Podcast | Actionable steps to automating customer communications

Jean
Shin

Director, Strategy & Content | Podcast Host of Mobile Interactions Now

25 min podcast
tyntec podcast

In this Part 2 of the chatbot development episode, Malte Kosub from Parloa lays out the steps involved in choosing which use cases to automate, setting up teams for chatbot testing, deciding when to go live on which channel – and why messaging is becoming the most popular choice for customer-facing chatbots.

Podcast Script:

Jean:
Malte, welcome back to the show. In our previous episode we talked a lot about what it is like to have a chatbot and the customer experience and we talked about patience we had to have to teach the chatbot to get the kind of experience we want out of it. And now, in this episode I would love to delve a little bit deeper into it, because it sounds like there are some things are really better at, as is now, and some things that are a little bit more difficult. But first let's talk about use cases, what use cases that are really acing as is now, and you can share some of the things that are not really happening yet in you're opinion.

 

Malte:
Yeah. Use case is something every company asks us what we think about it, and it is always a really easy answer to companies and we are always saying, okay, what are the repetitive questions customers ask you, and that's really easy because a lot of companies know that. They have their documentation about call center questions or about email questions. So, just look what repetitive questions you get. That's the first start. In most cases 80% of the customer interaction has just 20% of all the potential questions. So that helps to get first set of use case I would say. And then you always have probably high volume use case because those are the most repetitive ones.

 

Malte:
Then we should think about use cases which can be automated. So, if I'm an insurance and I had my whole house burned down, I might not want to talk with the bot but with a real person because it's really, really challenging. But if I just wanted to change my address, I don't care. I just want to make that happen and... It should work and maybe I want to have an email at the end that it will work, but I can also talk to a bot. So first of all, what repetitive use cases you have first and then second, do you really think that the customer would be satisfied by doing that with a bot? And if there are hundreds of people calling you because of birdhouses, which would probably not happen, but if this happened, the first thing, is it a repetitive use case? Yeah, check. That's fine. Secondly, could you do that by a bot? Probably not, so you should not do it. But if both of those things are right, I think that is a good use case.
 

Malte:
I'm sure we always need to see that's more or less a third way of thinking about it and we need to see is it possible with the current AI to understand all the things which the user says? So for example, when someone is calling you and you have a call center bot and the user needs to give their email address. That is something which is quite complicated because the automated speech recognition has some problems with names because names are very complicated. So there are some things which might make it a little bit tricky to use a bot. So we also need to to take a deeper look into what data do you really need from the customer. When do we have a use case? So repetitive. Do you really want to do that by a bot? And is it possible from a data perspective and automated speech recognition perspective? If all three things are okay, then yeah, you should do it. If one of those are not good, you in the first step, you shouldn't do it.
 

Jean:
I think that echoes some of the concerns talking with a lot of people who are integrating checkbot with WhatsApp Business API and part of the requirement is there needs to be a clear path to hand over to a human agent and some of the scenarios you are talking about, you know how human interaction goes. You call in for change of address, all of sudden you realize you have to talk about something else as well and this being able to hand over when it's needed smoothly from bot to a human agent and carry on that interaction that just happened, how are we doing this? I mean, where are we at, is this something that you are doing more and more and seeing it as a real possibility?
 

Malte:
Yes, definitely. I think that's something we talked about in the other episode. And so I deeply think that it is very, very important to have a human handover when you're developing a bot because it is not possible to automate 100% if you are good you can automate 30 to 50% so the handover I think is very important. Sure, it is very challenging because what I doing when it's like 11:00 PM and there is no one in the office. So you need to have your processes to have someone in the office and doing your real chat with customers but to build a satisfying experience. I would suggest to every company to have real chat agents who take over when the bot can't answer the question anymore. Yeah. So, it's challenging. It also has some structural challenges, but I would suggest that.
 

Jean:
Addressing some of the issues that arise, I think it is, a lot of it has been traditionally handled by customer service call centers. And guess what, correct me if I’m wrong because you're a data guy, it sounds like chatbots and all those AI power technologies are being deployed at call centers first. Can you tell us why that is the case?
 

