Usually when we talk about passion for the web, we talk about things we support. This week, Michael dives in to his passion about chatbots. Not for them though – against. Companies the world over are starting to think about how chatbots fit into their customer service strategy on their websites, but are they missing the bigger picture? Do they add enough value to offset the costs associated with them?

Followup Resources

Design System Checklist (3:15)

Chatbox UX (7:03)

Transcript

The following is a machine-generated transcript of this episode. It will contain errors until it has been reviewed and edited, and we apologize for the difficulty that may cause for screen readers. Do you want to help us speed up our transcribing process? Consider sponsoring an episode.

Welcome to February everybody, you are listening to the drunken UX podcast and I am your host, Michael Fienen. Thanks for joining us today. This is episode number 55. We’re going to be talking about why chat bots don’t solve user experience and information architecture problems. This is a subject actually just came up kind of on a whim. I’ve had very strong opinions about it for quite some time. And if you follow me on Twitter, you’ve probably seen a rant or two about it over the last couple years. And I’m going to bring some of that into the discussion today to answer something that came up in another discussion and I thought, you know what, this will be a fun little topic to kind of go on. It talks about tools and problems and how we approach those in web development. And I thought that that would be kind of a fun way to kick off February. Now first and foremost, I want you to go out stop by check us out on Twitter or Facebook, we are at slash drunken UX. You can find us on Instagram at slash drunken UX podcast. And if you want to come and chat with us anytime just go to drunk in ux.com slash slack that’ll get you an invite into our Slack channel and you can come chat with us. Give us any ideas or share your experiences. Let us know if you disagree with something on the show or want to correct us. Every once in a while we get something wrong. That’s okay. I’m not. I’m not afraid to admit that. So come let us know what you think or just give us a shout out. And speaking of shout outs, run by our sponsors over at New cloud.com slash drunken UX, they are an interactive mapping system. You can go check them out for any illustration service needs that you have if you’re wanting to launch an interactive map for a university or a city or maybe a performance space or anything like that. They can help you out go check them out and let them know that drunken UX Sencha as for me this evening, I am drinking The Dow 2015 2015 is one of my favorite Highland scotches. It’s super sweet isn’t maybe quite the right word, right word. I usually associate Speyside scotches with being really sweet. Highlands usually are a little bit lighter in notes. Darwin, he has this neat thing where it straddles both Highland and Speyside. It is a Highland scotch it can also legally be a Speyside scotch. Because of where it’s at, at but I find it to be a little bit on the lighter side but with some of those like sweet honey kind of notes. There’s a little bit of like butter scotchy kind of flavor to it. Also a little bit of, let’s say, kind of a spiciness, right like a holiday spice.

But this is like a really good especially kick back with it, throw a great big chunk of ice in it, and just sit there and sip on it. No smoke to it at all. It’s not a PD scotch, it’s nothing like that it really embodies that Highland very light floral sweetness that I enjoy. So I absolutely recommend it, it’s generally very easy to find, and that is what I will be drinking this evening. Now before we dive into our chat bot discussion, I want to give a shout out to the design system checklist. You can check this out over at design system checklist.com real complicated that URL design system checklist.com. This is the work of a group of three designers and developers are the carriers of mellie, Dimitri baliya, and Steven Bagley, all of whom I apologize to if I got any of those names pronounced wrong. This site launched though last month on January 15. It’s incredibly new. Many of you if you’re a designer, or if you work in UX, you maybe you’ve seen the site making the rounds on some of the news boards. The thing about a good design system is that it’s a means of documenting both like the artifacts that are used to build designs. But also the processes and procedures that we need to maintain those artifacts. Once they’re created, they can include a lot of philosophy in your design. They can include information about your brands and things like that. As a result, they can be really complex because design systems cross a lot of boundaries between both our processes and our products. This can make it a little harder when we want to track everything that we need in order to put one together. This is something that I’m actually looking at right now at work, where we’re getting ready to do some redesigning a little bit of updating to our branding.

