🧠 Creating a Sound Mind

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. 5:

  • Female-led ML that can identify depression from 20-seconds of freeform speech

  • Dissecting the different roles across psychology and neuroscience

  • Helpful tips for breaking phone addiction


🎙️ Interview with Grace Chang, CEO of Kintsugi

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

It was about two years ago when Grace Chang met her co-founder Rima Seiilova-Olson at a hackathon. They bonded over a couple of similarities. First, their immigrant path to the US, and second, their shared struggle in accessing mental health services. They were both Kaiser members, and each experienced an unnecessarily complicated journey of accessing care over a period of five months. Grace was dealing with work-related stress, and Rima was working through postpartum depression. The two of them recognized a need to rethink the space from a different lens. Drawing on their technical backgrounds in machine learning, they came together to build Kintsugi – what they dub as “smarter mental healthcare.”

Kintsugi uses voice biomarkers to score clinical depression and anxiety from less than 20 seconds of freeform speech. They’ve secured millions in funding from the National Science Foundation’s SBIR grants and were selected as one of just ten companies from thousands of applicants to join the latest cohort of the UnitedHealthcare Techstars accelerator.

I spoke with Grace about how Kintsugi became what it is, and where she’s taking it next.

Your background is very technical, but not healthcare-focused. Why did you decide to build a company in this space?

Usually, people who go down this path do it for a very personal reason, and it’s no different for me. I had a really hard time accessing proper mental healthcare at a time when I was struggling. It’s easy to shake off looking back on it now, but at the time it was significantly impacting my life.

Before Kintsugi, I was the head of product for another machine learning startup in the Bay Area. We were extracting feature vectors from wearables and mobile devices to uniquely identify individuals based on how they move. The company had a lot of early success, and we went on to raise a $20M Series A from NEA with just 5 people on the team.

That was a pretty transformative experience, in that we were able to learn quite a bit about taking signal processing in one space – security – and apply it somewhere more personally meaningful. I think based on Rima and I’s backgrounds of building highly sophisticated products in pretty regulated fields, it felt really nice to hone our efforts into something that felt like it needed change, and in a space where we felt we could make an impact.

And how did you approach building Kintsugi?

We started with this kernel of an idea around applying our expertise to the field and began thinking about why mental health doesn’t work the way that it should. For us, we saw it as, well, in physical health we have our Apple Watches and Fitbits which give us data such as our steps as a proxy for how we’re doing, but in mental health, there really isn’t anything analogous. There are patient questionnaires like PHQ-9 and GAD-7, but those can come with a whole host of compliance issues and can be difficult to get patients to consistently complete.

As with anything, it’s really hard to improve something when you don’t measure it. We knew that there was a long body of clinical work related to measuring voice biomarkers, but these studies were very small. At the end of 2018, Jim Glass produced a paper at MIT that took contextual representations of what people said along with how they were saying it, and then assigned a binary “depressed” or “not depressed” score from the data.

Rima and I wanted to see if we could solve this problem for ourselves. When we looked into which clinical sets of work were most impactful, we quickly came across Pennebaker’s paradigm, which instructs participants to write about either emotional events or neutral topics, and shows that those assigned to the emotional writing condition typically display both physical and psychological health improvements after the exercise.

So basically after journaling about their emotions?

Exactly. I used to journal in middle school and high school, but life happens, you become an adult, you have other responsibilities, and I just stopped. But I thought, since people are talking to their devices already, would people speak into an application to journal? That didn’t really exist at the time. I wanted to create a very rudimental voice journaling application so that what you say could be transcribed and kept over time, which was the initial MVP. At first it was just Rima and myself using it, but eventually it spread through word of mouth. Today, we have over 100K downloads with users across 250 international cities, and that has given us the world’s largest annotative voice repository for machine learning, and specifically for mental health.

This enabled us to take the work that Jim Glass’ group had completed on a small scale and apply it to a much larger data set. And what did we find? We found that dropping the contextual representations of what people were talking about, and focusing just on how people were speaking, we could beat state of the art detection with our models. We could produce real-time predictions with a much more elegant architecture, and our models were language agnostic – we weren’t looking at the words, we were looking at which features of how people spoke were most predictive of depression and anxiety biomarkers. It still amazes me that I can speak both English and Chinese and get the exact same results from Kintsugi - it’s fascinating that human emotion isn’t really that hard to detect, and is quite universal. You don’t need to understand exactly what someone is saying to understand how they’re feeling.

When you think about audio data, it’s just a wave, and there is a shape to that wave form as well as patterns associated with the dimensions of that wave that we can correlate with an individuals’ self-reported depression or anxiety scores.

We took that as an endpoint, and started to license this technology to major players across healthcare.

Speaking of licensing the technology, who are Kintsugi’s customers?

