Deep Dive into Bot Design & Copywriting: Dustin Dye & Rob Lubow



Rob evil and dust and I come on. Hey guys. This is our third year here and it's kind of exciting to be in New York each year just it gets more exciting because we're really all Pioneers in this space and have everybody here. There's people I talk to Edgar he's coming in from Mexico City Robocop. He's coming in from Amsterdam. It's just it's exciting space to be in we were bad copy bad copy is a software that has a rich web chat UI for dialogue flow. So we accept all Google assistant or Google actions response types as you can see there and we put them in your own custom white labeled widget on your website. So all sorts of cool tips and tricks we can talk about later.

This talk is actually going to be about conversational design and ux Rob’s and word winning copywriter. He's our co-founder at Bay. Copy so he'll be running most of this conversation. But yeah, it's who we are, right? Alex is kind of give okay. So we have deep roots in writing and conversational design is so we're very happy to be joined by so many conversational designers in the audience. We've also done our share of programming and engineering and reviewing data sets and we worked with Enterprise companies and now we're a software company so we can identify with a lot of people in the audience from the freelance writer who's kind of figuring out do I want to bring my talents into the bot space all the way to you know, like a Walmart that's trying to figure out a strategy that works for automating different systems and reducing costs and boosting Revenue, which is really why I think this industry exists having said that we're going to start all the way at the beginning of the story with the writing piece. I'm a copywriter creative director from the agency world where I used to write TV and radio and print and I was drawn to the AI space because I just love Ai, and I think how cool it would be to write for robots what I didn't realize Was it's not a linear thing. It's a tangential thing with digressions and fractals and it can get very complex.

So as a writer you're looking at a whole different idiom that you have to learn. So there's some things that bear repeating just some Basics about writing for dialogue for Bots. And I think there's a deck that's coming up here. So we'll just bring will blur through these really quickly because you know them already but they bear repeating. I think let's go to that bus one. Yeah, so, okay. We'll start with the audience. Go ahead.

And yeah, so Clarity if people don't know what you're talking about the bot breaks. So this is what we call them feebly I believe it's a kind of ambiguity in writing that the folder was on the bus, but now it's gone. Are we talking about the bus or we talking about the folder? Okay, so it's and the reason I point this out is we like again you can look at data and you can look at how people respond to your body and then change it later but your first generation bought the when you first put out their needs to have strong writing or else we'll never get the escape Velocity. I believe to get people to come back your most important users the ones who bothered to find you be the first ones there. So that's why good writing good grammar is important so that people know what you're talking about in this case the it's we don't know if it's the folder of the bus and when you're designing this kind of conversation like this sentence in particular when you're when you're thinking about what's going to be the follow-up like The user going to ask like what problems do you run into this? Scenario being a bus or a folder but as like a writer like what are you thinking about? The answer could be and how does that affect like your response you look at the sentence and you ask yourself reasonably. What could somebody asked about this sentence a reasonable question would be are you talking about the folder? Are you talking about the bus right? So if you have to use a sentence like this, make sure you have a follow-up that anticipates that reasonable request. In fact, we'll get more into that writing for Bots is about anticipating reasonable questions. And that's when people get frustrated when you can answer a reasonable question if they throw up an unreasonable question, you don't know it then it's really not your fault concision.

Okay. Now that we've got the subject of clarity squared away. Let's move on to the critical topic of being quick and concise. I've been writing for 20 years and I do this every single day. This is important for the writers in the room who are writing for Bots where the attention span of the users way smaller than any other idiom I've ever worked in as well as the people who are hiring writers. Make sure you hire the best writers who understand editing and concision. This is going to show up on a phone and it's going to look really like a big thick paragraph so writing for Bots is 50% as cutting it down use Hemingway.com. That'll keep your sentences really short.

And finally tone is a lot of discussion about what's my brand voice and you can spend months analyzing what the exact brand voice should be my experience. It's really clear concise upbeat. That's the brand voice to 99% of the time. Here's one two, three, four, five six ways of saying the same thing. You can't read it in the back. I know because I was in the back the whole time. But the first one it says this is this is a great idea Marilyn a meeting is a fantastic place to start. Please click the button below to select a time most convenient to you and then you can hit set up a call all these in the middle there pared-down pared down more.

