Understanding Behavioral Bot Design
Fabian Reinkemeier – Elaboratum
About this episode
In this episode we speak with Fabian Reinkemeier from elaboratum, a firm that advises customers on eCommerce and technology projects. Fabian is a consultant and PhD candidate in the field of e-commerce and conversational AI, with experience in project management, usability & user experience, research methods, and testing.
Today we’re discussing with him the design of conversational bots.
You might also want to listen to…
- Episode #29AI GuidesJohn Stroud and Jen Shellinck
Podcast #29 – Why Do So Many AI Projects Fail?Listen now
- Episode #28PhoneMyBotLivio Pugliese
Podcast #28 – Voice bots and IVR Systems – The Future of On-Call SupportListen now
- Episode #27Mitsubishi Chemical Advanced MaterialsRoy van Griesven
Podcast #27 – Customer Service Digitalization in Manufacturing IndustriesListen now
- Episode #26HeadlessCreatorMarcelo Lewin
Podcast #26 – Headless Technology and the Future of OmnichannelListen now
Jordi Torras (00:02):
Hello, hello, and welcome to the Future of Customer Service podcast. Today, we have Fabian Reinkemeier from elaboratum who is going to tell us a lot about intelligence for chatbots interactions and customer service. So Fabian, thank you so much for being with us today.
Fabian Reinkemeier (00:34):
Thanks for inviting me.
Jordi Torras (00:36):
Absolutely. Tell us about you, Fabian. What do you do at elaboratum?
Fabian Reinkemeier (00:42):
I work as a digital consultant there. It is a specialized e-commerce consultancy firm called elaboratum in Germany. And there I mainly deal with the topic of conversational user interfaces, meaning chatbots and voice assistants. We work in an evidence-based way and always want to be up to date with the latest developments and findings, so I’m also doing research on the exciting topic at the University of Göttingen.
Jordi Torras (01:15):
Wow, that’s amazing. So I like the idea of using science in the space of artificial intelligence and customer service because there’s a lot of hype around the space. So I have read about your work and there’s something that particularly intrigued me, which is the idea of behavioral bot design. What is that? What is behavioral bot design?
Fabian Reinkemeier (01:45):
Yeah, perfect. Let’s jump right into it. Behavioral bot design means for us to optimize chatbots and voice assistants with behavior patterns. At elaboratum we have a behavior design unit to do this, also for websites. And this is a team of experts focused on applied behavioral economics in the digital world.
Fabian Reinkemeier (02:09):
But let’s break it down. First of all, what is behavioral bot design? For us, it’s important to understand what makes people tick when they make decisions. So, to our podcast listeners, and everybody, go ask yourself, how do you think you make decisions? Rationally? Intuitively? Consciously? Unconsciously? I know most people will probably say rationally. But is that really the case? Sorry, I have bad news. We make our decisions unconsciously. Okay, now some of you might think, “That does not sound like me at all. That might be true for some people like that one colleague that always decides so spontaneously or that friend that uses always their gut feeling to make decisions. But not me. I’m a rational person.”
Fabian Reinkemeier (03:05):
But trust me, I really have to disappoint you here at that point. I know you probably wish to be like Mr. Spock from Star Trek, who thinks rationally. But we are all more like Pippi Longstocking. We think intuitively, instinctively. But the good news on that is there’s research.
Jordi Torras (03:29):
Wow, so that’s an interesting approach. So when designing artificial intelligence and bot conversation, you need to think about the audience not as a rational entity, but as some sort of an emotional organism, which is what we are as humans. So tell me, what is this theoretical perspective that you mentioned about this irrational or this subconscious behavior from humans when interacting with chatbots?
Fabian Reinkemeier (04:09):
Yeah. So if we put on the theoretical perspective on this, the psychologist Daniel Kahneman, many may know him, has made groundbreaking findings on this topic and received a Nobel prize for the theory. We humans have two different decision-making systems in our brain. We have system one, which is our intuitive, instinctive decision-making system. It’s unconsciously, fast, doesn’t consume any resources. We can call it autopilot. And then we have system two. This is a rational, logical thinking system. It works unconsciously, slowly, and consumes many resources. We can go it our internal computing machine.
Fabian Reinkemeier (05:00):
Looking now at the split between these two systems, research shows that 95% of our decisions is actually made by our intuitive system. And only just like 5% of our decisions are made by system two. Why is that? We are making so many decisions each day. As a consequence, we base our decisions on what we know as behavior patterns. These are like psychological patterns that can be understood as templates or pathways to our actions. That’s the theoretical point.
