🧠 The Robot Will See You Now

The latest on the business of mental health and wellness

Welcome to On The Mind, a collection of stories, news, and analyses on the startups, investors, and thought leaders in mental health and wellness.

Here’s what’s included in issue No. 11:

  • How conversational AI can be applied to tackle mental health at scale

  • The power of honesty, even from robots

  • Promoting small talk, in-person, with real human beings


🎙️ Interview with Alison Darcy, Founder and President of Woebot

Conversations with founders, investors, and thought leaders in mental health and wellness.

I recently spoke with Alison Darcy, who in addition to founding Woebot, is a PhD, clinical research psychologist, software developer, Stanford lecturer, and repeat entrepreneur.

Alison founded Woebot in 2017, and it has since grown to over 50 employees, has over 4,000 5-star reviews, and exchanges millions of messages every week, with a 98.9% accuracy rate in detecting when a user is going through a crisis. With the well-documented shortage of mental healthcare providers, can a ‘robot’ like Woebot help people manage their mental health?

Can you walk me through your background and how you came to start Woebot?

I initially got into the mental health field with a focus on eating disorders, though even that was something I accidentally found myself in. It was the dot com boom, and I went to London to work at an investment bank, where I taught myself how to code as a developer. It was a small American bank that got taken over by a Swiss bank, around ‘99 or 2000, when the internet was still fairly new.

Eating disorders were prominent, and there were networks of support groups, but the existing solutions weren’t sustainable. We built an online group to translate things to the internet. It was an interesting challenge to deconstruct what’s helpful about an in-person thing and reconstruct it in a digital way.

From there, I went to graduate school, continued researching eating disorders, and went on a treatment development track for several years, leading me to adolescent psychiatry. That’s when I started paying attention to what happens to patients once they return home - they don’t have anything. Therapy was becoming more of a mainstream model - this idea that if you empathize with someone, they get better. In reality, therapy is much more complex.

I was getting deeper into the field, working with larger data sets, and eventually called Andrew Ng to be on one of my grants. We hit it off and collaborated, and I went on to run Andrew’s Health Innovation Lab in Computer Science at Stanford.

A lot was converging to nudge me out of academia and launch a company. It seemed the only way to make my work sustainable was to make a good product and turn it into a company. I did a stint at Stanford’s GSB where I participated in a program called Ignite, which gave me the language I needed to start Woebot.

What are the Woebot offerings today?

Woebot uses AI and NLP techniques that have the ability to form a therapeutic bond with users that leads to better outcomes. There’s a wide variation in sophistication of what we’re talking about when discussing AI - it’s simply a tool with good or bad uses. People are turning to their phones when they’re miserable, so we see that as a perfect opportunity to meet their needs.

We offer digital behavioral health products that address adult mental health, adolescent mental health, maternal mental health, and problematic substance abuse.

There’s something we call Woebot Live, which is our free baseline product. Woebot Live serves as our R&D engine where we gather data for our algorithms.

Then there are our prescription digital therapeutics (DTx), which are still being piloted and require high standards of clinical validation. These come down to whether you really believe software will eat the world, and if that applies to psychiatry. Psychiatry won’t be the fastest to adopt DTx, but almost everyone has a cell phone on them, which makes DTx a powerful solution. Behavioral health products are already being recommended by physicians in closed network payer groups, patients are using them, and they’re working.

We have three DTx in our pipeline: one for postpartum depression (which just received FDA breakthrough designation), one for adolescents with mild to moderate depression, and one for substance use disorder.

You mention the adolescent population specifically. We’re seeing more and more companies in the mental health space build products around specific segments, be it parents, teenagers, men, women, LGBTQ+, BIPOC, etc. - are distinct products necessary?

We built Woebot for adolescents to start with. Our first randomized controlled trial was with young adults.

What we discovered subsequently was it was appealing to so many people and the age range was much broader. We get a ton of emails form people in their 70s or 80s.

I think the products for distinct segments is more about choosing a beachhead, crossing the chasm, and doing one thing really, really well. Mental health is episodic, and we’ve been missing the longitudinal perspective for a really long time.

That’s how we try to design Woebot, even if we start with specific populations or points in time. People should be able to use Woebot for awhile, then stop, then come back and use it again, and have something that remembers what worked well last time. If you’re in an episode your brain doesn’t function as well as it does when you’re not, and you may need a friendly guide to show you that you still have the skills to cope, someone who sticks around through your life. This continuity allows for more relevant care - if you had a baby 6 months ago, you’re not just getting generic mindfulness tips, for example.

How does Woebot assess the user and actually figure out the right level of care someone needs?