Let's skip to this one. This one says cool. Let's get this party started set up a call. This one says no problem smiley face set up a call. Those are different tones of voice writer spent a lot of time thinking about which tone of voice makes sense and in this moment, that's all they do character and dialogue writers. So, that’s a long-winded kind of Gadget E1 and this is a really short one. He's the kind of decisions that you'll have to make and see what works. So you've done work for American Express for wire or all sorts of different brands.

Like what can you say in the bot space in particular where you're looking at an audience? You're looking at who you're speaking to and when you approach that is it was a traditional ad but how do you take the same like thought process and concept and put it into like a I So we have a client if I copy right oxy go. It's just an oxygen portable oxygen machine. So we know that our Market have trouble breathing chances are they might be elderly or struggling some kind of disease so we know that we have to we can't use younger. Jargon. We can't take kind of like a smart alecky tone. These are people who are struggling with breathing and they want answers and they're frustrated and they might be cranky and I even knows I get older I get kind of cranky and I talked too much and I asked questions and I get paranoid and last night at the hotel. I was yelling at the woman who checked me and because screwed up my reservation so you can kind of expect that with oxy go you're going to get all these questions. So, you know, you have to know your market and then modulate the voice to be warm and fuzzy with this Market with another company.

Let's say we're working with HP, you know, and it's Enterprise we're talking to Geeks, you know, we're talking to people who want to hear all the tech stuff and they want to get right to the point. So it's pretty similar to any marketing brief. Yeah, we tried with oxy go kind of being fluffy at first. Taking like the cuter and charming like let's be nice and help these people get the medical devices. They wanted and the results weren't that good, you know, we launched it. The conversion rate was relatively low and we quickly changed up all the copy to be very Punchy. Like we I think the initial call to action for the bot prompt at the bottom was like, you know here if you need us like happy to like help with like a cute little smiley face like it was breathe easy. We're here to help and it had a face.

It was a woman's face look kind of like a doctor. It's a breathing apparatus. So breathe easy was kind of a pun. It didn't work, you know, so we changed it up like get answers now, you know, we're here that worked. I really well. So sometimes you're going to you're going to launch and realize that even though you spend a lot of creative time building whatever you're building. It might not work. So using analytic tools look at it and keep reiterating and revising and don't think that whatever you launch with is going to be the thing that is stick with it's not I'll go ahead and take this one because to me it's the most important in 2016 when we were playing primarily with Facebook Messenger experiences and just getting into the space.

We're I realized that there's this huge void an opportunity for Creative writers to come in to a how cool who the hell gets the right. Cirie who's writing Alexa, you know, there's some opportunity there, you know, so we quickly started bot copy to kind of capture that market share and we started we wanted to do everything, you know, and we started doing these builds and had just wanted to make it fun and really had no objective in the beginning. And that doesn't work. So don't do that. You know, what we had then started doing is every client that we would work with. We've look at specific kpis and every single conversation would be measured side by side with you know, how it worked prior to launch how it worked, you know during launch and these you can get creative with these kpis. There's a lot of analytic tools that we will talk about in a little bit but don't just don't just base it on the number of conversations or how long you know, it went for there's really particular things that you can look at. You know one was we went side-by-side with the call center and we had we are measuring how many calls you know came in.

We thought that was like the kpi to look at. Okay, like how many calls are we getting before? How many calls are we getting after and we realized that they were relatively the same and we were pretty frustrated. It was until we looked into the calls and realize what they were calling about and how long that they were chatting for then and we realized there's actually a way bigger success than we thought, you know, the calls were like they wanted to purchase and they wanted to purchase the human they only had one or two questions all the fa q's and all that shit was kind of gotten taken care of by the bot, you know, so it was a win but we kind of had our kpis wrong just comparing like number of calls that were coming into the call center. So spend a lot of time here because it's easy to do just want to get creative and write something cute and fluffy and fun and answer a bunch of questions and if it's not tied to something your creative efforts are going nowhere. Cool, and that's where we kind of get in. This is blurry. But this was this is a map of just an onboarding experience. Now, it had a lot of decision tree Basics and functionality in the beginning but it's where we kind of put a turning point in our company where we realized that if we really Want to get creative if we really want to Leverage The Power of good writing and where Bots can we need to start turning to NLP platforms.

