WAITLESS ORDERING: U-TEN, CONVERSATIONAL UTENSIL HOLDER
Team: Aadvita Aadvanthaya, Rufei Fan, Jing Hu
(March 2019 - April 2019)
ux research & design, VOICE/CONVERSATIONAL INTERFACE design
My team created a Conversational Interface that addresses the problems of feeling rushed while ordering, filtering menus by food allergies, and learning about unfamiliar food items.
As Lead UX Designer, I was in charge of visuals, including storyboards, dialog flowcharts, and the final iteration of the U-TEN concept. I also created interfaces to demonstrate conversation and wrote the final VUI/CUI script, as well as the final pitch.
Lead Visual/VUI Designer
Rapid prototyping and Ideation
Illustrator & Photoshop (for final design)
PRObLEM: ORDERING DELAYS & FINDING STAFF
Many restaurant goers experience embarrassment and frustration when trying to learn about unfamiliar food items. It is also often difficult for customers to know what items to avoid in a menu due to allergies. Additionally, it is getting increasingly harder to find qualified servers who want to stay in the restaurant business.
Solution: CONVERSATIONAL ORDERING UTENSIL HOLDER
U-TEN is a Conversational UI that takes the cognitive load off of both restaurant-goers and servers. U-TEN starts a session with a customer and keeps their voice in a profile to then assist them with menu information and filtering, and places the order for them. Due to the small auditory radius between U-TEN and the customer, U-TEN will not accidentally pick up conversations from other customers.
U-TEN is not meant to replace serving staff, as it will still defer to servers in the event of an issue, and it cannot physically deliver food. U-TEN’s conversational capabilities will enable servers to avoid having to go back and forth to tables to confirm if a customer is ready to order, and enable bypassing language barriers as well.
Reflection: FUTURE ITERATIONS
Designing for a Conversational Interface came with plenty of its own challenges. I’m generally used to working visually in the design arena, and guidance by ear via guidance by eye was a unique new perspective to have. VUI/CUI is becoming increasingly prevalent in our daily lives, and the feasibility of a product like U-TEN is not far away.
If i were to continue iterating on U-TEN, I would perform usability tests with U-TEN with both servers and restaurant-goers. The next focus would be on multiple people at a table, and ordering in different languages. I would also experiment with having U-TEN communicate to the kitchen on the back-end, via a combination of both CUI/CUI and data-driven display.
CHALLENGE: MEAL ORDERING VOICE INTERFACES
What would make a CUI useful in the context of meal ordering? What do users need that they can’t already get from a menu?
RESEARCH & PROTOTYPING: MEAL ORDERING
We started off with Experience Collection exercises. In filling out our knowledge and experience in this domain , we were able to see how differently each of us thought about the area. As a result, we were able to make sure everyone was on the same page along with introducing different perspectives.
In order to begin ideating on a solution, our team delved into research in the area of Voice Interface usage and food.
We found several insights about voice agents:
The market for voice agent shopping, including meal ordering, is still fairly small
People tend to be more expressive/less shy around voice agents
Alternatively, there are still issues with lack of trust (and overtrust) in voice agents
People tend to be more patient with voice agents, e.g. repeating commands
With high turnover rates in the restaurant industry, it is hard to repeatedly train employees. It is a hassle for restaurant owners as well as an expense.
Using the research above, we created 24 different scenarios concerning various uses of a CUI in meal ordering. Scenarios allow us to take several risks, communicate ideas, and explore different domains for a relatively low price. The scenarios forced us to think of potential errors and how the CUI could recover. The scenarios spanned over ordering from home, at the restaurant and on the way to the restaurant.
Each scenario covered pre-attentive states, utterances, attentive states, intents, contexts and responses.
Once we completed the scenarios, each member voted on a scenario we felt would be interesting for experience prototyping.
SCRIPT v1: EMOTIONAL MEAL ORDERING
After creating scenarios, we shortlisted which ones stood out to us. We finally narrowed in on a scenario involving a home ordering scenario with a distressed user. The CUI is able to respond to the user with empathy and tries to help the user’s emotional state.
We ran into several challenges while creating a script—would the CUI initiate the session if the user was feeling too much distress? We decided to experiment beyond the current capabilities of products on the market and make the CUI capable of recognizing a distressed voice. This was then modified in the CUI recognizing odd behavior patterns, such as the user not moving throughout the course of a day.