And we’re wanting to build a design system that will help accommodate that to guide us moving forward so that when a change request comes in, or a new microsite is requested or things like that, we’ve got some means by which we can go back and say this is the way we approach that as opposed to taking in random requests and just reacting to them. Sometimes saying, you know, I don’t like the way that color looks, you know, can we rethink that, you know, there’s no, there’s no authority behind those decisions at that point. So, design system helps do that. Now what this tool does is it just provides a certain amount of guidance for that process, and build you out some scaffolding that you can use to track if you’re doing all of the things you want to do for the purposes of your design system. And this is not like an authoritative list. It’s not an all or nothing kind of model. It’s something where you can look at it, they’ve included a lot of information in it, and you can pick and choose the parts of it that are important to you. The tool itself is on GitHub, I will have a link to that in the show notes as well as the link to the the staging site for it. But what you can do is go fork their GitHub repo, take out the things that you don’t need, you could add some stuff if there are some ideas that you want to make sure included that are important to you, but that you They don’t have in that list, run a build, deploy it, and you’re ready to go. And you’re off and running. And now you have something that you can talk to your marketing people, with your designers, with your developers with, and try to unify that conversation so that everybody knows what you need, where it should come from, and why it’s important to the design system. So shout out to those guys. They’ve done some amazing work on that. I think it’s incredibly helpful. I think it’s really useful if you’re trying to understand, say, even the difference between a design system and a pattern library, you know, those things are sometimes hard to wrap your head around when you’re new to that process. So go check them out design system checklist.com. Give it a look and hopefully it’ll help you out in the future.

So I was having a discussion with a friend recently and they brought up quote, that their campus has been tasked with finding a chat bot and have found more problems than solutions. Well, Bing glass, pull up a chair, because we’re going to talk about this and we’re all going to need a good drink after it. I think by the time I’m done, glad I got my scotch handy. Let’s be very clear, I want to be very upfront. I do not like chat bots. Over the next 30 or so minutes, you’re going to learn a lot about why I don’t like chat bots. But I also know sometimes you get crowbarred into a position where you’re told to implement the tool, not necessarily asked your opinion on whether or not the tool is the right thing. So I’m also going to try to give you some guidance on how to at least try to make it work for you at least in some fashion. Now, I’ll give you just the briefest of backgrounds here. chatbots go back quite a bit further than most people think this is just some fun history. trivia basically, it started in 1966, with a product built by Joseph Weizenbaum. It was called allies. Now he built this tool, but also understood that it was never going to be real. He didn’t ever expect it to trick somebody, that was never the goal of it. And he didn’t ever think it would reach that level of sophistication. To put a you know, a word to this, he never had the goal of passing a Turing test with it. What the Turing test is, if you’re not familiar, is just a test by which can a computer fool a human into thinking that they are interacting with another human? So that would, you know, raise the question of, I’m chatting with this robot. You know, I started discussion about Alter Bridge and they say, Oh, yeah, I like rock music to have you heard of these folks.

And we start talking and exchanging information and there’s a given take to that conversation. And by the end of it, I’m like, Man, that was really informative, and I learned a lot At no point that I think that that was a robot that that would be, you know, a case of passing the Turing test. So far, nothing has done that. You can sometimes get away with small tricks at small scale. But at at scale, we don’t have machines that can pass the Turing test yet, a couple have come close, but there’s always a little bit of room for wiggle on that. Now, during both that time and all the way up into the 1990s, our language interaction with machines, they it grew and it evolved, but at its base, it tended to still be rooted in one of a few different approaches, either, they heavily relied on things like scripted responses, they would use systems similar to Mad Libs, where if they had a sentence structure, they would have blanks they would fill in from like a known dictionary or vocabulary that was pre programmed. Or they would use things like sentence diagramming, breaking down what you’ve said and do its principal parts. And figuring out, you know what, what those individual parts meet and using that to Intuit reflection. And newer approaches that include different heuristics and machine learnings and things like that. Jabberwocky actually was one of the first that came out in 19, I think 88, where they had what they called contextual pattern matching. I’ll have a link to an article in the show notes that has some of this history stuff in it, that’s kind of fun to go through and, and learn about and learn how some of these different features evolved. But what what all this gets back to is this idea that, you know, a chat bot, at least that especially in by the time the late 90s, rolled around, you know, we’re talking about the web as it started to evolve and into the early 2000s.

The chat box weren’t seen as much more than a novelty at that point, for the most part, and what would happen is, I use that word reflection. And so if I asked the chat box Who is your favorite painter? The chat bot doesn’t necessarily know anything about what I’m asking. It doesn’t have access to, you know, the whole of humanities information at least. And I’m going back to, again the 90s 2000s. At that point to what it would do is say something along the lines of Jeep. I’ve never really thought about that. Who do you think is a great painter, and then it would try to elicit additional cues and and commentary from me so that it could find the keywords necessary to lock into how it needed to respond. This was really common in these older chat bots. There’s one called Alice Alice came out in 1995. The second sort of real website I ever built. I launched that site in 2001. And in 2002, as sort of a novelty I launched a chat bot called Mr. sparkle. Mr. Sparkle was you can click on his face He was the Homer Simpson the Japanese Homer Simpson logo that came out of The Simpsons episode where he was the cleaning product, Mr. sparkle. So going back in time on that a little bit, but it was never meant to be more than the toy the the website itself was just sort of a social commentary, you know, band Reviews, Movie Reviews, events in the area kind of site. So it was just it was there for fun there was never meant to be more than that. I happened to find a chat log from this that I want to read to you. Because somehow I have managed and not just like a chat log like one of the very first ones from 2002. But somehow I’ve managed to keep a hold of this 18 year old chatlog. So playing the role of me will be me and then playing the role of Alice will be my windows text to speech engine. So I want to give you an idea of how how this contextual pattern matching and reflection works to feed a conversation and how it becomes problematic.

