A virtual therapy chatbot for
Coco is a virtual therapy chatbot for caregivers of children with asthma. It uses established therapy techniques to help caregivers solve challenges and address their own needs as well.
I worked as the only designer (user experience and conversation design) on the team collaborating with two researchers, two academic experts, and a team of developers.
I lead the design of the chatbot "Coco" which included creating the chatbot persona, sample dialogs, Wizard of Oz testing, interface design, and design handoff for implementation.
Coco page cover
Why create a chatbot for caregivers?
Raising children is a lot of work. It gets even more difficult when your child is suffering from a chronic condition like asthma which means more doctor visits, supervision, sleepless nights, and added costs. Through all this, there is hardly time for caregivers to take time for themselves and attend to their own needs.
  • Provide a self-care program via Coco that is tailored to their specific needs as a caregiver of a child with a chronic illness such as asthma.
  • Provide actionable therapy techniques and resources for dealing with stress in a conversational manner which can easily be built into their routine.
  • Create an accountability system with personalized reminders from Coco so that the caregivers prioritize their own needs along with those of their children.
Understanding our target audience
Coco has been built keeping in mind caregivers of children with chronic conditions. To understand our audience we conducted participatory design sessions with parents meeting our criteria. Our goals with these sessions were:
  • To understand the parent/caregiver, the child, and their relationship by creating parent-child personas that illuminate aspects of both.
  • To understand the stress triggers for the caregivers as they deal with their child's condition on a daily basis. This would help Coco understand where it can intervene and help out.
Darren persona
Persona of a caregiver and their child: Darren & Blake
Designing the chatbot's personality
I spearheaded the design of Coco's personality by understanding the research team's findings on the needs of caregivers and their children. My goals for Coco's personality were:
  • Coco should be relatable to the caregivers as someone they can feel comfortable sharing their personal struggles with.
  • Coco should be seen as knowledgeable and someone who gives sound advice.
  • Coco should not be seen as a substitute for a medical professional.
To be effective as a companion for a caregiver's troubles, the persona was modeled after a health counsellor who guides the caregiver to the right resources in a friendly manner.
Coco persona
Persona for Coco
Key use cases
  1. Build a therapeutic relationship i.e introduce Coco and the team behind Coco and get to know the caregiver and their family.
  2. Identify and assess an issue the caregiver is facing and present solutions using Problem Solving Therapy.
  3. Set goals to implement the solutions and eventually evaluate progress.
We envisioned the above use cases to be cumulatively completed as a result of an ongoing conversation between Coco and the caregiver over a period of time. Below is an example of a sample dialog and user flow based on the above use cases.
Sample dialog
A sample dialog for Coco's first interaction with the Darren persona
Sample flow
High-level flow for the sample dialog
The Wizard of Oz technique
Our next step was to test out the conversation flows with caregivers and iron out any kinks that they may discover in the conversation. We adopted the Wizard of Oz testing method to test our conversation where members of our research team took the part of "Wizard" and interacted with caregivers as Coco. The caregivers were unaware that a human was responding as Coco and not the bot itself. Below is a snapshot of the tool that we created for this purpose in action.
Woz interface
Wizard of Oz interface for dialog testing

Becoming smarter through conversations
We continue to improve Coco's responses using machine learning and natural language processing with further Wizard of Oz testing sessions. Coco is set to launch in the next year, once the implementation is complete and the accuracy of responses is at the desired level.
Our work was published in the 2020 CHI conference as a part of the late-breaking works.