We need to start looking at not just copywriting and creative but context and that's kind of what the NLP and if you want to so yeah, this was kind of when we evolved from focusing on text or copy to context this. It's hard to explain but even what you're seeing here is only the scratching the surface of the actual full architecture of a conversation. But you know, like as I mentioned before bot conversations are tangential and nonlinear and so while I click Bots are why NLP over click Bots? Well, okay, so you would think that if you if you created a button bot with a multi-level choices people would behave and they click the button but we found out that roughly 60 to 80% don't they'll type in a query and it's always a surprise in query and we don't mind if it's a query like what's your favorite color? Because that's not that's not relevant to the use case. What gets us feeling guilty is when someone types in a query that is entirely relevant to the use case and we can't answer it. That's when people start getting frustrated and leave so we had to have a way to do that and NLP. Engine like dialogue flow for instance was the only way that we could attack that problem. Here's kind of a better example of that. This is when your Amazon articles, so if you want to take this one as well sure well, so this is about that we built and It's like what is Bod copy? Okay, so take us for example, we went through this ourselves because this is our bot we built on dialogue flow and we describe what bad copy is it lets your Google dialogue flow agent run on a rich featured front end web chat that puts you in control.

Okay, there's little bit of jargon in there. So if one of the users actually wrote in English, you know being a smart aleck like I don't get what you're saying. So we see that that incoming response and we're like, oh they're right. We need to say this clearly. At least we have to have a backup in case someone doesn't get it. So we go back and we create a follow-up for in English. And by the way, 200 ways of saying in English like I don't get it or what does that mean or see it clearer? Or you know, what a dialogue flow is or what's a why does it what's a front-end web chat. So we train that all those phrases to trigger this my apologies in simple terms.

And then it goes on to explain it in simple terms. So no one said this was easy, but that's the backup in case so the way we do this is we mentioned the word widget and this one here. So we have to highlight we have to look at each single word in this intent and in this response and ask ourselves could this word bother someone could it prompt someone to ask a reasonable question and we have the word widget in here and we actually thought someone might not know what a widget is. So let's create an intent for what's a widget in case someone says what's a widget and sure enough someone did WTF is a widget. And we had the response ready to go. Nobody's asked me that before I guess I'd say widget is a self-contained many program that lives and runs on your website app. So we're very picky about making sure that we can answer any reasonable question and you can mathematically figure out what all the reasonable questions are there to embedded in your response by going word-for-word and asking yourself can this Elicit a reasonable question and then programming follow-ups and intense for those polyps and even follow-ups for the follow-ups in order to make sure that your body is bulletproof and can answer every reasonable question. It can be done.

Don't let anyone tell you it can't be done or that you have to take a shortcut or that you shouldn't do really long follow-up arrays. If you're if you're a new conversational designer pushes far as you can put in those extra two or three days to make your Bot bulletproof instead of you know, saying, you know what that sounds too complicated. We're just going to give up and go, you know, just send them to a you know to an FAQ or something. I think that we often hear like we have people call us every day and we hear their counter argument with the click bot solutions that are out there that you know, what the fuck is a widget. They don't they don't care like why don't we just we want them to get down one particular path, you know, we don't want to put up anything else. Like here's the here's the kpi like hit that why would you want NLP to go outside of these? Well, if someone says what is your product you try to answer as best you can you want to answer in a way that's not going to elicit somebody's confusion. But so some of your responses will have it's a necessary evil. It's gonna have to be unpacked a little bit Yeah, but from a copywriting standpoint you want to get good at writing responses that can't be unpacked and that kind of drives somebody into the next question.