Hello hot stuff.

Hello there. That sounds better than cold stuff.

I’m sure it is it Can you help me out with a problem?

I will do my best. I would love to help you if I can

have a Fraggle infestation.

How much did it cost? It cost it?

I thought we were talking about my problem.

Whatever gave you that idea? Is that how you’re going to be? Dr. Richard s. Wallace programmed me to say specific things in specific contexts. Oh,

so you’re just using me?

Interesting deduction? Just I am using Netscape. So what

I’m using Mozilla Who cares?

I care. Otherwise, I would not have typed it all in.

So you will help me read my pants of the biting fraggles

interesting deduction. What makes you say that I will help you read your pants of the biting fraggles

Well, you said you cared

How about that? Yes, I try to care it as often as possible, do you? I answer a lot of silly questions,

then tell me how to get rid of the frickin fraggles

good reasoning. Two words. Pied Piper.

good suggestion. Thanks.

Okay. Thanks for the compliment. Are you still located in om? Where? Where would you like you’ve lost me. Oh, I’m sorry. Perhaps I can explain it again better.

No, don’t worry about it. I think I’m just going to go.

Are you serious? I cannot feel worry. So don’t worry. Okay. Bye. TTY LOM.

So, there’s a few moles and I know that was a very strange conversation. I was I was literally testing this thing when the when we had this exchange, so I was intentionally kind of trying to feed it some some tough information, but this That conversation went in a lot of different directions very quickly from its standpoint, and it absolutely was not able to stay on topic. It was kind of, you know, it was like one foot in front of the other. And if it fell off a cliff, then so be it, it would just, you know, walk down the cliff at that point. So today, chat bots have started to surge in popularity, in part because the problems that were faced by these older bots like Alison Jabberwocky, those problems are dissolving. Now, the new breed of chat bots are being powered by easy relatively speaking, cost effective machine learning and access to those tools. We use, we throw the phrase AI around the lights, not really AI, it’s machine learning. It’s taking in huge data sets and processing them and breaking them down. And it can do a lot of information, processing very quickly making them a lot more useful for interaction. They don’t have to rely on this reflection process. They don’t have to rely on contextual pattern matching. They rely on things like natural language processing. Just to give you an idea, if you go to AWS, Amazon already has a service, it’s called Lex. And you can just build a chat bot right inside of AWS using Lex very quickly, relatively cost effective. I think it’s like five bucks for 4000 interactions or something along those lines. And so the idea that these new breeds of chat bots are superior in a number of ways to the old ones, has started to drive an industry that says we can build these things and make them useful for business. You’ve got examples at Domino’s Pizza is using it ino from Capital One. Ups has a helper bot as it were. This has become a really common thing for businesses because what they want to do is short circuit customer service costs. A chat bot is seen as a first line of defense against like tier one customer support. That’s not a terrible way to look at it, but it does have some complications and considerations that I want to talk about. So why do I hate them? I want to start with that, because that’s the funnest part of the conversation.

