As part of the work I;m doing with Lola Tech I recently penned a few words on lessons so far in creating voice and chatbots:
We’ve been working with voice interfaces (VUIs) and chatbots for a while, and have 5 key lessons to share that can help you make the most of this new breed of interface.
With the likes of Alexa, Siri and Google Home capturing the hearts and minds of tech enthusiasts around the world, it’s clear that a new kind of interface is upon us. With analysts estimating Alexa could be a $10B megahit by 2020, the sky seems to be the limit for conversational interfaces.
That’s not to say that clicks, swipes and taps are in the past just yet. For all the good voice interfaces and chatbots can bring to customer engagement, anyone who has spent enough time with the likes of Alexa knows that there are limits to what the bots and voice design can do today.
The voice interface is a young technology, with developers and consumers alike working out better ways to handle and operate this new design paradigm. Here at Lola Tech, we’ve been tinkering with voice user interfaces (VUI) for some time, and have had some success releasing our own products – notably Dazzle: our voice-activated digital concierge. We’ve learned 5 key lessons along the way that can help other developers and tech innovators make the most of this new kind of interface.
Lesson 1: Build a human into the loop
The ideal bot may in future create a completely closed loop of interaction that doesn’t require human intervention. The user asks a question, and the bot gives a satisfying answer – empowering your staff to spend time on more high value tasks. That closed-loop future isn’t quite here yet though.
Even the most advanced bot in today’s world will be flummoxed by some of the bizarre requests users will inevitably make. To account for those questions a bot can’t answer, you need to build loops that can also include a human assistant if needed. To that end, it’s important to get out there and start learning from your user base ASAP.
Using Dazzle as an example, it’s well equipped to tell a guest about restaurants and attractions near the hotel they’re staying in. If someone asks for a bespoke tour of the city, that might be a bit beyond Dazzle’s scope. Rather than give a dissatisfying “I don’t understand” message, Dazzle can pass the message on to a human concierge who’ll give the guest a call to try and resolve their issue. That kind of failover provides a much better experience for users.
Lesson 2: Voice and text interfaces can win users over faster
Voice interfaces are novel, and exciting; the cool new thing people want to try out and explore. But if you only have a voice interface, the experience can be limited – leaving people with the disappointment that the VUIs of today aren’t quite as advanced as the expectations set by decades of sci-fi.
Today’s VUI is limited in comparison to text chat interfaces, because voice is a far more complex thing to process. It doesn’t help that people tend to be less concise when asking for something verbally, adding in “umms”, “aaahs” and other verbal spacers that can confuse today’s voice assistants.
Using voice and text together is key. Voice acts as the draw – the exciting innovation people want to see for themselves. Then, when you have that initial engagement, you have a potentially more stable alternative through a text chatbot. Users tend to trust a system far more if they receive feedback and confirmation through another channel. Think of it like how a retailer sends you a dispatch to your email for an online service, or a bank sending you an SMS to log into your account. In these examples, you trust the system is working when you see evidence across multiple channels. You can achieve that same level of trust by confirming a voice command through another device.
Lesson 3: It’s all in the copy
When you build a webpage or an app, you need to have an innate understanding of what will appeal to the folks using it. What colours will they like? What font will make them feel at ease? How can you add nuance to the user journey? There are plenty of small concerns like this that have a big impact on how users respond to your product.
The same things apply to VUIs. Without the visual language many experiences rely on, all the emotions you want to convey to users must be shared through audio cues and responses. Voice and tone are a big part of this, but the writing is an even bigger one. When all a user has to go on is words, they’d better be some damn good words.
Beyond responses, writers are also important for considering the questions users will ask. We have hired screenwriters and poets to the team to help us not only craft the conversational characters, but also to build the many different ways users make requests and deliver delightful responses. Of course, we also capture questions a given system doesn’t know and build it in for use later. The initial load of questions the bot needs to anticipate is best captured by a writer, we’re also using a lot more machine reading these days.
Lesson 4: The online and offline worlds are very much linked
If you think of the big online influencers – the Twitter heavyweights and Instagram darlings of the world – they normally also have a very interesting life offline. The on and offline worlds are very much linked, and to ignore that link in your chatbot or VUI is a big mistake.
When designing your chatbot, always keep focussed on how it will enrich a user’s life in the real world. The end-goal is never the platform itself, but how the conversational interface delivers great experiences. That’s a very important focus to keep in mind as you develop, test and workshop new parts of the conversational interface.
Lesson 5: Be prepared to go back to the drawing board
Developing a conversational interface demands a new set of skills, and a completely different design language to what has come before. Many of the lessons covered above are about developing the skills needed to function in a world where clicking and tapping aren’t the primary interactions. But there are plenty more questions you will need to ask, and be prepared to answer.
Can theatre techniques help us write better chat interfaces? What’s the optimum length for a question or response? How do you keep people engaged in a longer booking process? These are all questions you can (and will need to) answer through workshopping and user testing sessions.
There are of course many more questions that will need to be asked to further develop the richness and complexity of conversation interfaces. Here at Lola, we’ve been doing this kind of experimentation for several years, and are looking to do even more in the future. If you have any questions about the conversational interface, or need help building your own, please get in touch at email@example.com!