EXPERIENCE PROTOTYPING & FEEDBAcK: CUI TOO INTRUSIVE and “CREEPY”
By experience prototyping and having the scenario acted out by another team, we were able to see flaws and get feedback about the scenario. Our colleagues felt the CUI was creepy and too intrusive when they saw the scenario acted out. This got us thinking about privacy and how we can make the CUI seem empathetic without it being intrusive. Therefore, we decided to pivot to a more public domain.
After our first round of rapid prototyping, we decided to go deeper into restaurant ordering. We spoke with several users about meal ordering and learned that:
Unfamiliarity with food items can be a source of embarrassment
Patrons are interested in learning more about different, unfamiliar kinds of food
Patrons are hesitant to ask the serving staff for detailed information on food
Patrons with allergies have to take extra care to ensure their food isn’t contaminated via utensils and cooking services
All of the above can take extra time in serving, and ultimately receiving food
PIVOT: RESTAURANT MEAL ORDERING
To support our earlier research, while our team members were together, Advita and our classmate, Peng, were having a discussion about ordering from an Indian restaurant. Peng wasn’t very confident about ordering as he wasn’t familiar with the cuisine. Advita explained items on the menu as Peng pointed them out and showed him images so that he could better understand them.
As soon as we saw Advita and Peng’s scenario play out (it was almost like another version of experience prototyping with Advita as the CUI), we decided it would be great to pivot. The scenario seemed to be more compelling and the CUI had better value for customers and the restaurant industry, solving issues we saw in our research.
This new scenario seemed to be more compelling and the CUI had better value for customers and the restaurant industry, solving issues we saw in our background research. We proceeded to quickly sketch the scenario out for our movie.
We tried to think of which form our CUI could take, thinking of objects commonly found at any kind of dining table. After discussion, we decided the CUI would take the form of a utensil holder with an indicator on its body.
PERSONAS: PATRON & MANAGER
Using all the data we’d gathered just far, I created Personas for the team.
PROTOTYPE DESIGN: I & II
In order to visualize our Utensil Holder Design for the purposes of testing out concepts, we used a lo-fi paper & plastic prototype. I then iterated on this prototype to create a digital version in Illustrator & Photoshop. We decided on the name U-TEN, as it was more distinct and easier to pronounce than our previous names, and did not invoke memories of indigestion.
PIVOT 2: WHAT MAKES A CUI NECESSARY IN A RESTAURANT SCENARIO?
We had to think more about how to make the CUI more compelling for restaurants to install as well as to the user. How could we provide more value to the restaurant and the user?
One way was if the CUI body could project a menu onto the table. A digital menu would allow restaurants to easily change the menu whenever required and would also present a more interactive experience to the customer.
But that still didn’t answer the question: how can we make the CUI do something that a phone isn’t smart enough to do?
We looked back at our research: many restaurants need to exercise caution when customers with specific needs come into the establishment. It’s important for employees to not forget these needs as it could lead to potentially life-threatening situations. A CUI can help provide a reminder of that data to the staff and kitchen.
Via our research, we synthesized that adding customizability via the freedom of a conversation was easier than having customers visually crawl through a menu, or have constant back-and-forths with the server on which items are, for example, peanut-free. As menus are constantly changing, and there is an evident high turnover rate of restaurant staff, a CUI can be a source of easily updated, reliable information.
FINAL CONVERSATIONAL SCRIPT
Our final script involves a scenario where the user discusses food preferences and a peanut allergy. I created a dialog flow chart to illustrate a sample scenario.
(Currently working on Interactive Dialogflow Prototype—coming soon)
WHAT U-TEN DOES WELL:
U-TEN’s dialogue is friendly and trustworthy
Can provide value to several different stakeholders
“I really liked that customers can see how to eat [a certain menu item]. I’ve been to restaurants where I had no idea how to eat the food.”
Accurate portrayal of the problem, especially w/ other customers
FURTHER CHALLENGES TO ADDRESS WITH U-TEN:
What’s better? Initiating conversation or having U-TEN initiate?
Letting users know U-TEN’s true capabilities without over or under-estimating
AI would have to be very careful with allergies & this information would need to be checked by servers, still—it is unrealistic to just rely on a CUI.