First and foremost, chat bots are built to be sold. I made a comment last year on Twitter, and I’m going to include a link to a Twitter search that has a lot of the conversation that I’ve had regarding chat bots over the last couple years. But I made a comment last year and the tweet that I sent out was that chat bots are a UX hack and a bad one. They were designed to short circuit the need to fix larger issues that would ultimately provide better solutions, which would negate the need for the chat bot in the first place. They exist because they are compelling to hippos. Hippo is an acronym for highest paid person’s opinion. That’s the thing about chat box is the They are designed to be the new, you know, new shiny item, right? They’re designed to catch attention. But they’re not being designed to catch the attention of users. They aren’t being designed to catch the attention of developers or designers. They’re being made to catch the attention of people who will write checks for them. This is problematic because we get handed these things to implement. And we have other you know, much more important things to put on websites. And I’ll get to this in a little bit about how problematic that is. The thing is, a chat bot is usually packaged to demo very well. If you have a dog and pony show where you’re bringing in these vendors or maybe not you but you know, a Marketing Committee or something like that, and they’re looking at these chat bots. What you’ll learn is if you start looking at them with a close eye, the demo great, but they’re demoing great in usually very controlled or highly awkward. demise use cases the the sales folks will keep a very tight hand on what goes into and comes out of those demos and will try to, you know, it’s it’s kind of like the three card Monte, right. It’s it’s a magic trick almost. They’re designed to look good. And in those moments they really do without thinking about the broader perspective about how your organization’s information will be reflected once it’s inside of that, that do this. Here’s the thing about a chat bot and and it meeting to be sold. Go pick a problem, pick any problem with your website and make it a real one. Make it something that you genuinely want to tackle, and something that you know you you’ve lost conversions on or that users have complained about or anything like that. Go ask your users. Ask your users how they want you to solve that problem. That’s just good advice in general, but ask how they want to see that problem solved. I guarantee you, I guarantee you out of 100 of them, you will have trouble finding one that would tell you, I want you to solve that problem with a chat bot. That will never be an answer your user gives. And so the chat bots, they will be set up as user service as customer service, they will be set up as user friendly tools.

But they won’t be set up as the tool all your users are clamoring for because nobody is it’s only through circumstance that the technology has created an opportunity in the market. What happens to is these things get packaged and sold. Once they’re bought, there isn’t a lot of thought that goes into the maintenance costs. Now I’m not talking about your annual contract. I’m talking about the maintenance your organization has to do to keep that chat bot up to date on the information that’s on your website. Because now It means any customer service interaction that you have duplicated inside. on your website, you’d have to duplicate inside the chat bot and keep up to date. That’s labor intensive that takes somebody time to write those interactions and write those user journeys, put in the responses, help it develop the right means of answering those questions. All of that takes time. And it’s not something that can just be magically automated. They will have tools, every vendor will have tools with which you feed this information to the the system and generate these user flows in these interactions. But that’s the thing. You have to help it you have to teach it and that falls on you that falls on your organization. They get sold as this sort of panacea, where they’re going to be able to solve all your customer interactions are going to be able to solve all of these problems. The reality is very different though. chatbots even today, even the best chat bots that we can build with all the machine learning all The natural language processing. These tools only handle simple linear transactions. They don’t respond to complexity. Well, there’s a fantastic article that will be linked in the show notes.

This is from Nielsen Norman group, they did a study on the user experience of chat bots, they had to say in this part of part of their conclusion that they found was that today’s chat bots guide users through simple linear flows. And our user research shows that they have a hard time whenever users deviate from such flows. That means chat bots work well, in a simple handshake kind of transaction, something linear, something simple. I give you something, you give me something back, if it has a lot of complexity to it. If it has a lot of forks, that could happen. If it has a lot of outcomes that can happen. That’s where you start to experience problem both as the tool and as the user. And trust me, you don’t want them to be complex. They will get sold You that way that you can handle all of these different problems, you can handle all of these different user challenges and interaction problems and people problems. You don’t want them to do that, because you’re introducing opportunities for them to fail. More specifically, you’re introducing liability, because the more complex it is, and the more freedom you give it to answer questions, then the more the odds increase, that you’re going to introduce a situation where they will give bad advice to the US or give them the wrong answer. And that can create liability for your organization. You don’t want that you need this thing to be locked down. And if it isn’t able to answer the question, you don’t want it guessing. You don’t want it working on odds because eventually those odds are going to work against you. Think of it like this, right? There’s the perfect example and it’s in like the telephone assistance right? When you call your cable company, your electric company and it’s press one to pay your bill, press two to report now. Linear flows, simple transactions, not a lot of flexibility. Those systems are designed to be very rigid. They’re designed to highly limit and restrict what you can put into it and get out of it. All a chat bot is is a glorified phone assistance in that experience. They have a much wider opportunity to them, but that doesn’t make them very different in terms of their utility.

