Keys to a Great Bot User Experience
  • Chatbot Conference Team

Keys to a Great Bot User Experience



Transcript:

For those who don't know bod copy is a web chat interface for dialogue flow. so, if you're using dialogue flow and you're looking to put it on your website bad copy could be really good solution for you, but enough about us going to go ahead and talk about some ux things and I use this click yet. Well, this is again like Dustin's that our fourth time speaking. so, it's going back to 2016 sides of head because at that time, we, as you know, pioneers were very interested in copywriting same one. Remember Poncho Jovan's Poncho. The weather bought Poncho's have nice fuzzy cat. We kind of fell in love with it for years ago really got us into the space. The point is it was a there was a lot of copywriting that went into it a lot of cleverness. And every one of the rooms was pretty much we went around the panel and said who's your favorite ux and everyone said Poncho hands down. Even from the big companies and Pancho's no longer around because the ux wasn't really a good use case. The copywriting was phenomenal off the charts, but the ux didn't really the use case didn't really matter in the end of the day. You didn't really need a very smart whether bot what so, we would constantly question ourselves. Okay, what really makes better ux and we'd play around with other NLP and nlu engines the time and we still do it's what we integrate with today we found in fell in love with Log flow because of everything we could do with it in context you could add to the conversations as you can see. Here's a pretty image of all the response types you can add. This is Vlad copies web widget but enough about really how we've come and where we've come and why we fell in love with NLP, let's kind of dive into some copywriting stuff right on the agenda. It says Dustin and I are here to talk about copyrighting and ux the first rule of copywriting is editing, as you already know? Okay, so, we're not going to talk about copyrighting you just heard from the greatest copywriter in the industry Hans. If you want to you know, if you take copywriting very seriously, you know go to robocopy. But if you walk out here with one message is that copy should be clear concise and upbeat. We spend a lot of time spinning wheels about personas brand DNA and that's all great in theory. What we've seen in practice with thousands of bots is really comes down to those three things. And those three things are not so, easy to master. You still need a great copy writer who's been honing these skills for years and years just to achieve that and if you can achieve that you're 90% of the way there what about like Governess in being cunning and funny and cute.