Jordi Torras (05:39):
I see. So to make sure I understand, we have these two systems. The first one, the rational system, is actually the one that consumes more resources? So it makes sense for us to make decisions emotionally because it consumes fewer resources. So it makes sense from an economic or, let’s say, energetic point of view, to make decisions emotionally. And then those decisions that are more complicated, we make them rationally because those are going to consume more resources. That makes a lot of sense. But tell me, how has that anything to do with chatbots?
Fabian Reinkemeier (06:18):
So we take this knowledge and this need of customers into account when designing chatbots. From developing user persona to the selection of suited behavior patterns, it’s always important to make the decision process as easy and intuitive as possible for the users, because 95% is intuitive thinking.
Fabian Reinkemeier (06:45):
And we actually did a study on that. We challenge a non-optimized voice bot against an optimized voice bot with a behavior pattern. The goal was to recommend books to the user. And the guiding theme for us was social factors. So we had this non-optimized bot, which was really transactional, using only minimal vocabulary, actually getting the transaction really done as quickly as possible. And then we have the optimized bot on the other hand, in which we implemented behavior patterns, which had an extensive personality and a more personal style of speaking. It would pause like a normal human being would do before answering. And we had great results on that actually.
Jordi Torras (07:47):
Wow, wow. So if I understand correctly, you say, “Okay, this is your chatbot but it’s not optimized from a behavioral design perspective. So let’s optimize it. Let’s take that bot and let’s optimize it. Let’s make it ready for our audience because they will make intuitive decisions. That’s fascinating. And what are the advantages? Why is that a good thing?
Fabian Reinkemeier (08:26):
So, after all, it improves the quality of the dialogues and the user experience. We really want to achieve a natural conversation. Language is the most natural way of interacting. And we have this chance with conversational user interfaces. So we take that and we build an intuitive path for the customer.
Fabian Reinkemeier (08:55):
And for example, in the study with the voice bot, we had great results. The study showed there was 80% higher purchase intention rate. I think there was around 40% higher user satisfaction with the optimized bot. We’ve applied that many times after that and we keep seeing it’s working. We knew that before from optimizing websites and now we’re using this method to optimize chatbots. So yeah, it’s not only about the technical part, it’s also about the design.
Jordi Torras (09:34):
I get it. I get it. This is when you have this aha moment. Okay, the same idea that’s been applied now for decades on designing user interfaces and user experience, it’s the same idea but applied to conversations, having conversations with a behavioral design. And that, you said increases by 80% the purchase desire, right?
Fabian Reinkemeier (10:06):
Jordi Torras (10:06):
Because very often purchase decisions are intuitive or let’s call it irrational, sometimes. That’s amazing and really fascinating. So could you tell us a little bit more, or maybe some ideas or patterns or some of the points that are part of this behavioral bot design?
Fabian Reinkemeier (10:32):
Yeah, actually, as there are so many great patterns, my colleague, Philipp Spreer, even wrote a book about it with 117 patterns, which is on its second edition now.
Jordi Torras (10:44):
That’s a lot of them.
Jordi Torras (10:45):
Okay. Could you tell us three patterns that we should always remember?
Fabian Reinkemeier (10:52):
Yeah, I will give you an example, actually from a real-world example of a customer. I will just give you a bit of background because behavior patterns are not applicable to every chatbot. You always have to consider what’s your target group, what’s the use case. So we had the opportunity to create a chatbot for a backpack manufacturer in Germany. And the use case was to guide the user through the product search process, just as an experienced salesperson would do in real life. And we analyzed the user behavior, we interviewed the sales staff. And we determined, again as a guiding principle for the chatbots, a social fabric. And with this in mind, we began the design process. We gave the bot a cheerful appearance with some outdoor background. So again, two behavioral patterns.
Fabian Reinkemeier (11:47):
Firstly, we implemented the behavior pattern called ‘Likeability’. We tend to have more trust in people and processes when we find them likable and friendly. Consequently, we started the conversation flow with a happy, waving chatbot and the statement was, “Great that I can help you today.” And by doing so, we set the tone for the entire communication. Both are really working towards the same goal, but the first one is likeability.