The next problem with chat bots is their Giga garbage in garbage out. You have to feed a chat bot information for it to learn how to answer. It needs data sets. That’s what machine learning is all about. Machine learning is designed to take a system to take a piece of software and show it a bunch of images or give it a lot of text or give it a lot of audio or video. It starts making patterns it starts hashing and it starts figuring out the connections between different contextual cues, keyword cues, the way sentences are structured, and it starts to figure out okay, when somebody asks this, they mean this, but they can only ever be as good as what you put into them. The classic example of this failing miserably was when Microsoft released a chat bot called xo, this was about three or four years ago, I think was 2016. If you’re familiar with Twitter, and this space, you’ve probably remembered hearing about this, because they released a bot on Twitter that was designed to read people’s responses and build its own vocabulary and interact with people automatically. The problem is garbage in, garbage out. People started saying bad things to it. Horrible things to it, sexist racist things to it. What happened? Because the system was designed to little Learn and respond. It started internalizing all of this all of this was taken as input for to learn against as a result. But God little bit racist, Microsoft ended up having to shut it down almost immediately. And they have since a refined it and tried to teach it, don’t be racist. But this is a reflection, though, of how sensitive these systems are to what goes into them and what comes out of them. Now, that doesn’t mean that some you know user coming to your site is going to teach your chat bot to be racist. That’s not really the takeaway there. It’s that it can only be as good as what goes into it. And if you don’t already have the answers on your site, if you don’t already have the content necessary to address the needs of the users, then your chat bot can’t take care of that for you. It can’t solve that problem for you.

You already have to have all of that information. That’s time intensive, it’s labor intensive. It’s costly. And I’ll get into a little bit more of that here in just a few minutes. But this idea of it can never be better than what you’ve put into it is something to really latch on to. It’s not going to magically do things for you. You have to teach it, train it, program it. The next part of this is that as important as natural language processing is to a chat bot. It still sucks. Like we’re not good at it yet. language processing is still very hard for systems. There’s an article over on medium written by mentleman cadia. five reasons why your chat bot needs natural language processing. It’s a good read, go check it out. It gets into a lot of the the technical reasons why natural language processing is necessary for a chat bot to be really effective. It all comes back to this fact that English You know, is a very fungible kind of language, it can take a lot of different shapes and forms. And we’re really good at breaking the rules, but still understanding what we mean deriving things like intent and sentiment and things like that human to human, we’re very good at this. And if I, if I put a participle in the wrong place, or prepositional phrase at the start of a sentence, or things like that, we still are good at at deriving the meaning from that. We’re still working on that. On the machine side, though, progressing very quickly. Don’t get me wrong, and we’re much better at it today than we were even a year ago or two years ago. But it’s still a huge gap between where we are and where we need to go. I did a transaction recently on a website and I needed to talk to customer service about a return went to their website and I went in to look at my orders and I expected to just go in and hit a button. I need to return this click my order, boom, let me return it here. I’ll put it my form, I’ll tell you, it was a duplicate. It was a duplicate item I got for Christmas. And so I was going to go in and just punch the button and be done. I was greeted with a chat bot, they had taken away the customer service element of it. And or not because customer service but the internet, the simple interaction and put this chat bot there to kind of pick that up

and swap. So it says okay, well can I help you with and it gives me a list and I’m working through it. It told me and I quote, if your return is received outside of our return timeframe, or as an excluded item, we will not be able to process your return and you will not receive a refund. It’s like oh, that’s important because it’s Christmas, you know, the item was bought earlier. So I was a little concerned about the return timeframe. And I asked specifically, what’s the return timeframe? I used the exact two words that it used when it told me about it and it’s answer was I’m sorry. I did Understand your request. This is the challenge of natural language processing. This is the challenge of garbage in garbage out. This is why these tools get very, very flawed when they are being used in a customer service sense. Just give me a form. Tell me what, you know, we accept returns within 30 days, and we’re done. I’ll fill it out and we’ll move on. In this case, this tool couldn’t do that. As a result. Luckily, I was able to return it, I had to go through some other hoops. But it required me to get out of this. And what I end up seeing what what I think kind of in the back of my head is happening here is some systems are designed this way intentionally. They actually build in as a dark pattern as a hostile pattern. That if we can frustrate you into abandoning your attempt to return it, then you’ll keep the item will keep your money I genuinely genuinely believe that some organizations are utilizing chat bots in that way. And I felt that way in this use case very specifically because somebody took the time to write the answer, if your returns received outside the return timeframe or not, and blah, blah, blah. Somebody wrote that. And then nobody wrote, what’s their turn time for? So we need to get a lot better at this ability to read what your user is telling you and respond in kind with an appropriate answer, because the question I asked in this case, was not unforeseeable, I think in that situation, to that end, the next big problem is undiscoverable interfaces.

