it has to share. you just want to put that on a canvas and maybe very detailed about that because you're going to use it for the next step. aren't you, Dustin? yes, so the next step you're going to do two sample dialogue and what it pretty much means is that one person is going to be playing the chat bots and the other person is going to be playing the user and you're going to be sitting back to back from each other. are you sort of going to study what you just put on the canvas right? are you going to get into character? are you going to sit back to back from each other? so there's no visual communication and it's just your voice that you're designing with and you'll do that role play and it's very messy in the beginning, but you'll get better. are at each time the user asks the question that's sort of a sign of a concern or an anxiety that they might have.
So, you want to done proactively communicate that and you'll do a couple tries and you'll quickly get to the most natural conversation pretty much. and then the next one is to do wizard of oz testing. So, once you have that dialogue designed, you're not going to just implement it in rasa or dialogue flow or whatever you're using. you want to validate it very quickly. So, you probably have from the roleplay bunch of post-it’s on the war on the war on the wall, which are France now. we just going to bring people in like bringing the coffee later the cleaning lady or a colleague and just run them through the dialogue you're going to quickly validate if that design that you just created if that holds up and make sense. So, you got feedback in 30 minutes all your entire dialogue.
It's also a great way to collect the utterances for your training model. right because whatever people say that is what your users are going to be saying. and then the next step really is the detailed conversation design where you going to be looking at the long tail where you're going to be handing it over and particularly the error handling in to repair flow. So, a lot of chatbots aren't working because you're not putting much enough attention to error handling. right? so a lot of companies are a lot of users are being handed over to an agent way too soon. she can just do use are prompt to get them back into the dialogue. So, yeah that that would be the next step and then no, you should have your good conversation and should be ready to go live. So, really the benefits of that human centered design process are that you're going to save costs during the implementation because you're going to go live much faster. you're not going to start building stuff that you have to rebuild later on going to save time optimizing get higher completion rates and get a much higher nps score go through the next one.
Yay. Yay. all right. talk about persona a little bit of got to speed up a little bit here. yeah next one. it's called pareidolia. you see a bunch of them just skip through them. So, there's not a single person here. and now that doesn't see your face in this it's quite interesting how that works. right? so we all know that it's not a face, but we see a face in it. it's the same with voice as if we hear a voice within one second. we're actually creating an image of would advise belongs to and we'll assign gender to and it will figure out how old are they and where they live and how could it be of value of me? and they create this entire image of who they're talking to right? so if you have a million customers talking to your voice assistant or your chat bot then they will all create a persona for it. and that's quite tricky right? that's not then you're not in control. It's the same with chat as well. right if you're chatting. someone tinder is an example of this like if you start texting with someone, you'll build up this image of who they are and what they're like and how wonderful your life can be together and then you meet them and it turns out that you know, it's completely different right so that just how that works. and so, we don't want to do that. So, if we need to craft a persona to take control of that entire process what's taken place in that user’s mind, otherwise going to be inconsistent less likeable and less trustworthy and we get to that uncanny valley, right? right, and that's what we want to avoid. So, one hand we want to make sure that we sort of we have these users that are going to get carried away with these conversations and going to lose themselves in it. So, we want to make use of that to create longer dialogues. but at the same time, we want to continuously communicate. Hey, this is only a game that we're playing and we're just having fun with each other right and that's how we sort of steer around that uncanny valley that we want to avoid. So, let's show you some writing stuff really quick to wrap it up. yeah, right conversational, please state your last name should be dusted. what's your last name? you can go which gender are you we see this one a lot is it? Mr. or Mrs. bro. let's keep them going turn-taking don't take is really important. we see that go wrong a lot in chat bots. So, can you order me a cap and your chat bots as sure why do you want to go starting fee is 750 etc. you see this a lot in chatbots. it's actually really bad because what you're doing is, you're asking a question, but now you're adding more information immediately and then you get those situations that you sometimes half when you're talking to people on the phone. it's like you're not sure whose turn it is to speak right because it's very confusing. all right, he's confused too. So, let's go sure. where do you want to go? lights are blind and writes about 25 years. is that okay? right. So, you want to have these very short prompts with very simple questions to which people just say, yes, and no pretty much. let's go to the next one. it's called an explicit confirmation. I would like to book a table. you would like to book a table. is that correct? so that's the explicit confirmation and I will show you the implicit confirmation, right? I would like to book a table. sure. that's an acknowledgement. this is an implicit confirmation. I can book you table. what time should I book it for tonight? and what's interesting is that people do this as well? right if I go to a bar during the day and it's very quiet and I say hey, can I get two beer then? the bartender will probably say are two beer coming right up, right if I go to the same bar in the evening and there's a lot people there and there's a lot of music he'll probably say to beer is that correct? so we did as to when our confidence level is low. We use an explicit confirmation versus an implicit. distant yay. going yeah, so the way to look at it where we want to use bots this some model by bill price from amazon and pretty much their stuff that's super valuable to the customer and super valuable to the business right and stuff. that's not valuable to either. we just eliminate altogether, but for the other stuff that that's valuable to the customer but not so much to the business. we can automate that right? so that's where we use