Why I Spent A Year Designing For Voice
Since the dawn of computing, we have dreamed of (and had nightmares about) machines that can think and speak like us. But the computers we’ve interacted with over the past few decades are a far cry from HAL 9000 or Samantha from Her. Nevertheless, machine learning is currently in a renaissance that has started to provide designers with a wide assortment of new tools for better engaging and understanding users. Hence I decided to dedicate time exploring how designers can better engage with users via voice-based interactions. 

Process - How Did We Approached The Project 
Unable to find clearly laid out the path for this nascent space, we looked inward at how humans communicate. We started with the basics of conversations, learned how to establish dialog and trust and then decided to translate acquired knowledge to create a better experience between man and machine using wizard of oz technique for our initial prototype & API.ai for testing with users. Once we created our prototypes using api.ai we deployed on the Google Home & Amazon Echo platform. 

VUI Cornerstones - UX Ingredients To Designing Great Voice Based Products
Based on our research on human-human interaction, we identified the fundamentals of conversations. In order to build great conversation products, we need to solve at least six different user experience challenges ranging from learning how to create dialog, assisting in feature discovery, getting started or learning how to build acquaintance, breaking interactions into granular exchanges, mitigating faulty assumptions and learning to seek clarity.

Cornerstone One - Assisting in Feature Discovery  
Despite its many advantages, an open-ended dialogue does little to make the system’s capabilities known to the user. To make the most of the interface, the user must hold prior knowledge of the available features or make frequent reference to the documentation, which is likely to negate any efficiency gains over menu-based workflows. For intelligent interfaces, offloading the process of feature discovery to auxiliary documentation would be even less practical. Imagine the experience when we are opening the Echo App, there’s a booklet that provides us with all the actions that are possible. The potential limitations of a conversational user interface can be seen in our interactions with first-generation intelligent assistants such as Siri, Cortana, and Echo.

Solution One - Providing On boarding Examples 
One solution to the problem of feature discovery that is offered by many intelligent assistants is to provide the user with a set of example inputs that demonstrate a variety of features available within the system. This helps new users to become acquainted with the system’s mode of interaction and the kinds of knowledge it is able to access. 
Solution Two - Providing Contextual Assistance
Demonstrating an application’s full range of features through a series of embedded examples can, however, begin to make the software look suspiciously like a reference manual. Rather than handpicking a static group of examples, machine learning can be used to dynamically select examples that are contextually relevant to the user’s activity. Going further, machine intelligence could be used to generate speculative extensions of the user’s trajectory, providing a series of possible next steps in a range of conceptual directions for the user to consider. This mechanism would not only extend the user’s understanding of the interface and its possibilities but also enrich his or her fluency with the task or domain of interest at hand. 
Solution Three - Offering Suggested Meaning
Another common mechanism for helping users to locate features within a conversational interface is to offer suggested meanings when their statements are not clear to the system. This can be helpful in correcting the improperly formed expression of a command that is already more-or-less known to the user. But if the user is completely unaware of some potentially advantageous feature, he will not think to ask for it in the first place, and this mechanism can provide no assistance.

Our prototype built using Dialog Flow which contains intents and training phrases

Final Thoughts - Lessons Learnt
In my perspective, as designers, we should strive to build systems that not only respond to user requests but also enrich the user’s understanding of the domain itself. It is not enough for intelligent interfaces to offer an infinite variety of features; they must also find ways of connecting the user with the possibilities they contain.
Back to Top