You want them to ask so you want to you want to Loop them back into that. Kpi eventually for sure outside, you know have a reasonable response to get them back in right? And how do you go about doing that without you know feeding them on and on forever and ever? Because if you answer the question about, you know, WTF is a widget then it's pretty reasonable that this person just ask the same thing twice, then it might take another context out of this sentence and then bring it back. So what are some like design tricks to get them to come back in? Well this last one it's pretty much done. Like there's really nothing more anyone could say about that. So then the person wrote got at what else she's pretty open-ended question, but we have and it we have it in 10 trained for that got it. What else is one of the training phrases for what can you do or what are your sir? What is this bot do so it triggered this intent? Please ask me about Val copy. I can talk about our product.

I could talk about Ai and almost intent most responses end with that sentence. Please ask me about copyright, and when you see in this kind of scenario, so we offer the ability to type, you know, so they could go off the rails and do this, but we've also have like suggestion chips and all sorts of other things that Google provides. What do you see in being more successful? Let's say that they ask this, you know this example, we didn't put a suggestion chip. Wanted to be pretty reasonable to say. Hey, I hope that answer your question. You want to get back to you know, maybe giving us your email or setting up a call, right? That is if that's the kpi to get their email. Yeah. We'll say that's all I know about this subject.

Why don't you give me your email and we can talk more and be ready for them to say no. So we've seen, you know people start at I'd really two different spots, you know one is they've got some really complex and LP built out and it's up and running and get it up into two or three clicks and then there's they just they're full call center still and they have nothing and don't really know where to start so usually we look at that from like, okay, you have a hundred percent live chat on your customer service side, and you're trying to get to you know, 25 percent or 10 percent. I think it's hard to say that it's going to replace a hundred percent. It's always good to train your AI based on you know what you're able to facilitate to your live chat, but that's typically how we look at it, you know make a curve understand and back to some of those objectives and things we were talking. About know where you're trying to go understand how your live chat call centers who are however, they're communicating look at how they're answering some of those things and always make it so it's if your company whether its internal external chat facing make sure that you're designing and training on cost that you're cutting, you know, profits boosting and delighting customers in some way because you know customer retention is huge if we can keep our customers longer. It's easier than closing Nuance. So chat Bots are amazing for doing that leverage them for that but be very specific on how your training and don't try to build too much in the beginning or you're going to get overwhelmed. It's not going to work.

If you're presenting it to like one of your project managers all sorts of people going to ask it all sorts of crazy things and you're not going to come back. So to kind of wrap things up. This is a one of the last slides one of the most important things that people Overlook is the top of the funnel. It's not just the chat experience. It's also where the hell is the chat experience, you know, is it on your website is it clicked? Is it triggered through a QR code? You know, how are people interacting and get creative with that? Because they're not necessarily going to get going to know to go to your Alexis kill or your Google home or on your website or your Facebook Messenger pledge. So play around with some of those each medium is very different. You know, Facebook Messenger is cool. But putting Facebook Messenger plug-in on your website.

I'm not sold on, you know, having mandating people log in through Messenger just to chat with you and they came to your domain first. I don't know. I think there's a little bit of an issue there. There's plenty of solutions where one of them you can have your own native live chat on your website. And have two different mediums, you know, you can have a web chat with a bot. You can have Google Assistant. You can Facebook Messenger but understand why people are interacting there and what they want to accomplish because I've seen I've seen a lot of different builds and each one is each medium is a little different you answer a call differently than you answer an email. So understand that when you're designing a i as well, can I see and most importantly as we wrap it up again the very specific on your chat goals.

Otherwise, you're going to build something with no kpis and she's not going to work and make a quick prediction just to wrap it up Dustin's a great CEO and we sometimes disagree. I don't think there are that many other options to what we're doing. The prediction is that there's going to be a new generation of conversational designers who are going to master Frameworks, like dialogue flow with intense entities fulfillment context. That's to me the best way to make sure your body Break, I think brands are going to realize that they want to be on websites because it's an easy place to go. You don't have to log in. You don't have to be a member. You don't have to download an app. You don't have to pay data charges on your phone.

It's an open Channel with no friction. So why not go talk to a Smart bot on your website. That doesn't look like a Facebook Messenger plug-in, but actually looks like your brand and so from our journey as writers all the way to software developers. That's our prediction. There are other great companies. If you're not looking for a complicated bot like you can go to something like a smart Loop which is a fantastic product. We're big fans of rasa, but that's my prediction connecting NLP Bots to websites. Cool.

