Designing for conversational user interfaces: A Q&A with Stratis Valachis

6 min read Max Koh

Stratis Valachis, senior user experience designer at Aviva’s Digital Innovation Garage, took some time out of his busy schedule to answer some questions about designing for conversational user interfaces (CUI). Learn more about his processes for research and design for CUI, what he thinks the future will look like, and some of the biggest challenges he’s faced while designing for CUI. 

Stratis will be speaking at MUXL2017, the third annual conference around Mobile User Experience in London on the 10th of November at City, University of London. Using case studies through talks and workshops, the conference will cover Core UX principles as well as emerging topics such as AI (Chatbots), VR (AR) & IOT.

What does the research and design process for conversational interfaces look like? 

Like any design project, you should always start by identifying user needs and real problems. Research how users solve that problem currently and then evaluate for which use cases you can remove friction and enhance the experience by utilizing a conversational interface.

Don’t try to chat-ify or voice-ify your product just because it’s a cool trend. In many ways conversational interfaces (CUIs), both voice and visual, have more usability constraints than traditional GUI. For example, it’s hard to interrupt the conversation to recover from errors, you can’t easily skim through information, progress is linear and you very often need to rely on recall.

Users make conscious compromises about which type of interface they want to use.

This means that a solution utilizing a CUI needs to offer an obvious benefit for your chosen use case, otherwise users won’t use your product. That’s why special emphasis should be placed on early research about the context in which users will use your product and on why a CUI could provide a better experience. 

When you begin the design phase, a good practice would be to craft a personality for your interface. Studies have shown that because humans are empathetic, they will assign human character attributes to your CUI anyway, so it’s better to make sure this is defined through design. This works really well for platforms like Google Home and Facebook Messenger, which make it clear to the user that each product built on them is a different entity from the default assistant.

Some channels like Alexa, though, don’t make that distinction clear. In these cases, you need to make sure that the character of your CUI doesn’t significantly deviate from the personality of the default assistant, otherwise you’ll mess with their mental model and create confusion.  For example, when you’re ordering an Uber with Alexa, it’s Alexa that speaks back to you: “Alexa, ask Uber for a ride.” “Sure, there’s an Uber less than a minute away, would you like me to order it?”. While on Google Home, the Google Assistant makes it clear that it passes you over to Uber “Hi, I’m Uber, how can I help?”.

After you define the personality, start drafting out the core experience of your product.

If you’re working on a visual CUI, type the conversation down like a screenplay. If you’re working with voice, act the dialogue out with your colleagues and use voice simulators to see how it feels in the channel you’re designing for. This will make it easier to decide the direction you’d like to follow and will also help you initiate conversations with stakeholders.

At this stage, you will be ready to start designing your user flows to define the functionality at a granular level. Again, understanding context is crucial. Make sure you think of the different scenarios in which users will interact with your product and the ways they’re likely to phrase their input. User testing is key for this.

What are some of the biggest challenges you’ve faced designing for CUI?

Setting the right expectations for users. That applies to both visual and voice interfaces. There’s a gap between the mental model users have of what most AI products with conversational interfaces can do, and what they are actually capable of doing. That was a common pattern I’ve seen in user testing sessions even with users who had previous experience in the conversational channel that was being tested. As a designer your challenge is to make the affordances and constraints clear in a way that feels like a natural part of the conversation and mitigates disappointment from unrealistic expectations. 

Another challenge is trying to cater for all the different ways people will phrase the requests. The key here is to invest time and resources in user research and NLP (natural language processing) services. If you feel that this is out of scope for your project, you may consider limiting the options for your users as trying to guide them to say things in a certain way will not work. Good examples of this are Facebook Messenger bots which now allow developers to remove the input field entirely from the experience in order to prevent users from making requests that can’t be supported.

How do you think CUI is going to change the way designers and researchers do their work?

It might require designers and researchers to slightly alter some techniques they’re using (for example thinking aloud during user testing doesn’t work with voice interfaces) but the fundamentals will stay the same. You still need to focus on understanding the problem, explore different solutions through divergent thinking, converge, develop and continuously iterate based on user feedback. The exciting thing is that these new technologies significantly expand our toolbox and offer new interesting ways to solve problems for our users. 

What improvements to this kind of technology do you wish to see? How would you like this technology to progress in the future?

I would like to see a more widespread integration of voice interfaces with visuals and GUI interaction patterns. A good example of the benefits of this approach is Amazon’s Fire TV. Users can converse with the system via voice when it’s more efficient than the alternative interaction options (for example, searching for a movie) but use their remote control to interact with visual UI elements for tasks that would be tedious to perform through voice. For example, selecting a movie cover to reveal descriptive text and then skimming through it helps you gauge whether the plot is interesting faster than if you had to consume this information through a conversation. This hybrid approach utilizes the best of each world to create a stronger experience. I think we will see this type of interface a lot more in the future. Think of Iron Man and J.A.R.V.I.S.

Any advice for young designers and researchers hoping to get into this part of the industry?

Invest time in learning best practices for crafting good dialogue. It’s a crucial skill for designers in this field. Google and Amazon’s design guidelines are a good starting point. This doesn’t mean you should omit training and improving your knowledge in usability for traditional interfaces. Most of the principles are time-proof and channel agnostic and will help you greatly with conversational interfaces.

Another thing you should make sure you do is stay up to date with the latest trends. The technology evolves very fast so you need to stay ahead of curve. Attend meetups, work on personal projects and participate in hackathons to practice and learn from the experts.

As long as you’re really passionate about the field, there will be plenty of opportunities for you to get involved and contribute. We’re still in the early stages of mainstream adoption of the technology, so we have the chance to make significant impact on the evolution of the field and shape best practices for years to come, which is really exciting!