[PODCAST] In-Ear Insights: How To Plan for Chatbots

In this episode of In-Ear Insights, listen as Katie and Chris discuss chatbots in marketing. What are chatbots good at? What are they bad at? Will chatbots take your job? How do you build the training library of questions a chatbot should answer?

Katie and Chris also reference some of the common platforms like SnapEngage, MobileMonkey, IBM Watson, Wit.ai and many more. Listen in!

Listen to the audio here:

Download the MP3 audio here.

Machine-Generated Transcript

What follows is an AI-generated transcript. The transcript may contain errors and is not a substitute for listening to the episode.

Christopher Penn
 This is in your insights, the trust insights podcast,

in this week’s in your insights let’s chat about chatbots. They are the one of the biggest topics in marketing in terms of, you know, should we be using a hashtag. We’re using them. We have seen the a lot of people deploy them very badly. So Katie, what do you think is important about in the use of chat bots?

Katie Robbert
 I think chatbots I think as a marketer, and as a business owner, you need to think about what is the purpose of having a chat bot. So chat bots, you see them everywhere, and you might not even realize you’re interacting with them. So you have some companies, especially in the financial institution, that create these very lifelike chat bots as if you’re talking with a real customer service person, someone has taken a lot of time to program and the response is based on certain keywords that you say, but then you see very basic chat bots on programs such as Facebook, which are more they prompt you with what questions like asked me about what hours my businesses open or asked me about my services. And so I think in terms of chat bots, you know, this is very much a broken record, but start with what your plan is what your goal was the question you try to answer, but in the sense of a chat bot, you can sort of go very basic or very high tech. And for someone who’s just starting out, you know, we would always recommend start with the basics and sort of think about those initial questions. Chris, what are your thoughts on how chat bots are going to start

I guess revolutionising how we market to people?

Christopher Penn
 I don’t think it’s a revolution. It’s an evolution instead of when we think about the fundamental promises of AI, which is what chatbots a good number of them are based on its acceleration, automation and accuracy. And the big one is that automation, what can we do? What can we deploy, that will free up time for our employees to do something else? So, so if you have 20 people working customer service phone lines, you know, can you use the chat bot and get that down to 10? Or can you get take those those 10 people who are no longer answering the same questions over and over again, and have them do something, you know, some more productive work. I think that’s probably the most important consideration is Do you have enough regular, frequent common questions that automating the answers to those questions? Will you know what follows the Pareto principle of is 80% of your volume 20% of the questions that are asked if so, that’s a good candidate for automation. The question is, does your audience use technology like that, you know, I think a good proxy for that would be to measure how much of mobile traffic you get, if you get zero mobile traffic, there’s a good chance that, you know, people are not technologically savvy, and therefore, you know, they’re they’re not contemporary audiences that would feel comfortable working with automation, you’re pointing your finger?

Katie Robbert
 Yes, because I want to respectfully disagree, okay. It’s your analysis of, of whether or not a chat bot is right for you. So if you think about Facebook, for example, a lot of people who use Facebook these days are using it desktop, not people who are just scrolling through their feed, but people who are actually interacting with it. And the Facebook chat bot for a business actually works really well on desktop. And so I think that it’s a matter it’s not what device are they using, but it’s more Where’s your audience coming from? Are they coming from desktop? Are they coming from tablet? Are they coming from mobile, because that will help dictate how sophisticated you want to make a chat border chat bot or even if you need a chat bot, because a lot of times chat bots on mobile are really tough. Especially not all phones are like the big walking like iPhones. Some of them are still normal size for like a normal person. And so so I respectfully disagree. I think we’re getting to the same point, though, have you need to take a look at where your audience is coming from. And that will help determine whether or not a chat bot is right for you, and how much time and effort you put into programming it.

Christopher Penn
 Yep, exactly. And

the the platform also dictates where you build that technology out. So if your audience is on Facebook, then yeah, using chat bots that integrate with Facebook Messenger is the smart way to go. I mean, I really can’t think of any reason why you wouldn’t do that. If you if you had a large audience on Facebook, if you have an audience on your website that you care about, then going to want a platform there. If your audience likes to text, there are chat bots for that. So the prominent vendors in the space would be like chat, fuel, snap, engage mobile monkey, there’s a whole bunch of them out there. And you can also build your own to like IBM Watson has the ability to build conversation bots. And you can even build and train using Watson services. I think to your point, though, starting with one of the dedicated providers that has sort of the drag and drop interfaces the way to to dip your toe in the water and figure out like, is this right for my business? Can I trim off those 80% of questions. And then once you’ve proven the value mainly through cost savings, then you might want to level up to something that uses true natural language AI.