Fabian Reinkemeier (12:23):
The second pattern is the picture superiority effect. This describes how we remember things much more easily when we see a combination of text and image. So by giving the bot a relatable character, people are not only much more likely to remember it but also to come back or even recommend you to others. That’s the second one. So second one is the picture superiority effect.
Fabian Reinkemeier (12:53):
And the third one, I would point out, is the primacy effect and it’s actually applicable for every chatbot. It means that the first information sticks to our memory. We all heard the saying ‘the first impression counts’. And this does not only apply to first dates or meeting the in-laws, but also for the first contact with a chatbot. So, in our case, we didn’t just set the tone with the cheerful greeting and the shared goal, but also the fact that the bot advice is based on 100 plus years of sensible expertise, which by the way also reinforces the bot’s authority. So the third pattern that you should always remember is the primacy effect. It’s really important for all cases.
Jordi Torras (13:50):
Okay, okay, that’s amazing. So there are these three specific takeaways here. One of them is the image. And I believe that’s something that’s been around for a while or the idea of avatar and chatbots with avatars. And sometimes our customers say, “Hey, should we use an avatar?” And I was never sure. But now you have here some data and scientific way to say, “Well, actually it is important.” And that makes a lot of sense.
Fabian Reinkemeier (14:22):
Yeah, but it’s really the combination of it. The combination of having text and reinforcing the things you want to say.
Jordi Torras (14:37):
Got it. With consistency. There has to be a consistent language, with an image, with all these patterns that create the emotional connection.
Fabian Reinkemeier (14:47):
Jordi Torras (14:47):
That makes a lot of sense. Listen, you know at Inbenta we work with artificial intelligence. And because we do, we know it is not perfect. So, sometimes you’re going to ask a chatbot and the chatbot might not have an answer for your question or that chatbot might not quite understand the question. So what happens when the chat fails? What would be the best course of action when you have your chatbot not actually being able to answer a question?
Fabian Reinkemeier (15:22):
So yeah, if you look at the market, there’s one common strategy. It’s just saying, “I’m sorry, I didn’t understand you.” Or, “Please try again.” That’s actually all.
Jordi Torras (15:34):
That one you will see most of the time. “I didn’t get that. Try with fewer words” or something similar. That’s what you will see in many, many chatbots today.
Fabian Reinkemeier (15:47):
Yeah. Actually, these messages cause frustration because they do not help the customer understand why the chatbot failed and more importantly how they can repair the situation. So we see this common problem with technologies. Many chatbots are developed from a technical perspective. They don’t sufficiently integrate anthropomorphic design, a more intuitive design, which is the key to customer-centric success. Again, we can look actually at human-to-human communication. You won’t see this kind of message. For example, Jordi, if I tell you, “We have to meet ‘Laura'”, and you didn’t catch the name or what I said, what would you say to me?
Jordi Torras (16:42):
Could you repeat that, please?
Fabian Reinkemeier (16:47):
Yeah, that’s actually really specific, but actually in the first thing you were like, “Huh?” And that’s sometimes the first thing humans are doing like, “Huh, what were you talking?” Then, if I tell you the same, you would upgrade your strategy, like repair strategy. You would repair the situation. So you would say, “Who are we going to meet?” And I, as a speaker, will rephrase myself as well. I will maybe just say Laura or a totally different sentence. So, what we can learn from human conversations is that they are not as simple. So you shouldn’t just say, “Huh?” and then repeats the same, huh again, which is what the chatbot is saying. So yeah, I know some chatbots after two or three fails, give way to a different channel. But that’s not a strategy really. You should try to change the strategy actually. So it’s possible and we know from research that you can have behavior changes when you interact with chatbots. So our goal is always to develop constructive, user-friendly failure messages for chatbots and build an optimal chatbot repair strategy.
Jordi Torras (18:18):
Oh, that makes sense. I like this idea of repairing. Like something has been damaged, in the sense that there’s a miscommunication with the chatbot. Something has been damaged and then the repairing, it’s absolutely important. All right, let’s shift the topic a little bit and just talk about Inbenta because I believe that you guys know us a little bit, we’re working together. So what is your experience working with Inbenta and what kind of results did you get? Do you think that someone can be kind of programmed or configured with this behavioral bot design?
Fabian Reinkemeier (19:10):
So yeah, first of all, if you are active in the field of chatbots, you need to be familiar with Inbenta, please. Because you’re one of the big players. So we know Inbenta, we also know other technologies. Which is great for us because we can always see what’s the advantage and disadvantage. It gives us perspective. Yeah, I would say everyone should know Inbenta if you want to build a chatbot. And actually, with a good customer of ours, a leading insurance company in Germany, we have already created several chatbots with Inbenta in the last two years with great results.