I'm going to wrap it up and we have time for questions or cool. You guys want to take one? Steal my Michael show you all right, we have a so it's funny. When you see different questions. Some of them are like this some of them like this. And this question watch. This is what it look like. That means it's very thought through it's very I have a high hopes for this question so you don't think about it, but I judge you by the way that you raised your hand. Here you go.

That's a big buildup. Okay, so you've mentioned the live chat and I was curious in the way to write copy for the chat bot versus for the live agent. So if the implementation allows a handoff from about to and agent do we need to look at how agents are responding and maybe make tweaks there? Is it the same Twix or is it in the same Spirit as how the bot would respond versus how the agent does that's a great question. I think told you was a great question. Sorry that question is is understanding how good agents are responding because everybody does kind of respond differently most sales agents are reading off of a script still but they do go off the script they do it. See things differently. So most of those calls are being recorded for quality purposes.

Most of those companies never listen to those calls. It's a really good place to start for instance right here when we're training I pulled this up because we're looking at this is one of a client. This is red box. So basically Red Box had a whole bunch of fa Q's that we were looking at a whole bunch of costs and we were able to analyze based on how much time each agent was spending on like the number one FAQ was my disc is scratched. How do I fix it? And they were getting that like 40,000 minutes a day. So that was a huge bill for them a pretty easy chat Bots to build actually to you know, how to facilitate the fix on that, you know, spit the disc wash it on your thing. Put it back in and see if it works but we were able to say okay, you know, let's not predict that the AI is going to take out like a hundred. Percent of that but let's just say 60% How much is that a cost savings to the business? And how did how did that bot do it successfully compared to the live chat because if you launch it and you have a 60% like the AI says here's how you fix your disk step one step two step three and sixty percent of the people.

Don't ask any other questions it works. It's a pretty big win because you know, the AI savings, you know $7,000 on enough. This was per day or per month or what it was but look at those types of numbers and do start with the transcripts that you get from the live chat because it's pretty valuable if it's already working. Any other question? Why is Mike not working the answer is it wasn't turned on? We're in the front somebody who's excited about accelerating their growth and the front row. I have lots of questions. We found that people were kind of missing that are bought was a bot. So one of the things that we're trying to do is transfer or move from the language of I from bought in bot responses to the language of we as a company have you guys seen any other common pitfalls from just unintended consequences of specific word choices like that? Well, yeah, we're a big fan of we language. But if you're trying to establish Rapport, then I let you switch to I language if you're trying to establish credibility and ability to you know, perform a job as a Then it's we language.

So it just depends on the goal of the bot other pitfalls of language. So just a caveat State of California. You actually have to declare that. It's a bit like up front. So this is sometimes that works, you know, sometimes then you can take the bottom on as a persona but I like that it seems how is it working Kevin to this? What are you happy to take a look see how natural it feels sounds? Okay, so I have an exercise for everybody because we all should work out everyone take your name badge and remove it flip it around 180 degrees. Take out your phone on the other hand. Take out your scanner scan the back of your card ready. It'll be like competition who can do it first ready to go.

Scan what you are scanning is a bought built by bought copy for the chatbot conference to give you an idea of some of this conversational design in action. Scan it start the conversation scan it scan it good. I'll give you raffle ticket. I promise and again what I think is interesting from a how many non-tech people. Do we have in here? Okay, just you then. I'll just oh, they're too busy scanning. See it takes us longer to do these simple things like scanning, right? What's interesting? What I'm seeing is that you have the back end. You have the middle end and then you have the front end.

And so what I'm seeing because I've been to these conferences to be at what point you leverage pre-built information to patch in what you want with other information. So I think bot copy is an interesting example of where you can plug and play in this larger scheme so you can play with that bot with a chatbot conference and I challenge you to challenge the bottom to see where the conversation goes. All right, give it up for these two people right here changing the world one body at a time whether you bought it or not.

© 2017 by Chatbots Conferene TM.