Malte:
That's a good question. I think it might be because there are the most repetitive use cases coming in, so maybe internally also a lot of repetitive use cases, but there is a huge, it's a high volume field and simultaneously a big cost center. So the companies have a lot of costs when it comes to customer service and they have the feeling that they can save some costs there. And I think that's totally true. If they do it right, they can save some costs. So I think it's a pretty straightforward way to automate, partially automate customer service. When we compare it to internal cases, it makes a lot of sense. But when we talk about it, I think there are, I experienced some companies who start with internal cases because there they can't break anything because it's just the employees and if something is not working then it's fine.
 

Malte:
They're just testing it and then they're rolling it out to their customer servers externally. But I always also see companies doing it the other way around, starting with the customer service because they really want to decrease the costs and want to automate stuff because they have maybe some pressure from the market, from the CEO, and so I think they're both companies doing one of those two ways. And maybe to add something on that, of course there's also commerce. You could also, well we always talk about customer service, marketing and commerce. I think marketing is something which companies also started with when we talk about Facebook messenger bots, a lot of companies that are with some marketing stuff there, which is not very critical when it comes to customer data. Then now a lot of service things come up and I think in the future also commercial will play a roll here. But I think we just started, it's still developing because commerce has some challenges when it comes to convert and when it comes to pay and so on.
 

Jean:
That actually reminds me of a recent recording I did with an analyst from Ovum who was talking about this trend changing from conversation to commerce and each player being ready with a certain part but not entirely the whole experience. And it is still something to be shaped and it is certainly an interesting time. You talked about one of the use cases insurance company as an example. And I saw some statistics related to some airline industries where incoming customer communication is increasing, like 80% is almost messaging versus other forms of interfaces. Do you see any differences in industries taking on this differently, where do you see more opportunities?
 

Malte:
So you mean which industries are focusing on bots? I see a lot of companies from the financial sector investing in that space, banks and insurances, of course, they have money to to invest in innovation. I also see a lot of e-commerce companies trying to start thinking about it, but some them have already implemented. So obvious cases. My feeling is, and that's what I experience every day when talking to them, we have some e-commerce clients, commercial is something they're thinking about, but this will follow in the future. Media companies are thinking about that very, very deeply because when we talk about digital assistance, this is a new interface which connects customers to the internet so more or less a new search engine, and for media companies, search engines are very important, so here we see a lot of movement. Of course airlines, when we talk about, again the digital assistant are doing a lot, but also KLM for example did a Facebook Messenger bot and they are also very innovative, and so they're doing some stuff there.
 

Malte:
So we could name a lot of industries. I think that nearly all industries have had their first experience with that and started developing their first bots and what I experienced three or four years ago when Facebook announced that they open up the Facebook Messenger, a lot of chatbots emerged. Then some time later Alexa came up and all the voice bot development emerged and when the chatbot development emerged, a lot of companies and innovation units from companies started to build first prototypes on Facebook Messenger and these were mainly marketing departments or innovation and digital departments.
 

Malte:
Then with Alexa, the same happened. Innovation departments did that, the marketing and digital departments did that and what we are seeing right now that companies are building own conversational units to do, not just one chatbot as a POC or Alexa skill as a POC. They really tried to think about it much more on a company wide context, so how can we implement conversational AI on different channels, how can we optimize it within one team? How can we learn from one channel for the other channel? How can we evaluate different AI components for each channel? So not, you have a silo and one POC for one channel like Alexa, one silo and one POC for chatbot, but really think about it as one important topic and we call it conversational AI. And it doesn't really matter which channel it is. It matters that it's conversational AI and that we need a strategy for that.
 

Jean:
And that's precisely why I'm getting a lot of integration related questions and it is no longer just a simple one task agent that just handles that. And so I mean you broke it down a little bit in terms of what is happening in the backend to make the chatbot tools to be available enterprise wide and just kind of feeding back some of the learning and using it.
 

Jean:
So where are we at in terms of using multiple bot task agents versus having one interfacing customers via, let’s say a messaging app, and integrating everything in the backend. More broken down tasks or more of a general purpose chatbot?
 