chatbots. that's our reverse domain the top left. that's what we want to be more persuasive. So, that's tough that's valuable to you but not so much to the customer so we can use chat bots there to simplify it and the difference between the two. is that the bottom, right? that's what your user actually has an intrinsic motivation to talk to your butt. So, you can keep the dialogue a bit more transactional a bit more straightforward and then the other one you got to be a bit more persuasive. So, you need to increase motivation, so there's some writing techniques that you can apply to her. So, the simple question technique does you ever think about the future if you have this on your website and it pops up these little questions, you know, simply just click in the button is easier than just sort of clicking away the entire screen. So, people start engaging with your chat bots and you know, you'll be able to convert them at one-point simple question technique. we have the exceptional benefit and this is traditional static copywriting right, but we can apply to chat bots when you get a discount by answering a few simple questions. yes. no. only more so all these techniques from advertising are going to work for chatbots also choice architecture want to know how to win friends and influence people, of course. yes, maybe no so if you add more positive options, then the odds of your user selecting a positive option greatly increase and these are techniques that you can apply everywhere in the conversations that you're designing. small commitments work to when I answer three simple questions. yes or no. it's also when people say yes three times in a row the more likely to give a positive answer to fourth time. So, we're working with a bank and we had the question of what's the origin of your wealth was an investment bank and everybody would just drop out of the chatbot that right. nobody likes to answer those questions. So, we said, you know, hey, thanks for working with me. am I doing a good job? yes. do you have another minute? yes. are you ready for the next question? yes, what's the origin of your wealth? and you see conversion rates go up 40 percent immediately, but just applying these little psychological tricks. So, here you can combine a bunch of stuff right to lock in the order. I’m going to ask you three. simple questions for most people there's only take a minute. does that sound good? yes, share tell me more. So, this is one prompt pretty much but it has social proof expectation management the reward at the end of it and choice architecture all applied in one prompt. and this is how you want to be designing your conversations to get better results and to make them more natural. we get the next one. yeah, and then there's managing emotions. there's a lot of psychology behind that. you can just apply for your chat bots isn't there, Dustin? yay. So, we developed the escalation prevention model and a pretty much means that you have these negative emotions that can pretty much turn into undesired behavior. right? so somebody that's very angry can become revengeful or someone. that's sad you can sort of and the relationship yay. obviously, so there's all these different copy writing techniques that you can apply to prevent that from happening. right? so we have an angry customer. maybe you don't say I we close down your account because you did not pay your bill. you want to depersonalize that situation. So, you say well, it's the end of the month. we always got the money we didn't get it. So, we automatically shut down the account. let's figure out what happens, right? so there's all these different writing techniques that you can just apply when you're designing these conversations. So, yeah, that's sort of what we do. right? so the last slide here pretty much so we have assessments to know which is I’m quite excited about actually it helps us identify who everything your company could actually, potentially become a conversation designer. So, how much l feel for language today have etc. are they intimidated by tech or do to actually like it we have the curriculum to train people and obviously we consult and help as an agency we are. oh, I guess it's the alien. it's here. yeah, so we put together a bot week promo. So, if you just go to conversational academy.com, you can just sort of check that out and see the stuff that we're up to. there's about video 50 videos of me explaining stuff. yeah, so I think we're all good. you're waving to one-minute flag. there any questions, I guess. thank you very much. thank you, Hans. Are we got time for five minutes three questions raise your hands one? So, one. alright, let's go here. hi, I love that presentation. I am also kind of research designer and I also see a lot of great design practices for ivr that most ivr designers don't follow so I’m just curious. what's your background? did you come from voice / iv out? like most conversation designers. yeah, so I always wanted to be a novelist and so I wrote a manuscript and then the publisher said let's not do that. So, I became a copywriter and I start working with tech companies and I worked in an incubator have my own little startup that went bust. So, I got a job in customer service and then when chatbots became a thing it's like hey, I understand the tech. I know how to write dialogue and I understand the service pay. So, let me go and explore this and then my co-founder stay had a behavior design agency. So, they have all the background in psychology, and they applied that. hi, so when writing a copy for a chatbot versus a voice board, what are the differences in the conversation wants is that you would recommend the copywriters to take care of? it's a good question. So, what we do is design for his first, right? so we sit back to back from each other. if you can figure it out with voice then a chat bot easy. all right, so we start with the bare minimum and we even if we're just doing a chat, but we still designed for us first so everything that we make you can use for voice and then if you want to do chat, you can add a picture add a button make it easier right? it's also if you can't figure it out doing it during the exercise, but you know, it's not going to work for voice, but that maybe you can do it for chat, right? so that's sort of mechanism that we have in place. Hey, Hans, have you all heard of anyone or have you looked into building a user profile around a user like with a personality and sentiment on the user end as well? right? so a lot of times a lot of companies have that and we take that as input. So, if we look at you know, if you have the canvas will know okay, we're dealing with this type of user where argan guardian to kill is it likely that they're in the kitchen or on a busy subway and we'll sort of you know make that character as rich as possible. and we zoom in on one user persona for the design and if you have multiple that are very different from each other then we just sort of, you know, do one by one and try to sort of meet in the middle a little bit. thank you very much caregiver on a pastor Hans.