More Human than Human?: Designing a Conversational UI
This was written for UI22, last year’s UI Conference. If you want to attend this year’s UI Conference. Check out the presenters for UI23.
You can draw a direct line in the UX family tree from User Experience Design back to Human Computer Interaction. What if we could make the “computer” aspect of that interaction, feel less like a machine, and more like a human?
Robert Sens, the Lead Product Designer of the restaurant reservation app, Reserve, sought to create a conversational user interface to help users get seated at restaurants. They settled on implementing a chatbot to simulate the interaction of speaking to a reservationist.
Steph Hay, VP of Design for AI Experiences at Capital One joins us on this podcast to share her experiences in crafting conversational UIs and her insights into Reserve’s approach.
Jared Spool:This is the UIE Podcast. I’m Jared Spool.
When people think of artificial intelligence, it’s usually in terms of the T-800s and HALs of the world. Or in more real-world terms, Siris and Alexas. But AI comes in gradations of fidelity. Siri isn’t HAL, and furthermore a chatbot isn’t Siri, but all have a level of programmed intelligence. People encounter far more of these lower fidelity AIs on a daily basis than they may realize.
The interesting thing is, from a user’s position, these interactions are still very human. That manifests itself in your voice becoming more stern after the third time Alexa says “Sorry, I didn’t understand…”
Users are expecting a comparable experience that they’d have with a human. How do you instill that humanity in something like a Chatbot that handles restaurant reservations?
Robert Sens:The traditional mental model really was this choose and go, right? Instant gratification. A diner would choose a time that was available, and they would get that. That’s a lot of what users were used to. What we were trying to do is have this request model. What this did was mimic what really a diner would do when they called a restaurant, they called a reservationist to say, “Hey, do you have any availability tonight for a party of two?” “Oh, let me check the book.” That idea of opening up the books and that negotiation process of, “I can’t seat you right now, but I could seat you a little later.”
Unfortunately, with that traditional web interface, web flow, that idea wasn’t coming off.
I am Robert Sens, lead product designer at Reserve.
Jared:Robert and his team at Reserve needed to find a way to emulate that back and forth, negotiation-style interaction that a user would be accustomed to when calling a restaurant.
Robert:A lot of our research was going into restaurants and observing how reservationists and how hosts interacted with customers, specifically on the phone, people calling and asking for reservations.
What really struck us as interesting is watching, again, this negotiation process that happens on the phone. It’s very conversational. It’s very fluid. No matter what way a diner comes in, what they’re asking for, what their initial request is, there’s always a way through a conversation to get to the point where you’re either presenting them with what they want or potentially presenting them with an option they want.
A key part of moving toward this chatbot or moving towards that was, how do we capture this conversational dialog? How do we bring this very conversational experience into a digital world without it feeling too robotic? It has to be a little bit about that negotiation and having some empathy for both the restaurant and the user. The diner is coming in looking for a specific thing for a variety of reasonings, and that could be special occasions, it could be I just want to dine, but how do we serve them and give them what they want while also taking into account the restaurant needs, right? The needs for an ebb and flow of the night or the need to fill different tables at different times, not have everyone come in at 7:30.
Steph Hay:They’re totally new UX challenges, like when you think about a beautiful sentence as it exists on paper and it explains everything you need it to explain and then you put it in a chat window, suddenly it feels off because the natural behavior in texting is often just the thumbs and emojis and Y for yes and N for no.
So you realize that a lot of the things you’ve come to learn in designing for GUI need to some degree be thrown out the window.
I’m Steph Hay. I’m the VP of Design for AI Experiences at Capital One. I am teaching a workshop at UI22 in Boston on designing for conversational UIs.
Jared:In a traditional interface, a user interacts with content on a screen and it’s a one way conversation. You put content out and users consume it, and it either works or it doesn’t. But opening the interface for a conversational interaction requires a different design approach.