Malte:
I think companies first need one database where all the bots get their data from. So one central, maybe it's your CRM, then you have your NLU and you should use all your data which coming in and it doesn't depend from what channel, to use that for your speech model. You should have one front and one system for your team to build those experience to make it better. So not to have one tool for Chad, one tool for digital assistants, one tool for phone that you should have one tool to train it and to make it better. However, a bot on chat can be totally different than a bot on a call center. So, the functionality can be different, also the training data can be different. So I think you need to have the same infrastructure, the same team, but you might have different bots. So some part is similar and you should try to use and reuse your knowledge you generated on one channel for the other. But in the future, there in a lot of cases need to be some differences on the different channels.
 

Jean:
That makes sense because a lot of channels have different features they expose that we can utilize. So all those factor into. Because at the end of the day you want to get the best out of each channel.
 

Malte:
Exactly.
 

Jean:
And I think this probably will be my last question by the way. And let's bring it all together now. You've been incredibly helpful breaking down things and looking at some of the complexity hidden behind that. As a brand who are looking at their customers who are completely changing and you have this landscape of different tools that there are some of the, you know, legacy things that they've been using and all this. What would you say if somebody is willing to do some innovative things, where to get started first, how do you approach first?
 

Malte:
So what are the first steps when a company decides to build a bot?
 

Jean:
Or determined to get the best out of all the signals that they are getting. What's possible?
 

Malte:
Yeah. So if they have already a chat in place, for example, really try to save all the data you have because they are very, very important because if you already have your data, you can reuse them to train the bot. So that's made me first try to do that as fast as possible. Then you really should think about what use cases and what channel I want to focus on first because really, first focus on one channel and maybe one use cases, well let's call it area. Maybe it's customer service and then you can start with the other because you need to learn how to implement conversational AI in your company first. So first think about use cases, so what are repetitive use cases and what do you think, which channel are important for your customers? If you're a company with a lot of old people, maybe a chatbot is not the best way to go, maybe it's the call center? If you're a very innovative new brand for young people, maybe WhatsApp is the perfect place to go. So really think about use cases in combination with the right channel.
 

Malte:
As soon as you did that and you figured out, okay, those 10 use cases are important and this is my channel, you really should think about what is the best infrastructure, what is the natural language and understanding I'm using, what is the automated speech recognition that I'm using? Is it from Microsoft? Is it from Dialogflow? Is it from the startup and which content management system or dialog management system I am using? Which infrastructure are you using?
 

Malte:
Then you should think about where do I get the right people from. So you can ask developers or UX designers, but they might not have the knowledge how to build a good bot. So you should either buy this knowledge from an external agency or you should hire someone who has experience with that. So after you have the right people in place, you should start talking to people. And maybe one example is Wizard of Oz testing. So without having any line of code or anything, you can just talk to a user. How would they interact with you? And then you should start with building the bot. And one of the most important and I think underestimated things building bot is testing. Testing, testing, testing. When you build a website, it's a red button. And yeah, you can also test if it's better, like if it's red or green. Yeah. But it's a button and you can click on it, it's pretty straight forward. But when it comes to conversational AI and bots, it's really, really crucial to test it.
 

Malte:
So as soon as you starting, you should really implement a good and real good structured testing process. And then yeah, just do it and keep on focusing, reiterating it and evaluating it and looking at the past conversations with customer and it's not a three months project and then you're done, you have your bot. It's when you start then it will probably never end because you can make it better. And if you are satisfied with the current accuracy, you are adding new use cases and so on. So maybe that's a little summary how companies should start.
 

Jean:
That's simply awesome. I thank you for that. But before I let you go, I had this little fun thing I ask my guests and this is going to get personal and I'm actually going to ask you to name three things you do most on your phone and they are, are you ready?
 

Malte:
Yeah, I'm ready. I need to think about the three because they are like probably 10 different things I'm using. But yeah, I have two in my mind and the third one is challenging because there are lot. So first thing definitely is my Gmail app for emails, so I think that's pretty, pretty straight forward. Second thing is probably WhatsApp. I think Gmail first and then WhatsApp and the third one is probably Spotify, so I think that, yeah, that's good. That's it.
 

Jean:
You are a serious business person because email, I was beginning to buy into the email equals business communications, but anyway, that was a ton of fun. I thank you for sharing your thoughts and all the hard lessons you learned. Thanks again.
 

Malte:
Thank you Jean.