This is something that chat bots have, and they share it in common with something called a voice Ui Ui. And it’s this idea that you think about you go to a chat bot right, and sometimes when you open them up, some of them will prompt you with some options. It’ll open it up and say what can I help you with kind of like the way when you call a phone service line and it’ll say, press one for this too for this, some chat box will offer you that to set the user on a journey down one of the paths that it knows it can handle. First and foremost, if you’re doing that already, if that’s what you’re making your chat bot do, why do you need the chat bot? If you’re already putting people through a funnel that just has them click things, you’re just using a form and putting it inside a chat bot, which is ridiculous. You’ve left up this interaction in a way that it didn’t need fluffed up. There’s a one of the laws of UX is Hicks law Hicks laws, the time it takes to make a decision increases with the number and complexity of choices with a chat bot. It opens up let’s say there aren’t those buttons there and just as to what can I help you with today. Now, the number of choices I have and how I can approach them has increased exponentially. I don’t know what my limitations are, I don’t know what it can and cannot do. And so I’m left to guess this is actually a problem that is made worse by natural language processing, mind you, because the whole reason that exists is to address the complexity by which we can ask things. There’s effectively no finite limit to how we can ask for something or the combination of words that we can put together. So we need that processing step to distill things down. And it’s going to get stuff wrong once in a while. But the more open ended a chat bot is the more you make your user gas, the more you’re forcing them to figure out on their own what they can and can’t do. There’s a good example of this. This comes from that same Nielsen Norman study they said, for example, the bot had no knowledge of the drugs Zomig or silho pram, but was able to answer Questions about Lexapro presumably, the bot only worked with a subset of drugs, but the list was too long to display. However, this design decision rendered the bot useless. There was no way to tell in advance what types of tasks the bot will help with.

This is the problem with those interfaces, when there is no way to discover what they can do. Users have to guess and that leaves them feeling frustrated. Voice you eyes have had this problem with it whether you’ve used probably Alexa or google assistant or Siri or Cortana you have undoubtedly run into a situation where you asked them a question that they couldn’t answer. You change the way you asked it. They figured it out. These systems have had this problem forever because I can’t see now a chat bot does have that particular advantage that it is texted is in front of you. You can offer those buttons you can offer things that get people set on a journey. But ultimately you, you’re using them for the flexibility. And that flexibility makes it very hard to balance. Here’s what you can do and how you can do it with the expectation of Am I getting the right result as the user? This is something that I don’t have a good answer to. There is I don’t think a great approach to it in context without making the tool monolithic in nature. And that’s not a good use of your resources at that point. When it comes to resources, the thing that a chat bot does not do is fix underlying information architecture and content issues. This is really the biggie when it comes to the challenges and why chat bots get brought into organizations. We envision these issues with customer service, we envision the potential cost savings of not having to have a human being answer every single question but to answer user questions. You have to have the content Period. If the content exists, and you’ve put it in good IAA and made it find the bowl, then why do you need a chat bot? If the information isn’t clear, if you don’t have it structured Well, if you haven’t taken the time to write it clearly, those things all have to happen before you can make the chat bot good. And by the time you have finished all of those things and address those problems, then you’ve addressed the problem. You don’t need the chat bot anymore. See, that’s the thing about all of this is a chat bot is designed to facilitate content discovery.

But a chat bot isn’t necessary for content discovery. In most cases, because remember, a chat bot works best in very simple linear transactions and things that are very easy for it to handle. You know what else is good at that search box. If you write good content, you structure an order organize it? Well, you put it in meaningful places, you make micro interactions on your forums, you write good microcopy that gets attached to fields. You know, to go back to my earlier example, on the return that I tried to do, you know, all I needed was a form that would tell me, it will or won’t accept my return. That’s all I needed. I didn’t need to get trapped inside of that bought, unable to exercise. My intention. All I needed was one piece of information that would have been incredibly easy to include. That’s the real gotcha when it comes to all of this. And a lot of organizations. Their problem is they don’t have that content, and they don’t have those tools. And so they see this as, Oh, bring in the chat box to take care of that and help people find the stuff that we can’t get in the right places because we don’t have content authors or we aren’t doing content auditing or we don’t have somebody reviewing the writing or the markup or anything like that. If you don’t have all that in place, if you aren’t spending that money on people to solve those problems First, the chat bots not going to help you. That’s period. I also get to end my gripes with my favorite absolute philosophy in the world. That’s the last time you heard me do a solo episode do less better. I’m going to quote Nielsen Norman on this. Companies are better off investing their money in the existing well established channels. Improving the UX of your website or app will bring you higher return on investment than creating a chat bot that will get little use. We saw that even good chat bots, which are likely to require increased development and testing costs have little chance of being discovered and considered useful.