Woebot is really good at meeting people where they’re at; we refer to it as the responsive layer. Some people need a little on the spot talking, some need a full course of therapy. As a conversation agent, Woebot asks people how they’re doing in a naturalistic sense, which captures data embedded in their lived experience. It lends itself well to shifting to whatever people bring to the table.

As far as identifying when someone needs what level of care, we use NLP algorithms for various specific tasks. That said, the best way to know how someone is doing is to just ask them. Someone conducted a principal component analysis to figure out the most predictive item possible, and it turns out that asking “how depressed have you been in the past 2 weeks?” is the best way to discern depression. Many in the field think of people with mental health issues as not being able to have insights. In reality, most people with depression or anxiety can actually just tell you how they’re doing if you ask.

Officially, depression and anxiety are psychiatric disorders, but these exist on a spectrum. Symptoms of depression and anxiety are something that all humans will go through at some point. In a fully comprehensive healthcare system, you don’t want to give everyone a highly-trained specialist, because not everyone needs one. A core element of Woebot is being as noninvasive as possible. In the US, we think of the gold standard as being high-quality therapy. Really, it’s offering the least invasive, lowest intensity care you can that still derives value and gets the patient to remission. It should be the easiest.

Who does Woebot serve today and where do you see it going?

Woebot is used by people of all ages, from 18-80+. About 55% of our users are female, 45% male. 94% of the world’s countries and territories are using Woebot.

Our consumer app is free. We monetize through partnerships with payers and big health systems. We’re in the DC Housing Authority and Ontario West. We want to continue expanding our reach and will keep working on our DTx pipeline, which we expect to begin commercializing in 2022.

Woebot is designed to be an untintimidating entry point. Wherever there’s a big challenge in reaching a lot of people, getting them well, and doing that at scale - that’s where I see Woebot continuing to grow.


🩺 Clinical Coverage

Discussion of clinical concepts, studies, or perspectives on mental health and wellbeing.

Using chatbots in mental healthcare is not new; the concept has long been criticized as a “wishful thinking” approach that can’t possibly have the same efficacy as human intervention. In a small study in 2019, Cambridge researchers found that, when compared to a human therapist control, chatbot-provided therapy was “less useful, less enjoyable, and their conversations less smooth.”

A recent study on digital conversational agents, using Woebot data, investigated whether users of cognitive behavioral therapy (CBT)-based conversational agents could create similar levels of therapeutic bonds to those established in other CBT-based modalities (e.g., face-to-face therapy, group CBT).

To test this, the study surveyed Woebot users with the Working Alliance Inventory-Short Revised (WAI-SR) test, which assigns a bond score between the user and Woebot. Data from over 36,000 Woebot users was included, and the average bond subscore for Woebot was 3.84, in a similar region to the average bond subscores for human-led individual CBT (4) and group CBT (3.8).

More interesting, Woebot performed stronger than other internet-based CBT conversational agents. Researchers speculate this is driven by the transparency embedded in Woebot’s design. While other chatbots attempt to seem like a real human, Woebot was built to do the opposite - to be unabashedly transparent. “Woebot explicitly references its limitations within conversations and provides positive reinforcement and empathic statements alongside declarations of being an artificial agent.”

In other words, because Woebot is honest about being a robot, it actually performs its job better.


💰 Recent Investments, Acquisitions, and IPOs

Rundown of recent investment news in mental health and wellness companies.


📖 Interesting Reads

Sometimes mental health-related. Sometimes just things I find interesting.


🧠 Mindfulness Tip of the Week

Tips to improve your mental health and wellbeing.

As in-person socializing returns, so does “small talk.” Inevitably, these kinds of conversations can feel inauthentic and forced.

After missing out on this for so long though, it might actually feel welcome. The Walrus goes so far as to say the recent lack of small talk is “breaking our brains.”

Small talk plays a role in helping us feel more connected and happier. Research has shown that a conversation with a stranger of 10-20 minutes can be beneficial.

So how can you bring more small talk into your life? This guide provides a few suggestions, including asking questions related to your immediate context, leaning on some go-to prepared questions, or bringing in knowledge of current events. The goal is to make a connection and actually get to know the person (so essentially, starting with small talk, but elevating it as you go).

When in doubt, turn to Larry David for some inspiration:


On Your Mind

Email me at tarockoff@berkeley.edu with any reactions to the newsletter.

If you’re working on something in mental health and wellness, let’s talk. You can book some time with me here.


Know someone who’d enjoy On The Mind?

Share On The Mind

Was this newsletter forwarded to you?


Written by Daniel Tarockoff, an MBA student at UC Berkeley and former healthcare strategy consultant exploring the future of mental health. Born in Michigan. Based in Berkeley, CA.