Robert:There was a lot of exploration and iterative ad hoc research … it sat in two different camps, right? One was, do we completely think about a different way to approach this reservation experience?
The other was, how do we build traditional interfaces starting out with just thinking about, are there different ways to structure this UI? Are there different ways to design around the problem and make people understand that this is a request and that this is how the restaurant works and there’s value here because traditionally where you would just run into a wall, we’re giving you the opportunity to get seated.
What we really found while testing those is that the interfaces that had more human language, that had more what felt like conversational or organic flows or organic process were resonating with people. That was an interesting aha moment. In retrospect, it seems slightly obvious that the process we were seeing in the real world is something that people were used to and that that would resonate with them. However, we didn’t go straight to the chatbot first. We were playing with, again, interfaces, different copy on the screen, “Where are you looking to dine tonight?” Or, “How can I help you tonight?” Those things were really striking with the users we tested with and saying, “Oh, this feels good.” “Yeah, this would be interesting.”
That was the main catalyst for moving toward the chatbot and the conversational UI is, how do we mimic a human experience in a digital way while maintaining some of the human attributes of negotiation and understanding and staggered empathy, right? This idea of, I know what I want and I want what I want, but I know what you can offer, so how do we bridge the gap and make both parties happy?
Steph:In some ways it’s automating the judgment of the booking agent. However, the work that has to go into that again, is so nerdy. We’re all getting into some really deep discussions about how do you manifest judgment? How do you build judgment into a system so that it can actually serve as an intermediary confidently? Because we all have examples top of mind of where something was so automated that it completely cut the humanity out of the equation.
That is the last thing that any of us wants, but is in some ways almost the easiest to achieve is just the automation itself. The judgment is the thing that makes the UI, or the CUI in this case, the AI’s, the intermediary, really special. That takes a lot of work to build into the constructs of the system that’s powering it.
Jared:Those constructs lay the foundation for not only what the chatbot itself is, but also whether it has a specific voice, how it communicates with users, and what users should expect out of it.
You want to mimic the human interaction with a non-human entity. That can be a tricky thing to balance.
Robert:We had a lot of conversations, early on, about potentially naming this thing, a lot of the early Reserve branding or Reserve tagline was, “Your concierge for a better dining experience.” However, the real dynamic here was, we as Reserve are working with you, the diner, on behalf of the restaurant. We don’t really want you to understand this is a bot, and we don’t necessarily want to name it, ’cause again, we’re that middleman, we’re the negotiator that’s helping you work with the restaurant to get seated. We never revealed that we were a bot, I think for a number of reasons. Again, one, like I just mentioned, we wanted to be that voice of the restaurant. And two, we were offering a very discrete service that had a finite start and end point.
Steph:The end to end journey when you think about a service design, mapping that end to end journey, whether it is point A to point B conversation is absolutely essential for any thoughtful design process. In the case of a bot interaction, one of the challenges is that some end to end journeys are really short. They are someone looking for a single answer and when they have that single answer they are satisfied and they go away.
Being confident as a provider that that is effectively a success is tough because what you want to see is the natural reaction that you want to see these long conversations, rich conversations that prove that your design is “engaging.”
Robert:I think a lot of the bots that name themselves are something that you can ask for at any time, like Alexa, I can ask it anything at any time, it’s very on demand. It needs to have that human personality, whereas we wanted to be a little bit more of that middleware in between the restaurant and the diner. There’s a little bit of an illusion there, right? Yes, you’re talking to the restaurant, and are you talking to them or are you talking to us? How do we make that a little shadowy and a little interesting so that you feel like you’re being taken care of? There’s a potential for the restaurant and the diner to have that relationship. That was really the goal of facilitating better hospitality.
Steph:The foundation here is the brand and the voice that pre-exists that we want to manifest in this new entity that we hope takes care of a lot of the things that are sort of the day to day top of mind of back of mind kinds of use cases that might come in and out of your head as you’re just going through your day.