Whatever isn't that important. I I don't think that cleverness is that should be on the top of your list with AI. I don't think the user cares if your bought is clever. I think the user wants their problem solved fast. I think it's not so, important as important what your Bot says as what your user says. And whether you understand what your user says and can respond in a smart and useful way we've come a long way from Poncho where it was about. How can we be like Saturday Night Live, how can we be really clever? so, let's give an example of that. This is actually a real use case. This is our bot at bad copy and not so, clever but really a good listener. so, this is now in practice a user says what spot copy what the heck is your product the bot answers. The user asks a follow-up because the first answer had too much jargon in it. so, the user says in English. What the hell you talking about? I don't know what that means. so, we actually had a fall back. A follow-up that said sorry that that was too complicated here. It is in an easier way and then there's a follow-up to the follow-up and we'll get into that but it takes a lot of work. There's no way around that. I wish I could say Hey, you can just have this beautiful technology and runs by itself, but there's a lot of elbow grease involved and that's how you get to about that's personalized in our experience. And really so, how do you make the bot personalize? Well the way we do it what worked for us was nlu natural you all know this already. We're not saying anything new today. We're just saying in a different way. Hopefully that's really easy for you to take away. Is that a bot that listens built on a framework that is has powerful nlu technology behind it machine learning technology that can statistically score, you know where the query should be matched to which Time but the problem with these NLU and NLP’s things is there's a lot of this new jargon like you're looking at intense entities fulfillments context events training phrases actions in parameters. so, it's new but really how hard is it? I don't know how many of you know what these terms mean some of you this is old hat, but many of you may not know what actions and parameters are or how to use them. so, we want to make sure that you understand that this is these are the new things that you might want to Wrestle with wrestle with this year or hire people to do so, and that this is not that new. Dialogue flow actually is ten years old. It was apia. I invented 10 years ago and we still aren't seeing Fortune 100 Fortune 500s digging into these Technologies and mastering them and yet we've already shown that to have a personalized experience. This is what you need to master. Why because there's tremendous demand we're going to explain why there’s so, much demand. If you look here, we've done a little bit of a study and just call centers alone. so, I'll just highlight Texas here. We see there's 591 call centers and over two hundred and forty-five thousand employees at each one of these call centers and employees answering very repetitive things and go ahead and next slide in call center. Like this and to be honest these scripts aren't Shakespeare know now, they're not Shakespeare and this doesn't mean that we're not going to need humans. We're not think we're not suggesting that there's always going to be calls that require escalation to humans. But we're the argent we're making what we're seeing is that we're eagerly handing off the Baton to the human too early. And there's going to be fewer humans needed. Now the we're not up here to be political. We think there's a two hundred-billion-dollar problem. And if you do the math ask yourself how much of that 200 billion are you paying into this problem? And how much can you reduce that by this year. Now we've seen in the past two years like a lot of dilemma between the difference or we call dbots for smart Bots and the rasa presentation, you might see the dbots more of like the Click Bots or the decision tree heavy Bots. With one and two, but then he kind of got further into the three four and here you see on VentureBeat even this is past two years. Why are Microsoft and Amazon ditching their smart Bots? And then you see examples like this bonjour. How would you describe the term bought to your grandmother? My grandma is dead. All right. Thanks for the feedback, you know, just and here's another example from Poncho and I'm not going to read the weather in Brooklyn is 46 degrees and then the user says this weekend and Pancho says excuse moi. It's absurd so. was the best of the best at the time. so, like I said before we're handing off the Baton to early the human hand off Bots the template Bots the lead gen box. They're great, but it's missing. It's not chipping away at that 200 billion dollars were spending on live human support. so, how did what about voice like, is that going help contribute to this problem? Like we just launched a new voice feature with Botcopy. I sometimes say to Dustin voice is overhyped and he gets incensed. Because we love voice and we all think voice is the future. But sometimes we think that voice the new voice technology is distracting from the brain behind the voice and the Brain behind the voice matters too of course and this brings me to maybe you don't know who this is. This is Masuku the name of a bot that went 19 minutes and 40 seconds on a Turing test. You're already a I can mention so, I assume you know, what? Turing test is it's you know, if you can pass if you can convince a tester that you're human you've passed the Turing test it's a big deal. This spot was built using rules-based basic simple AI dialogue flow. It was actually Pandora Boxx but it was it was an IML Nothing Fancy. No neural Nets no Markov Chains No, bio direct bi-directional learning 20 minutes and a general context where you could ask Masuku anything and it had to convince you that it was human. so, then why can't a Fortune 100 with endless resources build a narrow use case bot that can go 20 seconds. The Builder that bot so, the dbots problem is they hand the batons off too early. You often get frustrated and a human takes over and that's really not chipping away and Smart Box are complicated. so, it's kind of the problem here. This is Ronnie Coleman. He's great bodybuilder one. Mr. Olympia 10 times in a row and he always said everybody wants to be a bodybuilder, but nobody wants to lift the heavy ass weights. Everyone here wants to be an innovator but nobody wants to live to have you as a waste not talking about you personally, but in general, you know Corporate America corporate corporations all over the world talk about Innovation roll up the sleeves live the heavy-ass weights dig into the harder Frameworks to personalize your Bots problem is people think it's too hard.