Katie Robbert
 Well, and I think that one of the ways that people can evaluate is sort of writing out that script that back and forth of, you know, what would a chatbot be saying that my customer service folks might not might also be saying, and so, you know, it could be as simple as, okay, what are the most common questions, looking at your own search data to figure out what are people looking for they are they looking for the hours every business? Are they looking for the services, the address whatever those things are? Or are they looking for something deeper, and maybe it’s connects me with an actual customer service rep, but really sort of outlining what are those call and response types of things that a chat bot would be doing for you that you can sort of Chris, to your point free up a person from needing to do manually. So you don’t need to have people staffed 24 seven. And actually, that’s a really good use case is if you have an online business and you work with in different time zones, having a chat bot, at least field the initial inquiry so that it goes to the right customer service rep when they sign on first thing in the morning, then that’s a really good way to utilize the chat bot.

Christopher Penn
 One of the projects we did recently was actually for an agency that recruits truck drivers. And part of that project was to get 17,400 calls from their call center transcribed using AI and then doing analysis of those. And I think that’s probably something that people don’t give a whole lot of thought to is, what data sources Do we have to help us understand what those questions are, if you have a customer service inbox, there’s the the fuel for your chat bot is a travel plan.

If you have call center transcripts, that’s a great place to go. If you have questions or something sort of question form on your website. That’s a great place to get those questions. And a lot of those questions that you go through them and start to build out categories like ours, pricing, availability, whatever the the main topics are, that people keep asking you over and over again, that will give you the I guess that’s really the planning phase of a chat bot is let’s go through and figure out what that we need this thing to answer first, and then worry about the technology later.

Katie Robbert
 Hmm. Well, and I would add to that also sort of the volume of the increase. So if you’re getting a couple of hundred people to your website a week, you might not need a chatbot. But if you’re a larger Corporation, again, sort of thinking back to like a financial institution and probably gets millions of people hitting their website a week,

the way to scale your staff is to utilize in conjunction with your people, some chat bots to answer some of those basic preliminary questions and even sort of do some pre qualifying so that they know that they’re going to the right people to talk to, to about I just have some general customer support, or I actually I want to buy something because I’ve indicated with the right terms that I am ready to, you know, give you my credit card and take all of this home today,

Christopher Penn
 which I thought was really important question, when should you not be using chat bots? What is what are the wrong applications of this technology?

Katie Robbert
 That is a really good question. I think that the wrong applications. Well, I guess let’s take a step back. When you’re assessing whether or not you’re ready for a chat bot, it comes down to do you have enough volume of inquiries to determine whether or not a chat bot is right for you. If you’re a one person business owner, then yeah, you probably want to have a chat bot to field some of that information coming in, so that it’s a little less onerous for you, you don’t want a chat bot to collect sensitive information, you don’t want a chatbot to collect identifying information. And if you do need to have those things happening, you need to make sure that you have the right processes in place to encrypt that data so that it’s not just hanging out there. Because as we’ve seen in multiple articles, data is really easy to get ahold of. There was a Facebook there is currently a Facebook thing going around right now about your phone number, and how it’s available to everybody. You know, and so those are the types of things that if you are planning to use a chatbot So your question specifically was, what are the wrong applications at this time, probably any sort of serious medical diagnosis,

you know, but that said, there are now like the medical field is becoming more sophisticated with chat bots to at least sort of help you get answers to your questions. But at this time, where we are with technology, a chat bot cannot replace a human in certain situations, such as a true medical diagnosis or some sort of a mental health diagnosis or someone looking for help in those respects. So those would be sort of the two big things for me of don’t replace the human with the chat bot, because someone is actually reaching out for help.