Those are the kinds of daily behaviors that a bot is really well positioned to serve. The fact that it’s always with you on your phone and that you can ask during a meeting where it comes in and out of your mind, that is the sort of always on, everywhere, accessible, 97% of people who have smartphones have used text messaging, that is the ubiquitous nature of designing for this interface.
Jared:That ubiquitous access added to some struggles for Robert and his team. Restaurants keep specific hours. Prior to the conversational approach, Reserve’s users weren’t sure what was happening with the app. The Reserve team had to gauge what users’ expectations were and appropriately set the context.
Robert:The factors we were working up against was this problem of waiting, right? Again, we were working with the restaurants directly. We were the intermediary between the consumer, the diner, and the restaurant. Inevitably, there was a slight bit of waiting time there, a little bit of context setting that we’re actually talking to the restaurant. We, Reserve, are a company you’re working with, but we’re actually your conduit to the restaurant. You might have to wait, the restaurant could be closed. At primetime it might be harder to get in touch with someone. There’s some context setting there.
A big catalyst for the conversational experience was that, right? How do we just set that context that maybe an answering machine or not talking to someone can’t set. “We’re working with the restaurant. We received your request. We really value it. It’s just gonna take us X amount of time or a few minutes,” or, “While you wait, what can we offer you? Are there other restaurants you may want to get into? Can we get you seated somewhere else? Being that intermediary and attempting to set context and help the user get seated either at the restaurant they want or a comparable experience.
Steph:The second layer is setting the customer’s expectations when the particular use case is highly nuanced, highly contextual to the customer’s behaviors or their goals or their particular financial situation or their geography for example.
It would reset the status quo, having the context and having the data available to us to be able to make decisions about what to say when to whom is almost necessarily a design challenge because we need to have restraint. We need to have empathy. We need to have judgment.
Jared:Humans are conversational. The majority of their interactions in any given day are conversational. Designing interactions to align with what customers have in their day-to-day activities and experiences can have some interesting effects.
Robert:We had a general idea of our flows. We knew the general dialog that a restaurant used on the phone, so I think a little bit of the assumptions we made around tone were potentially erroneous at the beginning. There is an interesting give and take on the phone, that immediate back and forth, and also a little bit of the context around talking to someone in person that, I think, lends itself to an understanding, on the diner part, that they’re asking someone for something and that they’re waiting for an answer.
In the digital world, that was also true. However, we couldn’t be as punchy and direct with our dialog. We had to use a little bit of a softer tone, maybe a little bit more fluff, again just to make people understand, make them feel welcome, have a little empathy there, whereas on the phone I could be, “Well, I don’t have that time, how about this time?” Obviously, we couldn’t necessarily say it that direct. We had to say, “Well, unfortunately, we do not have 7:30, but we can offer you an 8 p.m.” Again, adding a little bit of that fluffy dialog and a little bit of that voice that felt more like a concierge, felt more hospitable, was a learning that came quickly, but not something we had expected in the beginning.
Really, one interesting learning was just looking at the things that people said, again, after they submitted that request and they got to the point where we let them know they were waiting. A lot of it was very gracious, like, “Thank you,” or like, “Great, can’t wait to hear.” Just understanding the impact of that experience, I think, could have really changed the way we potentially thought about structuring the whole experience, right? That goes back, not even just from the interface, but thinking about how that affects the business as a whole, right? How do we work with our operations team to better empathize with our users? Can we leverage that human tech experience in a different way to facilitate better hospitality, better engagement, or potentially even different experiences with the user?
Jared:Choosing the right tone can instill trust within your users. But like many decisions, there is a potential unintended consequence. Human nature is variable. Depending on where someone lives, or where they are in their life, they could have different expectations. And those expectations could be further complicated by the company’s brand or identity.
Creating an interaction that feels human while remaining universal brings with it some deeper questions.