So, let's dive into that next slide. And jump into the fundamentals sure. so, here's an example. First thing we usually talk about is what's the media that you're going to be chatting with your customers? Is it voice is it this is a website example of companies using odd copy, but is it Facebook messenger? And why are you using it there? so, usually that's the first thing is that your application then it's where you're going to build. You know, there's a lot of NLP now platforms. We know dialogue flow really? Well, it's why you keep seeing it but there's rasa which is Using there's Microsoft bot framework. There's Amazon Lex. so, pick yours. I log for those omni-channel. It's great. Ross is amazing to next. This is actually a picture when we would help clients and we would really break it down. Okay, we want for your top four things that you're going to start with and really map it out. And then after you've really mapped it out and great programs to map out a readily available. You've got bot society bought mock. This is actually designed on software. You used to be called real-time board. You can't upload anything to anything, but it was just kind of fun to design and play around with so far. Just the first two steps. Have a place to put your Bot. We think it should be on your website. If you want to know more about why we think that talk to us afterwards. We have a lot of sound argents to back that up and then step two was map out what your users want to know start somewhere. Those were little seeds that are going to grow into huge trees over time. And then and then map out the intense like whistle. If you were to hire a person to do the same thing that your Bots going to do. What what's like, what's their job what's their role? What defines them doing their job bad and then create intense around that? Go back. so, like for instance, what is Bod copy? That's the event and they're trying to figure out what bad copy. There are a million things off of our response to that that you need to be thinking about that. They're going to ask to those are follow-up intense. And then the beautiful thing about these NLP and now engines that some people obviously no and others that are just getting started. Don't is that people going to ask all sorts of questions and bounce all over the place? And these Frameworks really allow you to handle that so, hold up one thing you see follow-up to the follow-up we've seen follow-up to the follow-up to the follow-up to the follow-up and this is fractal because it's not just one array you're talking about a raise it split off multiple directions, but then again, you're only talking about what is a copy. How much is Just you know, who are you guys maybe for things? so, that's a narrow use case. If you think of your customer support questions that scenario use case you might have to go narrow, but very deep if you want to get satisfaction, maybe not but you don't have to go wide if you go, why'd you go shallow? If you narrow you go deep follow up snakes. And in that way, you know again keeping contacts. We were here to talk about copyrighting which is content but we think that context is Trp's content. so, that's why we spent most of the time talking about it and then really look at these analytic platforms that are available. This is an example - bot we integrate with - bot all your conversations are going to be pipeline there. so, for instance in this example, this is again from the oxy go that was on the previous screen. Which is a health care application now, they can't actually sell the product. so, a lot of their problem was we're getting all these questions, but it's not really valuable for us to answer because we just have to ditch them to the insurance provider that can sell the product. so, we created a chat experience in one thing was is it covered by Medicare, you know something that Is obvious but at the time we didn't have built into the agent, but we can then refer back to these analytics programs and say, okay. Well, yeah, that's pretty obvious. It's put that into the let's put that into the training phrase. so, here we say, you know, we just added in and select enter and then all of a sudden the bot can recognize that we did something pretty tricky here and pretty elegant is that we just put a little blurb about Medicare into our response about pricing in general so, that way anyone who asks Pricing also, got the Medicare information killing two birds with one stone. so, we build very deep arrays for narrow use cases, we don't just make them up willy-nilly they're informed by the data and then all of it starts to come together. Like there's a lot of resources out there and really the biggest problem that we see at bot copy is the time that the account is created in the time that it gets to live on their website and that's because everybody's training and building and then when they deploy it never ends like you got to continue looking at it. If you want this truly to become smarter, every conversation has been mentioned before you can learn from and not just by training the bot. There’s so, much information from these conversations that you can pull and add to your product feedback that is most direct feedback you're ever going to get its customers actually asking or even potentially complaining about the product service or technology that you're offering. so, use everything wisely. You can close that Gap to where there's no more holes and you don't have to hand the Baton off in many cases and really think about think about how you're attacking this this problem. Like the reason you're building these Bots is to save Money on repetitive tasks or questions. Maybe it's because you have lack of resources, or maybe it's because you're paying too much for the resources that are currently doing the same thing over and over whatever it may be think about that and put a number behind it and keep training to attack that number right and we focus mostly on customer service the 200 billion dollar figures about how much we're spending on human customer service doing repetitive tasks reason we focus on customer service is because we clearly Need humans because we're spending 200 billion on it. We didn't focus on a different thing like sales or slot filling and that kind of thing because a lot of times you can get the same result through an app ordering a pizza, you know and robocopy brought up some good points like with ways to attack the problem like adding different buttons or thinking about ways to like three yeses. And that's all great. And another thing is sometimes web views like inside of our chat you can I click into a web View and sometimes actually forms in getting three or four questions out the way works a lot better than a conversation back and forth. so, we don't need to get sucked into the idea that we just need to keep writing and answering questions and going back and forth. There are ways to display things better. And sometimes that is with the few graphics and I think that's it. I think that's it. Thank you, guys, for sticking around and big hand for yourselves. Let's give him a big round of applause.

