6 min read

The what, why and how of great conversation & conversational AI design

The what, why and how of great conversation & conversational AI design

It wasn’t too long ago that the only way customers could have a ‘conversation’ with a company was to pick up the phone. That often meant being greeted by the company’s automated phone system, the IVR (interactive voice response).

While the phone remains a key contact channel for most brands, the range of customer contact channels offered to customers is ever-increasing. The rise in the use of messaging anc conversational AI apps over recent years has seen an increase in customers having some of those conversations via Facebook Messenger, WeChat and WhatsApp, as well as through online chat sessions. There’s no doubt that the way customers want and expect to interact with brands is changing and will continue to do so.

One of the most exciting developments of the last couple of years is in the rise of conversational AI interfaces like virtual assistants and chatbots.  These kinds of interfaces are becoming more mainstream and companies are starting to invest in these technologies to connect with their customers. With the likes of Google Assistant, Siri and Alexa taking centre stage, conversation design is a hot topic.

Watch our webinar: Simple ways to assess conversation design

What is conversation design?

Conversation Design is the process of designing a natural, two-way interaction between a user and a system (via voice or text) based on the principles of human to human conversation. Conversation is the exchange of information by language. For a conversation to take place, two speakers have a shared mental model and communicate their goals, questions, intentions and emotion. These principles form the basis of good conversation and conversational AI design. Examples of conversational interfaces include voice interfaces like IVRs and voice assistants and text based conversational interfaces such as interactive SMS and chatbots.


But conversation design isn’t a new thing…

For those of us who have been working in the world of conversational interfaces for some time, it’s exciting to see our art thrive but also surprising that conversation design is being talked about as if it’s something new. Admittedly, the rise of these new conversational AI technologies means that conversation design is taking a new path and allowing us to rethink approaches to designing and using conversational interfaces. But when I think back to the start of my journey in the industry, some things really aren’t that new.

Back in the year 2000, one of my first experiences of working with a conversational interface was with “Wildfire”, a speech-enabled voice assistant. Wildfire did things like manage your voicemail and call people in your contacts list. At that time, it seemed the thing that science fiction had been predicting for years.

Remember HAL from 2001: A Space Odyssey? Maybe Wildfire wasn’t quite as sophisticated, but at the time, it seemed a big step in the right direction. This wasn’t just an interface that used speech recognition, but ’she’ was an interface with an intended character. That character (or ‘persona’) was a big part of the interactive experience and something that not only became an important part of the experience but also for research and insight into how people wanted to use that kind of interface. Conversational AI was growing.

It was during that project that my fascination for conversation design really started. Almost 20 years on and the conversational interface landscape is a very different place but the potential for conversation design is just as exciting.

Over the last 10 years, most people’s exposure to a conversational interface has been through interacting with an IVR. In our past post, ‘Designing chatbots and what we can learn from IVR’, we discuss the fundamentals of conversation and conversational AI design for IVR and how this can be applied to chatbots. We know that users hate badly designed IVRs, and we’ve made it our mission to rid the world of bad IVRs! But people hate any badly designed interface, and this is true for conversational interfaces too.

What makes conversation design different?

Conversation design, whether for IVR, chatbots or virtual assistants, relies not just on an understanding of conversational norms, but on a thorough understanding of context of use, users, interface design, affective (or emotional) engagement and great dialog design. And these things don’t happen by accident.

At VoxGen, we have spent the last 15 years creating, refining and optimising our processes for designing and evaluating effective, natural and efficient conversational dialogs.

If we’re expecting our users to engage with spoken language interfaces, it’s essential that we design with models of natural conversation in mind. In a previous blog post ‘Design IVR Conversations not Call Flows’, we discuss 3 key areas that are important to conversation and conversational AI design:

  • Persona: Your persona is what callers hear when they pick up the phone. The persona is more than just a voice; he or she is a character who engages your callers and embodies the brand values of your organisation. To get the best results, the persona should be played by a trained voice actor
  • Prosody:Does the persona sound genuine and empathetic? Do they talk in a natural way? Do they pronounce things – phone numbers, for example – in the way the customer expects? Is it all clear and easy to understand? In short, does the interaction feel like a conversation with a human being?
  • Collaboration:Real-life conversations are co-operative, with both parties collaborating to arrive at a clear understanding. If the system doesn’t understand what a caller said, does it help move things forward in an easy, natural way – or does the IVR just repeat the same thing before dismissing them

Our learnings from the last decade of designing smart, connected and conversational IVRs has highlighted the importance of user-centred design (UCD). UCD is critical to designing great conversation designs.

In many ways, the UCD process we follow at VoxGen for creating conversation design is like any good UCD process – a multidisciplinary team, early research to understand users and context of use, iterative design and evaluation involving representative users, and ongoing optimisation:

VoxGen user-centred design process

Figure 1: VoxGen user-centred design process

Some of the differences when designing conversational interfaces relate to the methodologies, tools and skillsets utilised:

  • Understanding contextual dialog norms – different conversational and conversational AI interfaces will need to support different users and different tasks. This means that a ‘one-size fits all’ approach to conversation design doesn’t work. An IVR for healthcare professionals will be used in a different context to users of a chatbot who want to find out the status of their order. Our own research has shown that these differing contexts and user needs impacts things like the formality of language expected, task priorities and even the specific words and phrases we use. We need to understand those specifics before we start the design through user research (e.g. user interviews, call listening and ethnography). This allows us to understand how context and the specific needs of the users we’re designing for will impact the overall conversation design.
  • Documenting the design – the way we speak and the way we write are very different – something conversation / conversational AI needs to incorporate. But the way we read and the way we listen can also be very different. In voice user interface (VUI) design, we produce sample dialogs that are presented to clients in audio format and in ‘persona’. It’s much easier to depict the conversational flow as spoken dialog if people can hear it. Reaction to something somebody reads can be quite different to that same dialog in audio format.
  • Usability research – when conducting usability evaluations of conversation design, we need to evaluate those designs in the same context in which they will be used. For example, if we’re evaluating an IVR, we use our in-house html prototype tool to play the IVR experience over the phone to participants (using the Wizard of Oz usability methodology). Voice interfaces rely on users’ auditory processing as well as cognitive processes such as short-term memory and attention. The only way we can realistically evaluate cognitive load, for example, is to present the interface over the phone in the same way users would use the interface in real life. It’s soon clear whether the amount of information we’re presenting to the caller is working or not. This also means that the ‘think aloud’ methodology commonly used in web and digital usability research (where participants verbalise their thoughts, feelings and experience as they interact with a product) doesn’t work here. We need to understand how the interaction flows, so asking participants to verbalise their experience as they conduct a task would be too much of an interference for usability evaluations of conversation design, especially for voice interfaces.
  • The design team – conversation and conversational AI design is a specialised area of design. While digital UX traditionally focuses on visual design, conversation design relies much more on an understanding of linguistics, psychology, voice user interface design and speech science. Ensuring your conversation design team has a solid set of skills in those areas will help in the creation of great conversation design.

Want to find out more about conversation design and how you can design great conversational experiences? Our Design Cloud is packed with resources to help you design user-centred conversational interfaces for any platform, in multiple languages.



Related Posts