Steph:You can’t even imagine the number of variables, right? But like regional dialect is one example. Age, there’s also the behavior that folks exhibit in channel. Like my dad, when he sends me text messages, he writes Dear Nooch, which is my nickname that he gives me. Then he signs it Love, Dad.
Every text comes from him like that, like it’s an email and I love it. It’s like my favorite. This is to some degree a business question more than it is a system question, at least at first, which is who are the audiences we want to speak with in a really complementary way and to what outcome? Because there are tons of research out there about a way people want to interact with other people and what they’re drawn to. There’s similarity bias, et cetera.
But interestingly in the research we’ve done so far there is a not actually so fine line between what a customer exhibits in their channel or in their behavioral language and what they expect from the language of their bank in our case. So if we started signing our emails Love, Grandpa Jimmy or whatever, that would actually raise a huge red flag for a lot of our customers. It wouldn’t feel like the brand that they’ve come to know and rely upon.
Even though our system, we could, by looking across a variety of data, make some assumptions about how to go about speaking to that person because in the Midwest they call it pop and in the east coast they call it soda so therefore all of our stuff that goes to Midwesterners should say pop. Those are actually, interestingly, moments that from a business perspective start to enter us into conversations about how we’re showing up differently depending upon those customers on aspects of our conversation that may not actually drive a deeper level of trust or connection.
On one hand you could argue that well yes, of course it would because it shows that we understand who they are and where they are and what language they use. This is a journalist thing. That too by the way, right? But on the other hand, if it gives somebody pause for a moment to say huh, that’s interesting that they used language or I wonder why they did that, then ultimately it may not actually outweigh the potential benefit of the connection because it’s actually raised into question why we were doing it in the first place. It’s not coming across to that person.
So this whole field of anthropology, especially for bot systems, is raising both the sort of business brand trust discussion as well as a system level delivery discussion, which is how do we make sure that our language complements and builds trust with our customers versus just reflects what they use.
Robert:What really surprised us is how quickly things get extremely complex. As we’re mocking up these Gator boards and these conversational flows, it’s just amazing how much we’re blossoming out into very complex storylines and how much conditional logic becomes a huge X factor of how we structure our flows.
Steph:It is a challenge to imagine designing a conversation that could go anywhere at any time. Everything that we know about information architecture and our human nature about that to put things in a bucket, to categorize things and make things fit together suddenly has exploded and everything is atomic level. Anyone can go anywhere. Sure, there are going to be some patterns, but designing the happy path is no longer the priority, you know? Designing UX constructs into a system that can handle any path and every path is a happy path is the priority.
Jared:User experience design is ostensibly, designing for humans. Humans are communicators, storytellers, texters, tweeters, and posters. Meeting users within the context of their normal methods of communicating creates interesting opportunities to allow them to stay within familiar flows while experiencing your designs.
Building conversational UIs, whether they are voice-based interactions, chatbots, or simply structuring your content in a more conversational way, makes designing for humans, a little more human.
This UIE podcast is brought to you by the UI22 conference that’s happening November 13–15 in Boston, MA.
I’m very excited about Steph Hay’s full-day workshop, Designing for Conversational UIs. She’ll spend an entire day sharing her approach to crafting compelling content to structure interactions that feel more like a conversation, whether you’re designing voice-based interactions, chatbots, or any kind of app.
You definitely want to check out her workshop’s full-day agenda. You can do that at UICONF.COM. That’s U I C O N F dot com. Use the promotion code HAYPODCAST17 for $200 off your full-conference registration.
Now, if you haven’t been to U I E dot F M lately, then you may have missed our recently revived podcast show, the U I E Book Corner. In this show, Adam Churchill and I talk about new user experience books that should be high on every designer’s reading list. Listen to the U I E Book Corner at U I E dot FM.
This UIE podcast was written and produced by Sean Carmichael. We’d like to give special thanks to Robert Sens and Steph Hay for appearing on this episode.
This podcast is part of the UIE Podcast Network. Thanks so much for listening and, as always, thanks for encouraging our behavior.