Christopher Penn
 How do you from a strategy perspective, do the risk assessment to figure out like, are we at risk? So I could easily see a technologist you know, someone like because like, I bet you I could build a thing for this. And, you know, like, let’s, let’s make a chatbot and I was suicide helpline. And obviously, common sense would say, that’s a really bad idea, because there’s so many parameters, but how do you how do we help an executive figure out? Okay, how do I assess the risks of determining whether this technology is appropriate or not?

Katie Robbert
 Well, I think it goes back to what you were saying about what data do you currently have, that will help you determine if chat bot is right for you. So it’s taking a look at all of your different data sets? And then really understanding what is the what is this chatbot going to do for my audience that my team is not currently doing? What can it supplement? What can it replace, and just breaking down those different scenarios. And so let’s say you’re a tech company who builds widgets, and you want a chatbot to sort of help pre screen, you probably actually need to factor in some of those scenarios where someone is trying to hack or somebody mistakenly reaches out for help, because that would be some sort of a liability. And you need to have those scenarios covered. And that’s part of your due diligence in your risk management. So let’s say, you know, you as the CEO of a tech company, find out that people have been pinging your chat bot saying, I need help, I’m feeling not so great today, what are you going to do about it, you need to have those controls in place. And so it’s making sure that your chat bot is programmed to do something about that, even if it’s just to say, hey, this isn’t the right place, here are some resources you need to go, you know, check those out, because maybe you guys have names that are very similar to something, you know, so I think that that’s part of your risk management is making sure that you have scenarios that are likely not to happen still covered, because it’s the internet and pretty much anything goes.

Christopher Penn
 So is there sort of a stock set of things that from a planning perspective, everybody and their cousin should have, like, this is a template that should just be part of the standard install?

Katie Robbert
 Yeah, absolutely. And so sort of going back to that scenario of you’re a tech company who has nothing to do with helping people on a personal level, but people are personally reaching out, you should have some of those stock responses. And I’m sure if you just sort of like do a quick internet search for what that looks like, you can find some of that information. But, you know, if someone reaches out to you and says, you know, I’m feeling suicidal, or I just accidentally cut off my hand, you know, there’s ways to respond to that that are very consistent is the word I’m trying to find. So, you know, you should probably call 911, here’s some, you know, local resources for you to check out and just make sure that those are just in there, and then make sure that you are auditing the information that goes into your chat bot on a regular basis, and that you are reporting those up to the proper authorities to make sure that they have been actually notified, and they can follow up with those things. So those are some of the processes that people don’t necessarily think about that they have to do. But they’re part of, you know, if you’re going to interact with humans on the sort of, like, blind faith basis, you have to plan for the unexpected.

Christopher Penn
 From a a automation perspective, one of the concerns that people continually expresses, you know, is this thing going to take my job? And from your perspective, how realistic is that risk, especially if, you know, your job is eight hours a day of answering the exact same questions over and over again?

Katie Robbert
 Well, I think that that’s a really key point that you just said, The answering the same question over and over again. So in some ways, the chat bot is going to take a portion of your job, but you might actually find some relief in that. So the chat box, you can program to answer those same questions over and over again, give those stock answers. Are you open? Yes. What time you open till five o’clock? Will you be open past five? No. So giving so variations of the same question, people can interact with the chat bot, which then gives you the human more time to, like, build those relationships with your customers? Find out, Hey, how are things going, we sold you this, you know, six months ago, are you interested in you know, so you can really focus on sort of that next level of customer service and take yourself out of the weeds of answering those same questions over and over and over again,

Christopher Penn
 I question that only because thinking back to when we had a you know, we worked in a company with lots of people at very different levels and very defined levels, like, yeah, you are this you are the lowest person on the totem pole, you get to only copy and paste eight hours a day, that person and typically didn’t, couldn’t or wouldn’t reach up except for a few folks who are ambitious, for the most part, people are very much in fact, there’s even an expression where you stay in your lane kind of thing.

And then when we think about these automation things, how realistic is it for somebody like that Junior most person to be able to reach up or feel like they’re, they’re permitted to reach up so that they can avoid the machine going up, you know, that jobs completely unnecessary now?

