Cucui: a Voice User Interface

The context

In recent years, voice user interfaces (VUIs) have been rising in popularity and functionality. Products like the Amazon Echo and the Google Home have given us the ability to have a conversation with a computer, while leaving our hands free to complete other tasks. As designers, our mission is to construct these conversations themselves, to feel natural, flowing, and technologically possible.

The Problem

How can we make ordering food an easier experience, especially with a group of people?

The solution

Cucui (pronounced: kuh-koo-ee) is a voice-based personal assistant designed specifically to facilitate and support food-ordering. Cucui can differentiate between multiple users and orders, understand personal preferences, split the bill, and organize delivery, freeing your hands and time to work on other things.

Duration

3 weeks, 3 people team

mY role

Competitive analysis, Scenarios, Conversation modeling, Prototyping

understanding the context

We started by asking questions to understand the role of a VUI in the food-ordering space.

Who will use, support, and influence this?

Our users can be anyone old enough to order food. Restaurants will be influential, as only those who have delivery options and comply will be able to participate. Restaurants will supply food with the facilitation of the VUI.

What are the additional products associated with this?

Users will need a credit card or digital payment method to use the VUI.

What processes and actions are associated with this?

Associated actions are placing the order, preparation, delivering, and paying. Users should be able to change their order and ask questions as well.

What is the desired outcome?

Often when ordering food, we have to stop what we are doing, grab a menu or computer, and find our credit cards. This process is not quick, and it’s even longer when we order with more than one other person. The desired outcome of using the VUI is an easy and effortless experience.

Finding a gap

To better understand our competitors, we analyzed the features available with existing VUIs/CUIs, such as Domino’s, Amazon Echo, Google Assistant, Tmall Genie, and McDonald’s via Facebook Messenger. We highlighted the interesting gaps in blue, such as group-ordering, handling special food preferences, smart recommendations, handling complaints, and handling budgets. Our goal was to fill in these gaps.

usage scenarios

We then created 10 detailed usage scenarios involving our VUI, highlighting the different features. Creating the scenarios helped guide us when deciding on the functionality and usage of the VUI in terms of our users’ needs.

A group of friends is holding a study group for a big exam tomorrow. It is getting late and they would like to order dinner together. However, they don’t have the time to go out and get food, and also can’t decide on a restaurant. They ask the VUI to smart recommend a restaurant based on their various preferences. The VUI then facilitates the ordering of food for the entire group and easily allows them to split the bill so they don’t have to waste time calculating who owes who money.

experience prototyping

We picked a couple of our scenarios to have our colleagues act out for a group. This was great in helping us bring our idea to life and getting a fresh perspective on the scenario. We received a ton of good feedback from the actors as well as the audience on things to consider when designing the VUI and modeling the conversation.

Feedback:

• What is the VUI’s persona? Is it a waiter? We were told to clearly define his persona and think about his voice (literally and figuratively).

• How does the system work? At what point does does the user input their credit card information?

• How will the VUI handle people changing their minds on their order?

• How will the VUI handle not understanding/hearing something? Something important to consider especially amongst large groups.

Introducing, Cucui…

Using our feedback, we wanted to clearly define the personality of Cucui (our VUI). We wanted him to take the role of a friendly waiter, able to handle user personalities and questions. After much deliberation and testing, we landed on a male machine voice for Cucui. It sounded the most natural and easy to understand, and we didn’t want Cucui’s voice to get lost in large groups.

Why Cucui? It just rolls off the tongue!

Conversation modeling

We first defined various intents in a conversation (what should be achieved at different points in an interaction with Cucui).

• Start a group food order

• Establish group profile

• Swap/change order

• Search user order history

• Order food

• Special food request

• Make a recommendation

We then mapped out the different intents (what should be achieved), utterances (what a user would say), response (what the machine would say), and the error recovery (how the machine would respond to not understanding the user). We spent a lot of time modeling how Cucui would perform and went through many iterations. It was very helpful to have people act out the conversations and gave us a better understanding on what was natural.

Error recovery

We wanted to address two types of errors Cucui will encounter.

Wrong Intent: The user wants something with a special request, but the VUI thinks the user wants the standard version. The VUI’s standard order intent is triggered. After confirmation, it understands and recovers from the error by triggered the right intent. (ex. “I would like extra soy sauce on the side”)

Mishearing: The user specifies a parameter about an intent, but the VUI mishears and specifies the wrong parameter. After confirmation, it understands and recovers by changing the parameter of the same intent. (ex. mishearing soy sauce as hot sauce)

Cucui can also handle multiple people talking at the same time, by prompting a specific user to speak first.

more prototyping

We then storyboarded our scenario and conducted more experience prototyping! We received critiques that Cucui lacked attentive feedback and too many confirmations from Cucui may make the user more aware that they are talking to a machine. To address this, we iterated once more on Cucui’s conversation model and added a foley to indicate his attentive state. If Cucui does make a mistake, the user can interrupt him at any point without waiting for confirmation (user interjection). The foley can be seen in our final concept video below!

Takeaways

Being a new domain for our group, this VUI project was both a challenge and an exciting experience. It was great to have to think in new ways and rely much more on how the experience felt (rather than something visual). I think for the future, it would be interesting to create a GUI to pair with Cucui!