MACHINE LEARNINGSTARTUPCHATBOT

Woebot

An AI-based mental health chatbot to make therapy more accessible.

Date January - June 2017
Platforms
iOS, Android
Team 
Alison Darcy, Founder & CEO;
Pierre Smith, Engineer; Andrew Ng, Advisor
My Role
UX Researcher

PROBLEM SPACE

How might we make mental health treatment more accessible?

Despite 1 billion sufferers of mental health disorders worldwide, treatment options are underdeveloped, and therapy is expensive, inconvenient, and difficult to find.

2/3 of people struggling with mental health problems will never see a clinician.

Stigma, physical location, financial hardship, and complicated healthcare systems decrease the likelihood of seeing a therapist in-person. The majority of individuals struggling with their mental health will never receive therapy.

SOLUTION

A machine learning enabled chatbot that ameliorates mental health conditions.

Woebot is a personalized, artificially intelligent chatbot that combats depression and anxiety through brief daily check-ins.

Backed by AI pioneer Andrew Ng, Woebot was released in June 2017 and grew from 100 users to 50,000 users in its first week. It received $8 million in Series A funding and now receives over 2 million messages per week.

FEATURES

Daily emotion check-ins make users mindful of triggers.

Woebot checks in with users for 5-10 minutes each day. Much like a gratitude journal, these brief encounters ask users to reflect on their day and examine how they felt during key events that unfolded to uncover any emotional triggers.

GIFs and stories provide teaching moments.

Woebot frames users' mental health challenges into digestible stories that create psychological distance and give users an alternative point of view.

GENERATIVE RESEARCH

We parallel prototyped 4 solutions with 100 college students.

Our team knew that we wanted to create a Cognitive Behavioral Therapy (CBT)-based technical solution for our target audience. We generated 4 out-of-the-box ideas to prototype with 100 Stanford students over the course of 2 weeks. These included:

Daily mood monitoring is beneficial for mental health.

Individuals responded the most positively to Daily Mood Monitoring "check ins" on social media platforms that were Wizard of Oz-ed by members of the research team in place of AI technology.

EVALUATIVE RESEARCH

70 participants used an MVP version of Woebot over the course of 2 weeks.

We moved forward with our Daily Mood Monitoring idea and built an MVP chatbot in Facebook Messenger. I conducted a clinical research study to assess Woebot's effectiveness as a therapeutic tool, randomly assigning participants to a Woebot group and an informational control group over the course of 2 weeks.

Users felt a noticeable difference after just 2 weeks of interacting with Woebot.

Results of the study were overwhelmingly positive, as users raved about Woebot.

Limitations in natural conversation were a point of frustration.

Despite positive feedback, users expressed frustration about Woebot's limited, pre-programmed responses. To solve for this, we made a more intricate conversation decision tree moving forward.

Woebot significantly reduced depressive symptoms.

We collected scientific measures on depressive & anxious symptoms throughout the duration of the research study. Statistical analyses showed that users in the Woebot group experienced symptom reduction while those in the control group did not, indicating that Woebot may alleviate depressive symptoms more than traditional sources of information.

PUBLICATION

We published the results of our clinical research in an academic journal.

READ OUR PUBLICATION

DESIGN DECISIONS

1. Emojis to describe emotional states.

Action: Users describe how they're feeling through emojis, rather than words.

Rationale: Users responded more positively to emojis than traditional texting with a conversational agent. Emojis create psychological distance, making a conversation about emotions easier.

2. Breaking down content into digestible chunks.

Action: Reduce users' cognitive load by allowing them to quickly choose an option rather than having to type it out. Integrate images, GIFs, and videos into Woebot's conversations.

Rationale: Woebot's conversations are meant to happen in the moment, making typing out a response difficult for users if they're on the go. Having pre-programmed selections makes conversing more carefree. Images and GIFs make the experience more engaging.

3. Short & sweet check-ins.

Action: Keep daily check-ins between 5-10 minutes.

Rationale: The brevity of Woebot's check-ins make users more likely to engage with the bot over time— the goal is gradual behavior change, rather than lengthy, in-depth conversations.

IMPACT

Woebot has reduced the burden of mental illness on a global scale.

Woebot spoke to more people on its first day of official launch than a therapist could see in a lifetime— and has since reached users in over 130 countries around the world. It currently maintains 2 million conversations per week and has been featured in The New York Times, The Wall Street Journal, BBC, and more.

Woebot received $8 million in Series A funding in 2018.

The investment will be used to further develop the Woebot's natural language processing abilities to expand the delivery of mental health care worldwide.

...it was also featured on The Today Show!

REFLECTION

Chatbots present unique design challenges.

Users consistently expressed frustration with the limitations of chatbots and the inherent lack of fluid conversation that can take place in an interaction. Unfortunately, as AI-based bots are still a nascent field in UX design, there is no one solution to this problem.

...particularly in the realm of mental health.

Therapy is a tricky subject to deliver via technology due to the inherent risks involved: giving the wrong advice, accidentally missing a warning sign of suicide, etc. We remained cognizant of these intricacies throughout our design process and were careful not to claim Woebot as a complete replacement for therapy.

Woebot's instant popularity indicated that we had tapped into an unaddressed market need.

One week after product launch, Woebot's user base skyrocketed from 100 to 50,000 users. We were absolutely thrilled to learn that we had correctly identified (and solved for!) a market need that was distinctly underserved.

< PREVIOUS PROJECT
NEXT PROJECT >
OVERVIEWPROBLEM SPACESOLUTIONGENERATIVE RESEARCHEVALUATIVE RESEARCHPUBLICATIONDESIGN DECISIONSIMPACTREFLECTION

© Molly Vierhile 2019