Voice This! Podcast: Episode 5

Prototyping with Robert Bruchhardt

How do you prototype voice products? How do you prototype in-car designs? What goes into the selection criteria of “the best idea”? To answer these questions and more, Robert Bruchhardt will take us through how to prototype concepts of Conversational AI in the automotive industry!

Defining a concept through prototyping

[10:12] What is prototyping and how does that make sense in the context of voice?

I would define it in a way that you basically think about a product, right? You have an idea. Usually you start with a concept that kind of sort, but in order to make the next step, right, how did you get from concept to the final product? You usually build a prototype, which is something, or actually multiple prototypes. We come to that later, but it’s basically the first zero dot something version of your product in which you try out some things, right. And they might fail, or they might not be, , or in the final product at all. But at the same point in time, you have a first version, which you can use for testing, which you can use to get the real experience outside from just this conceptual and theoretical approach.

[11:38] What does prototyping look like for conversational AI?

Technical approach: for me actually, as a more technical person, and I would also urge anyone who’s listening right now, and we’re saying who was considering themselves, a developer that are prototype can almost always be the real thing, you know, like really creating a first voice application, using the STKs where all which are out there.

Wizard of Oz approach: for folks who are not that technical, I think voice is also great area to prototype. One of my favorite approaches, there will be kind of the Wizard of Oz approach, which is basically you don’t have any applications running. For example, you use a chat bot, let’s say, you have a chat bot and you have that one person, , chatting with you, unaware that it’s actually you, but thinking that it’s an AI conversational AI, And the nice thing then is, , they would just type in what they would to a conversation at AI. And you can reply basically on the fly. Usually you would prepare some prescriptive answers, but, , you can reply on the fly, , acting as the conversational AI and, that involves basically zero technical skills, zero coding, but still you get the perfect, real life application on how users would interact with your application and, how users react to actual responses you are giving to them.

The prototyping process

[17:15] When does prototyping start?

Start prototype much early on: With voice who really have that advantage that you can prototype much earlier on in your process. So sometimes it’s easier for me to create a quick voice prototype than to actually create the concept behind it in more detail, or to create a PowerPoint slide to present it or something like that.

Get to it as soon as I have a promising idea: so when it comes to prototyping, I can only urge anyone who’s listening to this right now. I would get to it as soon as possible, as soon as I have a promising idea.

A prototype offers a much better understanding of your idea than any concept or any PowerPoint presentation: Because if you do that and you have a working prototype working demo, and even if you just, like film yourself, interacting yourself with a prototype, that’s often a much better pitch and often a much better, , it offers a much better understanding of your idea than any concept or any PowerPoint presentation or anything like that can do.

Iterate over and over: and then of course you can always iterate over it, right? That’s the beauty of software. Nothing, nothing is set in stone there. But at the same point in time, voice technology right now really gives you, , that, that really, really quick way of creating your first prototype. And that’s what I would definitely do.

[22:41] How long should the prototyping be? Is there any certain timeframe?

I would say it depends a little bit on the use case… it cannot hurt to establish some KPIs and then iterate over your solution until you kind of like reach the desired success rate you’re aiming for.

[28:30] What do you look for when you prototype your products?

Time spent to get to the thing which the user wanted:
I would say a couple of results I’m always looking for is time spent to get to the point or to the thing which the user wanted.

Productivity comparison between voice interface and other interfaces: So, it’s not only about, okay. I finally made it there after 10 minutes of interaction with voice assistant, but also is it, for example, much quicker than just having a button on this graphical user interface. So really that comparison, what kind of like added value does voice give us and this again, like going back to this hammer thing, right, is this really a good use case for voice or are other things or other sensors or any kind of information can be used to achieve the same outcome in a short amount of time? That’s always something I’m looking out for, especially coming from the automotive area.

[31:07] What does crowdsourcing look like?