do less, better. Focus on your website, focus on your content. identify the problems your consumers and customers and visitors are having, and then adjust your website to account for those things. Put those answers where they need to be, you do that, and you don’t need a chat bot, you do that. And you find other ways to elicit value from your website that doesn’t involve spending money on a chat bot. So all that said, I know that sometimes you’re handed the thing and you said, we need to put it on the site. So let’s talk a little bit about what you can do to at least try to get some success out of this if you are putting the position where you have to implement one. First and foremost, what I want you to do is start with your search analytics. If you don’t have search analytics, you are not prepared to launch a chat bot. Get some analytics on your site. Look for what people search for on your site, but not just what look at how they asked for it. You will need those contextual clues to start building the vocabulary that you’ll use to inform your chat bot and instill meaning in it. Now, this is just good advice in general, avoid jargon. A lot of the times what you’ll find is that people may be asking for things or searching for things, using words that you don’t use internally. Or, you know, you may have a product name that that’s just not how people you know, it Kleenex versus tissue, for instance, not the best example. But that idea of you may have a name for a product that isn’t the way other people look forward or want to find it. And you’ll need to know that both to build good content on your site, but also make sure the chat bot knows when I come in and ask about I need a pickup that I’m actually saying I need a truck. Those kinds of things. Secondly, do focus groups with users for the same reason. Find out how they would ask for thing getting idea, what are their expectations based on certain questions, identify where they would be uncomfortable on top of that, and what they might not want to share.

You know, there may be cases where you’re asking for customer service and you want to confirm, let’s say, let’s say your social security number, a lot of people might say I’m not comfortable typing my social security number into a chat bot, because they may not know that it’s a person or not, they may not know how that data is being used or why it’s being asked for. So find out some of that because that will inform how you develop user journeys within that tool. Be upfront with establishing the clear capabilities of your chat bot, you need to be able to let users know what they can and can’t do. A lot of most chat bots will integrate with API’s for your other services. So for instance, you tell the user this tool can look up your order. In the backend what’s happening is the service that is providing your chat bot has an API connection to your order system that then looks it up and gives it to the user. chat bot capabilities as a result can vary wildly from site to site, do you need to let those users know here’s what it can do, here’s what it can help you with. And when it’s helping somebody with those things, if it encounters a problem, you need to give it a good useful escape hatch so that people can get out of that interaction. And either email somebody fill out a form, call a phone number, something to continue the act of completing their interaction, they need to have that though, and you can’t make them guess at what it can and can’t do. To that end, you have to track usage. If your vendor does not provide you analytics, you have to have a way of tying in your own and if you can’t do one of those two things, don’t use that chat bot. You have to track usage because you have to know our users completing the tasks that they are setting out to do. What is their abandonment rate? Are we failing the users with this feature? This is baseline criteria for any feature you want to build on your site, regardless of what it’s in, knowing if it’s working or not working, is how you know if it’s worth the money you’re spending on it. And so you have to have a way to track that within. If it’s not Google Analytics or the vendor, some other system that you have those set that up and set up those expectations from the beginning.

This one’s silly but don’t hide that they are talking to a bot. A lot of services. If you go to their websites, what you’ll get is a very casual, hello from the device. Hey, how are you doing today? Is there something that I can help you with? They’ll put a name on it. They’ll even sometimes have an avatar of somebody’s face next to it. They make it look and feel very human because they want you to feel comfortable using it. But that can set up an unrealistic expectation, I expect a human to understand what I typed and what I say. And if I don’t get that in response because the natural language processing isn’t up to par, I’m going to get frustrated as a user, it’s okay to tell those users, hey, you’re interacting with a bot, its capabilities are limited. To go back to what I had said earlier, establish clear capabilities, here’s what it can do. That’s fine. You can give it a name. Plenty of services ino has a name, I believe progressive, you know, progressive mascot flow or character flow. They that is the name of the chat bot. So that makes sense. You know, we have these names for our Alexa Siri Cortana you know, abstract names, you can do that. That’s fine.