Jordi Torras (19:54):
Nice, nice. All right, thank you. So we have a customer in common. And that’s great because that kind of network is amazing. We meet customers that introduce us to great companies and teams like yours. Teams who have thought a lot about conversational AI and help make it work. So that’s amazing. And listen Fabian, what do you think is the future of these conversational AI? Where are we going with all that? What do you think?
Fabian Reinkemeier (20:28):
I believe that the decade of conversation is beginning. We are now moving more and more from omnichannel to conversational commerce. And this is good because conversational commerce offers unique opportunities such as ordering products online in a really intuitive dialogue. From an evolutionary perspective, it’s more natural. And maybe it’s like what Elon Musk would say: the perfect interface is to communicate only with our thoughts. The idea still seems a long way off. We see actually with recent advancements in AI that natural human-computer interaction is now possible. It gives us the interaction to move away from just exclusively, unnatural graphical user interfaces and more to taking customers by their hand, supporting them using an intuitive natural language dialogue, and provide them with a better customer experience. In the end, this is what is all about, together with having higher conversion rates.
Jordi Torras (21:45):
Absolutely. Well, and I think that the way humans have been interacting with one another is through conversation. And it’s been like that apparently for hundreds of thousands of years, right?
Fabian Reinkemeier (22:01):
Jordi Torras (22:01):
So, for us, the most natural way to interact is through natural language, which is English, German, Spanish… and through a conversation. So if we could make computers efficient, and I believe that what this behavioral bot design concept makes is make a greater step towards that target. So Fabian, what would be the main takeaways here for our audience? What will you tell them when it comes to designing their chatbots? What do you think they should know?
Fabian Reinkemeier (22:49):
Okay. I will give you three takeaways.
Jordi Torras (22:51):
Fabian Reinkemeier (22:51):
So first, always think about how your customers think. Remember that technology is great, but you also need to focus on customer needs and understand how they think and act. 95% of our decisions are intuitive, don’t forget Pippi Longstocking, and embrace the customer way of thinking. The second one is, think about your design. Don’t just look at these projects from the technical perspective. When building a truly customer-centric, natural conversation bot, a good solution is to use behavioral bot design. And also think about your chatbot repair strategy when the chatbot fails. And thirdly, always remember what’s your use case. So choose your use case and the AI wisely. Ask your customers, and also your employees if you introduce chatbots for intern goals, determine what they actually need.
Jordi Torras (24:01):
Totally, that makes sense. That’s three great takeaways for our audience. And I think they will appreciate it and I really appreciate that. All right, Fabian. That’s been a great discovery here today with you. What I would ask is if our audience wants to learn more and wants to contact you, what is the best way to reach you?
Fabian Reinkemeier (24:31):
So yeah, definitely, everyone, if you want to have a deep dive into a specific topic, or have just a chat about a certain problem, just get in touch with me, connect with me on LinkedIn or write me an email at [email protected], or Google my name. But yeah, the easiest way is simply to connect with me on LinkedIn and I’m happy to meet you.
Jordi Torras (24:55):
Awesome, awesome. Totally. All right, Fabian. Thank you so much. It’s been an exciting episode of the Future of Customer Service podcast. I appreciate having you with us. And for the audience, we’re going to work hard to have really smart people around talking about customer service, AI automation, and many other topics. Thank you so much, Fabian.
Fabian Reinkemeier (25:38):
Jordi Torras (25:39):
Take care. Bye, bye-bye.
Fabian Reinkemeier (25:40):
Thanks so much for tuning in. This podcast was brought to you by Inbenta. Inbenta symbolic AI implements natural language processing that requires no training data with Inbenta’s extensive lexicon and patented algorithms. Check out this robust customer interaction platform for your AI needs, from chatbots to search to knowledge centers and messenger platforms. Just go to our website to request a demo at inbenta.com. That’s I-N-B-E-N-T-A.com and if you liked what you heard today, please be sure to subscribe to this podcast and leave us a review. Thank you.
Do You Want to Build a Successful Chatbot?
Download our ebook!
Get your guide to build a successful chatbot and make sure you start the project off on the right foot.
Read about tips and best practices to frame your conversational AI interface and how to define goals, KPIs, the scope of the chatbot, and the team that will take care of it.