Designing a novel voice-based conversational user interface to support and facilitate healthy eating.
Conversational User Interface (CUI) is a big buzzword and a cool trend in design these days, but how can we use it to actually improve the user experience?
Changing your diet to improve your health is never easy. You have to give up your favorite snacks, the greasy food you crave after a bad day, and you have to prepare more food yourself. Finding food to eat for your body and conditions can be difficult, expensive, and time-consuming.
The aim of this project was to create a novel voice-based CUI to support and facilitate healthy eating, especially for the purpose of managing diabetes. We wanted to improve the accessibility of resources for managing your health, and to make the process easier in general. Being healthy shouldn’t be hard.
In short, Ina is a food ordering personal assistant with conversational user interface to help people with diabetes find the right food for them. Ina focuses on helping you plan your meals: from ordering groceries to step by step assistance when cooking recipes. Ina is also programmed to recover from errors in communication. Finally, and most importantly, Ina will make suggestions based on what you like to eat and what you can afford, based on programmed settings.
Of course, we couldn’t start without doing extensive research on the domain and our users.
We conducted research across three domains. The first was medical research — we read papers about the best way to encourage healthy eating. We also interviewed a medical professional to understand some of the barriers they see patients face when trying to eat healthy. The second area we looked at was user research, we spent time on diabetes forums to understand user pain points for eating healthy as well as what works well for them. Third, we looked at the food industry, we conducted a competitive analysis of other food ordering services out there to get a sense of trends and gaps in the market. We also talked with a restaurant designer on ways that he promotes different food options.
It is important to remember that, despite being a big trend in design, CUIs still have disadvantages that may contribute to less usability. However, we decided that a conversational user interface could still improve the overall experience of the system due to four primary key advantages it offered:
We found that for many, it can be hard to reach out and ask for help from friends, family, and even medical professionals. And often times, it’s just plain inconvenient to access or visit medical professionals regularly.
Scenario: “I am overweight and desperately need to shift a few pounds. Since being diagnosed I have only seen my diabetic nurse twice and my consultant once which was almost a year ago now since I’ve seen either. It’s a nightmare trying to get an appointment at my clinic.”
Unlike a real-life interaction, the help of a CUI can be requested at any time for the smallest of questions. They can provide advanced searching by understanding the user’s intent and preferences faster than scanning webpages for relevant information or waiting on phones. And, they can learn to predict things users haven’t even thought of.
Deciding what to eat is a very repetitive task but must take into account many different factors (carbohydrates? blood glucose? starchy vegetables?) that may take many Google searches, or years of experience. A CUI could easily keep track of and account for these needs and preferences when giving personalized advice.
Coaching is a popular and strong use case for CUIs. A conversational interface can make the experience more engaging, personal, and human. And, it comes without the inconvenience and fear of judgement from having a real person hovering over you all day.
Scenario: “When you’ve taken your insulin, you need to eat soon afterwards. I forgot once, when some guests arrived unexpectedly. I put some chairs out in the garden, forgetting I’d just taken my insulin. Next thing, I collapsed on to the kitchen floor, so my wife called an ambulance. I was taken to hospital and stayed there until night time. My blood sugar levels had fallen too low.”
Imagine if the CUI could remind you when to take insulin, how much insulin to take, when to eat, what to eat depending on your blood sugar levels and past meals, etc.
A great strength of CUIs is also that it can be used hands-free and vision-free. It can be used when you’re busy moving chairs or cooking, when you can’t see phone screens clearly, and so on. Considering diabetes can severely affect vision by damaging the blood vessels in the eye (diabetic retinopathy), it is critical that the system can be used without sight.
And, since conversation comes naturally to people, there is also less of a learning curve than learning the buttons, interactions, and ins-and-outs of a GUI app.
Before proceeding to writing scripts or conversation models, we wanted to clearly define Ina’s personality. She should be friendly and supportive, but be more of a coach than a best friend (not too snarky).
We also decided to opt for a more robotic voice rather than more human-sounding voices (we found a Gordon Ramsey voice generator but decided it wouldn’t be appropriate). We wanted to make sure users always knew that they were talking to a machine. There is often a fear of judgement or stigma when talking about food or health (even with a live nutritionist), so we wanted to allow users to discuss their personal issues more openly without worrying about such fears.
Yes, we named it after Ina Garten.
We worked primarily with the parts of the experience where the presence of a doctor or dietitian wasn’t necessary and that usually relies on the discretion of the patient, like meal prepping. So, we focused on the following goals to be accomplished through a conversation with Ina.
At this point, we began designing user flows through the dialogue. Throughout this process, we found user testing to be a key part of designing our voice CUI.
We separated the conversation into the utterance (what the user says), intent (what the conversation should ultimately achieve), response (the CUI’s response to the user), and error recovery (in the case that the CUI cannot understand the user). In the simplified flow below, the CUI shows the user’s food options for the week.
We went through many iterations of our conversation models. Each time, we acted the dialogue out with other people and tried out different voice simulators (this is when we tried using a Gordon Ramsey voice generator and received negative feedback) to get a better idea of how the design feels and performs.
Our user tests showed us many problems and opportunity for miscommunication in our design at the time. Here are two major points we received from users as feedback:
1. How would you say no to someone if they wanted to have food that they aren’t “allowed” to have?
We had to determine how to convince people to eat healthier options when they were asking for food they crave. We consulted some psychology papers related to persuasion and ended up reassessing our use of language.
Darya Rose, author of Foodist, gave us some great tips on convincing people to eat healthy:
2. How will Ina respond to requests it is not programmed to handle?
One specific error we decided to focus on and improve was the idea of the user asking for a dish/recipe that Ina could not provide. This was a great way to demonstrate alternate smart suggestions when Ina can’t provide exactly what is requested. By capturing key words that Ina knows, it can recommend alternatives from its database.
This project was interesting because it challenged me to focus more heavily on exploring very different solutions and techniques to understand and solve problems. Working on new technologies was very exciting for me; it presented very different problems than what I had been used to thus far and forced me to utilize a wider skillset.
While we focused on providing a voice-only experience, in the future, I would consider a hybrid experience that combines advantages of both graphical and conversational user interfaces, which could perhaps provide a greater overall user experience.
And finally, here's our concept video for our conversational user interface design, INA.