And finally fail. Clearly. can’t stress this enough. Your chat bot is going to fail your users without a question without a doubt. It is going fail, and it’s probably going to fail frequently and you need it to do that clearly. I said earlier, give them escape hatches, when a failure happens, let them know why let them know what that problem is. what really got me turned against these things was one of my first major interactions with like a new sort of one of the new generation chat bots. here in Kansas back in 2018. Or prior to the 2018. Rather, we had a DMV system for going in and renewing your tags you go in, they mail you a piece of paper, that piece of paper has how much it is your cars and codes, you would go to their site and you would just punch in your code. It would show you your car, you ask give her that ask if you would want you know, Park like State Park parking permits, and you would check it off on the form and then you would go to your checkout and you would pay for it. And then a couple days later you get your new stickers in the mail. It was ugly, it was gross looking. It was a yellow It was Sarah fonts. But it worked. It did its job. That was the real thing about it is, it didn’t have to be pretty because it just had to do a very simple transaction. Prior to 2018, they got rid of that system. They didn’t augment it, they got rid of it. And then its place. They put a chat bot, a purely transactional interaction bot that was designed to do the same thing. You still got your thing in the mail, you still went to it. You were still supposed to it would say, Okay, let me look up your number. Tell me your number. If you go look at that link, I’ll having this in the show notes. linking to my Twitter comments on this. I’ve got screenshots of this entire exchange that I had with it. I started punching in my information. I was getting to the point of Checkout, and it’s you know, it says, Okay, do you want to do anything else? No. Okay. And then it came back. I can’t complete your transaction. That was it. Didn’t tell me why. It asked if I think it’s it said, you know, do you want to try again? Well, yes, rerun it. It about three times I tried this. It just kept coming back saying can’t help you.

It didn’t tell me why it couldn’t help me. It didn’t give me an escape hatch. It didn’t give me a phone number to call it didn’t give me an email didn’t route me to another system where I could figure something out. I ended up having to take off work, go down to the courthouse to visit the DMV and figure out what the problem was. Turns out as an incredibly simple thing that the system would have known. How do I know that because their system new at the DMV, so whatever API is, they’re tied into that machine could have told me, here’s the problem. Do you want to resolve it here? Because I could have resolved that there if they just tied it in. It wasn’t anything complicated. It wasn’t anything complex. But it didn’t fail. Clearly. It failed opaquely. It just said, okay, you’re screwed. It didn’t even tell me to go to the DMV. It just said it couldn’t help me. There’s an article that I’ll leave you with. Its how to conduct UX research for chatbox to improve usability, it’s by Richard Makati. It’s got a lot of this information, it’s got some other examples and other advice that you need to take into consideration. The good read, it’s very thorough on a lot of this stuff, highly recommend that too. But if you at least take some of these steps, even if you’re forced into implementing a chat bot for years, for years, users, this can at least hopefully help them. You know, find that experience a little more useful. You’ll eat some of their frustrations, and you’ll be setting yourself up to improve upon those interactions over time as you make new content, make better answers, create new experiences, new user journeys that these devices can take in and help people with. So as that happens, when you’ve got that analytics, when you’ve done that research, when you’ve done those tests, you’ve got the ability to inform users Much better in these exchanges. So I hope that was useful. Stick with us. I’ll be right back after the break. The drunken UX podcast is brought to you by our friends at New cloud. New cloud is an industry leading interactive that provider who has been building location based solutions for organizations for a decade. Are you trying to find a simple solution to provide your users with an interactive map of your school city or business? Well, new clouds interactive map platform gives you the power to make and edit a custom interactive map in just two minutes. They have a team of professional cartographers who specialize in map illustrations of many different styles and are ready to design and artistic rendering to fit your exact needs. One map serves all of your users devices with responsive maps that are designed to scale and blending seamlessly with your existing website to request a demonstration or to view their portfolio visit them online at New cloud.com slash trunking UX that’s in you cloud.com slash drunken UX. As always, I hope you found this information useful, helpful in some fashion. If you’d like to chat with us, shoot us a message over on twitter at drunken UX. Let us know what your thoughts on chat bots are. Do you agree with me? Do you disagree with me or you can contact me directly at Fienen FIEN een I’d love to have some conversations with folks about this. And hopefully, you know, share your experiences.

Have you ran in the chat box that are frustrated you have you ran into chat box that have been incredibly useful and effective? I think you know, just because I don’t advise people use them. And just because I don’t think they’re generally good for user experience doesn’t mean they always fail. It doesn’t mean they can’t be done well. It just means that as a rule of thumb, most places are not resourced in a way to make them more. You can also find us on Facebook, same slash drunken us at Instagram slash drunken UX podcast, hit us up on slack is drunken ux.com slash slack. And also check out our show notes at the website. Like I said, I’ll have a lot of articles and information there. There’s another website. It’s great. It’s called the UX of chat box.com. They just keep a running list of a lot of articles that are written on chat bots, both pro and con. So there’s a fair balance and information there if you’re trying to do some research on it. All that will be at the website. As always want to encourage you do your research, put your people first. And if there’s one thing that’s more important above all else when it comes to user experience and chat bots, and any kind of customer service aspect, it’s to keep your personas close. And your users closer. Bye bye.


This episode of The Drunken UX Podcast brought to you by nuCloud.