Designing impactful conversational products with defensive design
Designing good conversations in digital products is a challenge. Digital conversations have never been more important. As I write this, most of the world is in lockdown due to the coronavirus pandemic, and most of our conversations have moved online. Being able to talk with friends and family over video and text is a blessing but it’s not the same as having a conversation in person. We are, I think, all missing the human connection you get from being ‘with’ someone, talking to someone face to face.
Being in the same room is something digital will never be able to replace. But conversation has a part in digital and in some cases does things better than traditional interfaces can. Designing those conversations successfully comes down to a designer’s understanding of the technology at their disposal, and then explaining the boundaries of that technology to their users through well-crafted interfaces. My colleague Emily, who helped design Cortana at Microsoft put it like this:
“Don’t lie to your users. Instead find the truth (of the technology) and show it to them.”
Different forms of digital conversations
We interact with digital devices in a number of ways. Most commonly through a visual interface made up of text, images and buttons. The majority of design still happens in this space.
But recently we’ve seen the rise of new forms of interaction, particularly voice. Alexa, Google Home and Siri have been at the forefront of a conversational revolution, with users interacting with products through voice.
The other form of digital conversations is through a conversational interface. Think WhatsApp, Facebook messenger and chatbots. In this form, users can chat with humans and bots and it’s this type of interaction I'll be focusing on.
The difficulties of language
The rise of conversation and voice has been built on the improvements in natural language processing (NLP), a way for computers to understand what users are saying and more importantly, what users want to do or the question they are asking.
NLP uses artificial intelligence and machine learning to extract meaning from human words. While NLP improvements have been impressive in the last 5 years or so, deciphering what a person is saying is still a challenge.
Humans don’t just use words to communicate. We rely on body language, facial expressions, tone, posture and the context of a situation to build understanding.
It is common for someone to receive a text or email and completely misunderstand what the sender was trying to say. However, if I walked up to a workmate’s desk at the end of the day and said one word, ‘pub?’, they would know exactly what I meant. As well as that, they would understand a number of other small, subconscious details that I might be communicating. I could say pub in a way that explained I had had a long day and needed a drink. I could say it in an off-hand way as a suggestion. That one word can mean a thousand different things depending on context and tone.
If I could recommend more reading around conversational principles, this talk on the cooperative principle by Google’s James Giangola is brilliant.
All of this context is currently lost in translation between a person and computer. The only context NLP has is from a user's choice of words. And as I’ve already established, humans aren’t always very good at communicating with words only.
What does your technology understand
How can you get close to replicating conversation in this world of technological constraints? As a designer your role is almost to act as the interpreter between the computer and your user.
You need to understand the technology. I’m not saying you have to go deep into the inner workings of your AI and machine learning, unless you really want to of course, but understand how it behaves. Understand how the responses will be delivered to the user, how much of a user's response the technology will understand, and how much it won’t understand.
You also need to understand how your users talk with a computer. You can do this with research and testing. Get a good understanding of user behaviour, how they form questions and queries. Do your users understand what type of questions they can ask? Do users know they are talking with a bot? How does this alter their behaviour?
Design for understanding
With knowledge of both your technology and your users, you can start to see where the gaps in understanding will be. In any conversation, one person can misunderstand another. It is the same between a user and your technology.
The greatest single thing a designer can do when designing conversational interfaces is to focus on the weaknesses of their technology and design interfaces that bridge the gap between technological weakness and their users. Even though it sounds counter-intuitive, focusing on the weaknesses in the technology is the most human-centred design thing you can do. It is the weaknesses of the technology that should form your interface.
In my current role, I lead the conversational design team at a health company. Designing our conversational Assistant, we knew from research that users didn’t really know what the Assistant would be able to answer. So, we added a suggestion at the start of conversations that if tapped showed the user everything the Assistant knew, in a curated list. It’s not a very conversational thing to add to an interface, but while our NLP was learning, we knew that it would fail a lot. The curated list gave the user the knowledge they needed to get around the weaknesses in our technology.
This might seem a little negative, but good design is defensive. Most companies are not a Google or Amazon in this space. Google can give users, through both its voice and conversational interfaces, a lot of freedom, because it probably has one of the best versions of NLP on the planet. Smaller companies will not have an NLP or knowledge base like Google’s which means designers need to help users more.
Digital conversation is essential these days. To succeed, designers need to fully grasp the weaknesses of the technology at their disposal so they can give users the experience they need. Defensively designing success for their users, rather than failure.
Read part 2: Giving users the right amount of conversational freedom
Back to journal