If you can add sky's the limit you can scale infinitely. That's right. We're going to do some QA. so, who's my first question going to be from? Vittorio, what's your question? You can ask him anything you want sure. Hi, so, what which is one of the things when you started to do copywriting that you to relearn for conversational copywriting so, something that was counterintuitive when you moved into the space, this is a good question. Good question for you Rob. so, Rob started as he's an award-winning add writer has written for Our can express wire all sorts of things. so, his transact like transition into the conversation only on space exploration like low-key. What's what do you got to think about is a conversation a rider compared to other compared to other writing? I yeah, I was a copy of her major agencies and I did a lot of TV radio print for big companies. And so, the quality had to be very high and sometimes in radio we do dialogue so, you'd have to have an ear for dialogue. So, some of that kind of transferred over when we started writing for Bots. But like I said and I kind of stand behind it is that the very beginning in theory. We went ran away with this idea of like brand DNA and in all these like Grand philosophies about the four basic temperaments of human moods and but then when we got to the practice part of it a lot of that went out the window, and we just needed to keep it really clear concise and upbeat and that's my take on it. As somebody who has a background in Liberal Arts and who's been a writer as whole life to whole career. I love writing. I love creative stuff and you know, but that's my take on it. Yeah, it's going on. so, could you comment on any differences? There might be for Bots that are deployed to a website versus say Facebook or slack or something like that. you write any differently or any other experience Facebook not have Facebook is a great me. Medi if they're on Facebook now two years ago when we were designing all sorts of things. We got a lot of Kickback from our clients because the Facebook widget actually takes you to Facebook if you go through mobile experience, it takes you off the website to Facebook. There are no branding tweaks to it. You're stuck within that framework. It was a huge problem for us. Now. This is great. If they were already on Facebook and discovered you through the Discover section on messenger or you had an ad and you click like and then the fought like added it's also, really good for re-engagement. But when people come to your website, that's like your domain we want very custom and simple widget solutions. so, that's what we moved into now follow-up engagement and one of the reasons we like dialogue close because then you can connect to it Twilio. You can do SMS You can pipeline it to email but we just saw so, much need and really want specially with companies using dialogue flow for white labeled widget, you know to take what they've built and put it on their website because they don't want to pass it to anybody else and in honestly, it starts a lot of customer service things. But we've seen a tremendous growth in HR and things like that. so, types of questions are a little bit different. Yeah, and I also, kind of get annoyed with Facebook Bots every once in a while, and it just kind of pinging me and feels a little spammy and there what used to be such a personal space now is kind of invading and Bots are doing that. so, I think there's an elegant way to do that and I'd be happy to talk with people about that but it is it is interesting. There was a lot of Statistics in the beginning that email open rates are Bots open rates of so, much higher than email which is true, but it's also, you know like fishing with dynamite because no one's used to about prompt you in this place. We're all used to getting these spam emails and just kind of ignore them. Right and don't open them when someone’s either SMS or in Facebook Messenger. It's kind of a newer experience of you. You're going expect a lot of open rates. so, that's slowly gone down. It's still higher. But yeah, the path of least resistance to a bot is through a website. We think right now anyone with a web connection go right to your url you as the brand you own that Nobody's, you know standing in between or making the user login. It's confidential we got I don't know if anybody's familiar with Swelly but we helped in the very beginning designed swell these conversations and I tweaked the whale into really cute fluffy whale it was times three Austrian guys trying to figure out how to write copy for 20 – 25 year old women. I was like English is your second language. Can I just try like add a little flavor in here and it worked like Swelly quickly blew up to 10 million users in like two months? It's like one of the most exciting things I've ever been a part of I'm not sure if Facebook's landscape is the same anymore. There's been a lot of privacy. We got almost shut down as a company when Facebook put a pause on, you know about development during the litigation last year. so, this is just all things to consider. Yeah. sorry one question. So, I think in the morning we discussed a lot about development and I think in the afternoon, we are talking a lot about design which is really good. so, yeah, I'm saying in the morning session. We talked lot about development and in the afternoon section, we are discussing a lot about design and writing. so, what I want to talk about is a design hand of which isn't technical way of making implementing the design. so, what is your opinion about an ideal design hand of what are the different artifacts which comes when you hand off the design to the development team? Okay, great and great question and it’s something that we're trying to help out with every day. I mean is great design resources as I mentioned about societies over here. I'm not sure Mach is still here. But those are great tools for your team to test. We also, integrate with Janice a I don't know if anybody's familiar, but it's a live chat that actually takes what you how you've responded and you can transition that to dialogue flow. so, instead of actually building out an entire experience. You live chat team can take it and then upload that to dialogue flow based on the Ponce so, there It's about that. so, as the live chat Personnel is answering the question. They're simultaneously training the bot to do it. Next time. It's pretty cool. Yeah, but designing is an art. I mean we're dealing with all these new jobs that are being created like conversational designer was in the thing for years ago. Either was like a conversational architect. so, there's all these new things being created and that's what it's part of the big struggle. We were talking some banking companies. I can't really name but they had a whole better introduce each line of copy to their legal department to make sure who's proved before they could perform like play it live. so, these are just new things that we're all dealing with and there's no right or wrong. There is great software out there to help with it. But yeah, it's not going to be perfect the first time either. so, just make sure that you're using these analytic tools to read and make it better. It's never finished. Thank you kidding around passing the road for dusting around. Thank you.

© 2017 by Chatbots Conferene TM.