Katie Robbert
 Well, I think that that’s a really good point. And, you know, it is the responsibility of the person whose job is redundant to think about, Okay, what else is there for me? What else can I teach myself? What else can I learn? What questions Am I not asking? So if you’re someone, as you’re describing, Chris, who’s just complacent with copying and pasting and not really thinking outside of, you know what that responsibility looks like, then yes, you are very much at risk of machine learning, taking over your responsibilities because it’s repeatable and it’s consistent. So you know, you as the person needs to be thinking about my own professional development What else do I want to do that said, if you’re happy copying and pasting and then you don’t actually want a job after that and great you’re good you don’t have anything else that you need to do the unemployment line is right over there.

Christopher Penn
 Uh, yeah, this is we should probably actually do a entire episode some point on sort of the the societal implications of machine learning and AI because there is so much happening on on that front about what jobs will be at risk. One of the most interesting things on the chat bot front is with lot of the new language models that have come out in the last six months. The old school style of building a chat bot, where you’d have to program in every response is rapidly becoming obsolete with particularly with Bert an elbow, the two not the Sesame Street characters, but

they can now ingest to create realistic natural language responses about pretty much anything. So now it’s just pulling guard rails on them saying like, when you get a question that’s off topic, don’t bother responding to or guide the user back as opposed to engaging in a long conversation about, you know, gender fluidity and Cinderella was just entirely possible if these bots just were completely unsupervised. But for people who are planning for chat bots, then you have to give some thought to what are the guardrails you’re going to put on it once you can respond in a very automated fashion to pretty much any inquiry it’s given and give reasonable responses back

for companies trying to figure out again, going back to that category of risk assessment of risk mitigation. How do you plan for, hey, what are the what should the guardrails be?

Katie Robbert
 Well, I think that it’s similar to outlining the job responsibilities as an employee, where you say, you know, to use your face, Chris, this is your lane, I want you to stay in this lane, and then everything else are things that you don’t need to worry about it. So you could even sort of use that same mentality with programming your chat bot, like this is above your pay grade, if this comes up, paying somebody else paying the manager paying whoever. And so that’s exactly it. That goes into your risk assessment plan, where if I’m a tech company, and I sell widgets, and somebody asked me about the price of bananas, then that’s not something that’s within my lane that I need to be worried about. However, it’s also not something that has that I’ve been told is an adverse event or somebody who is in trouble. So you know, I have now have two boxes checked one, it’s not in scope and to it’s not a crisis, so I can just go ahead and ignore that thing. And so thinking about it in those terms, will really help you put those guardrails on your chat bot

Christopher Penn
 and at what point do you how do you plan that out how to in a concrete fashion? Is? Is it a planning meeting as a set of documents? What’s the process for building out those rules?

Katie Robbert
 Yes,

so honestly, the best place to start is with your own data. And so one of the things if you have access to it is to do some text mining on your own customer service or customer support, or even your social media comments, data to figure out what are people asking about. So it’s starting with doing a very large data poll about what people care about the most what your audience cares about the most. And then my recommendation would be to sort of sit down with different disciplines around your company. So not just the C suite, not just the sales team, but really getting a cross section of folks who can really sort of think through from different perspectives, what are the types of questions are what are the types of appropriate responses? And so it’s a combination of deep data analysis, a planning meeting, and then documentation, you need that documentation, ultimately to refer back to. So once you’ve programmed the chatbot, you need that for QA, you need that to refer back to if you’re chatbot starts giving the wrong responses, then you can say, Well, this is what we intended it to do. So it is it’s not just an overnight thing where you can just sort of slap up a chat bot and say, okay, run with it, give some thought to it. Figure out what’s the goal of having a chat bot? What is it going to do to help benefit and grow the company? What are the types of questions that people are asking most commonly and then what are the appropriate responses? And then what are the things that I don’t want my chat bot to care about.

Christopher Penn
 So I think that’s a great place to wrap up it and certainly the the folks who are on this podcast will not be replaced by chat bots anytime soon, because I don’t think we would want to listen to two machines talking to each other. Maybe we would either way,

subscribe to our YouTube channel, add to our newsletter, and we will talk to you next time.

Thanks for listening to in your insights, leave a comment on the accompanying blog post. If you have follow up questions or email us at marketing at trust the insights.ai if you enjoy the this episode, please leave a review on your favorite podcast service like iTunes, Google podcasts, Stitcher, Spotify, and as always, if you need help with your data and analytics visit trust insights.ai for more.