Beginning of your project: Take your time to observe: So, for example, there is this really like time sensitive approach where you really take your time and really observe people, or you have someone external observe people for you. Right. And you just think about how could the test set up look like, and this could, for example, be this wizard of Oz approach. Right. And then usually you are really in the beginning of your project, usually in a state where you’re not quite sure yet on how the happy flow for example looks like, like, right. What’s the thing people would expect? Like how, how would they expect to interact with the assistant and so on? So then usually you spend a little bit more time on, and then at the same point in time, when you move further, you can standardize a couple of more things and you can involve more and more people.

Scalability: beta test program: If you really want to come up with an approach where you involve a lot of people and you want that scalability, you probably end up with having that kind of like beta test program, but you invite a couple of people to use it, for example, on their smart speaker to use it on the chat bot app on their mobile phone, and then try out a couple of test cases and then submit a feedback form later to you.

[21:10] How many people do you prototype with?

It kind of depends on the use case, I would say but honestly the more, the better, this is where I would definitely, especially if you’re in a business context, right then I would definitely always check out crowdsourcing possibilities. There are a bunch of platforms out there who are using crowdsourcing approaches that means you can give them test cases. You can give them your prototype. This could be, for example, the chat bot or this could be the invitation to a specific beta test program from your voice assistant or anything like that. And then you try to distribute that through this crowd sourcing to a lot of different diverse people, right from different countries, depends on where you want to release your final product, with different languages and so on and so forth.

[44:00] What do you do when you have unexpected answers from users?

There will probably crash or it will maybe handle it if you already have that in we’ll handle that error in some way… You need to make sure you got all of these responses and you treat them as they should be treated, which is the user did something you didn’t expect at all in that flow and you should tackle it in a future release.

I would say, again, Wizard of Oz. You can just quickly react to it and that’s why it’s so great to begin with a Wizard of Oz approach.

And then secondly, if you have your first prototype out there and you get to the situations and they will happen, right. And just make sure to catch all of those. Right. And don’t skip it. For example, if a user then gives the right answer as the next step, because it kind of looks like, Oh yeah, of course it wants to know this and this still take it as a hint to redesign it at this point.

Automotive protyping use cases

[45:35] An example of identifying use cases

Sensor vs. Voice: Let’s say it’s dark. You’re in your car and you put some stuff maybe on your passenger’s seat and you try to reach over like finding something you put on the passenger seat, but you cannot find it. Right. So what do you prefer? I do prefer to just say something like, Hey, Mercedes, turn on the lights or in my car, or turn on the passenger seat lights or something like that. Or the other thing is we have the sensors to know that it’s dark right now. We know that no one is sitting on that seat. Right. And we know that you are reaching over because we are monitoring your hands in the car. So we just turn on the lights for you without even saying it. You know, these are the, these are the things you also need to consider.

Not everything should be solved with voice. Sometimes it’s nice. And even more magic. If you think about other sensors and other things you could use to fulfill what you want to do. So as much as I like conversational AI, And it’s my favorite topic, my passion right now, as much I sometimes read to remind myself, to take a step back and think about, Hey, is it actually does this actually make sense?…Also for me coming out of this voice industry is sometimes need to take a glimpse in other technologies.

Conversational AI in the future

[53:45] Conversational AI in the future, what does that mean to you?

Seamless integration of different ecosystems: I think now it’s all about this seamless integration of all the different ecosystems and all the different players and, probably everyone who has their own smart home, or who has more than one voice assistant or different ecosystems on their phone and with a voice assistant and on their, computer maybe, It’s all so separated stuff, right? Like this thing has my columns and they’re all linked, but this thing doesn’t, and I have these different smart home devices, they all from different providers. So I think a lot of this in the future will be about aggregating those things or getting them all together under one umbrella, getting the different ecosystems, working well together.

About Voice This! Podcast

Conversations with the people who make conversational AI 🎙️Join Millani Jayasingkam and guests as they discuss voice technology, conversation design, AI trends, and the strategy of creating effective conversational experiences. Tune in for first-hand learnings, insights, anecdotes, and sometimes jokes! Say hi and send us your questions at: voicethispod@gmail.com.

Find us wherever you get your podcasts or Listen on Spotify!