During Techstars, we had over 430 meetings set up across 15 weeks, which was intense but helped us understand who the most important buyers were based on their willingness to pay and desire to move fast. It wasn’t immediately clear from the start who would be the ideal customer.

The program provided an incredible opportunity to speak with payers, and that led us to build a lot of Kintsugi around their needs. Insurance people are typically really hard to get in front of because people try to sell to them all the time, so having the UnitedHealthcare connection helped.

We had a lot of people who wanted the service immediately, but pricing was the biggest hold up - we needed to make the economics work in-line with the existing mental health ecosystem. Some standalone digital health services only bring in about $30 per coaching visit, so there were guardrails on how we priced our offering, for example.

Our platform is focused on IP licensing – essentially all of the data we’ve been able to amass from our robust models. We have a service where we charge for every patient we screen. Our focus has been on enterprise with companies like Amazon Care and Microsoft, but we also offer a consumer application. On that front, we use KiVA™, the Kintsugi voice biomarker API, which provides real time feedback from audio journal entries to document what your depression or anxiety looks like over time. You can also get some visibility into what triggers you’re prone to, such as the role exercise, sleep, or even the weather has on your mood. Our consumers love using the app because it’s a safe place to vent without judgement, and we have a community-based component that uses matching to highlight anonymized entries from other users around the world going through similar challenges.

Could you help me make sense of the audio-related competitive landscape when it comes to mental health, and how Kintsugi is different?

Audio is definitely an emerging part of the mental health industry, and there are a number of folks trying to get signal in what is a very challenging field: Mindstrong, Sonde Health, Winterlight Labs, Ellipsis.

Our main differentiator is our data acquisiton strategy. A lot of our competitors are acquiring their data by clinical trial, which is maybe a dozen or so individuals at a time. It’s going to be damn hard to get to something meaningful with any sort of machine learning when you’re talking about such a small sample size. We don’t think that’s the right approach, and the dataset we’ve collected from audio journaling provides a significant advantage.

We’re also really confident that we hit the sweet spot with the right team. It’s rare enough to have the expertise across security, enterprise, consumer, product design, and machine learning, let alone having a genuine interest and passion for solving the problems surrounding mental health.

Why do you think there’s a trend with audio more broadly right now? You see social apps like Clubhouse making waves, and companies like Spotify positioning themselves as “audio-first.”

I think, if I’m not mistaken, Mary Meeker put out her annual Internet Trends report about a year or two ago saying that audio is one of the media formats growing fastest, in the double digits year-over-year. I don’t know if it’s because people can listen to audio while doing other things, but I do think our ability to consume audio is relatively easy and that probably plays a role.

In terms of the machine learning side of things, it’s exciting because we have models that are able to ingest this kind of information and give us something valuable that we didn’t have before. There’s been a lot of advances in the space where this type of deep learning can now be built, enabling us to run new kinds of experiments. As entrepreneurs look at whether they want to build in something that’s red ocean or blue ocean, I think the blue ocean strategy where there’s limited competition and a chance to reinvent or define something new altogether is exciting, and we see that with audio.

What’s your fundraising journey been like?

We had initially only set out to raise a $1M Seed round, but we ended up well oversubscribed. We raised $3.5M from the National Science Foundation alone and additional capital from angels and a local fund.

With our seed funding, Rima and I have been able to make our first full-time hire with our Head of Operations, but we’re excited to bring on even more people to help us tackle our growing sales pipeline.

Once we finalize some new statements of work, we’ll be out fundraising again for our Series A - we’re ready to scale this.

It’s important to us as women founders that the folks we end up aligning with actually hear what we’re saying. Some of our conversations have been really one-sided. People want to tell you how to run your business without listening to your story. We really care about mission alignment and seek out a genuine interest from our investors in wanting to improve access to mental health. One of our first investors, Adam Grosser, who’s a former MD at Silver Lake and a former GP at Foundation Capital, is a great example - he’s awesome. We really couldn’t ask for better people on our cap table.

We’ve spoken to some investors who’ve said they don’t understand if depression is a big deal or not, and we’re never going to reach people like that.

What’s next for you and Kintsugi?

When it comes to long-term thinking, our mission remains broadening access to mental healthcare.

We’ll look at anything that helps us acheive that goal, from running new experiments to improve our algorithms to securing digital therapeutic status for our consumer app. As the world continues moving toward precision medicine, collecting more data and getting sharper about how we use it will help us play a more effective role in taking care of mental health.

And similar to physical exercise, there are a variety of different things to do to stay fit. I don’t think it’ll be just one tool to solve all of the problems. We’re excited to be one part of the solution.


🩺 Clinical Coverage

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

Outsourcing this week’s clinical coverage topic to this piece produced by Ness Labs.

One of the biggest factors in the quality of mental health care received is who the care provider is, and the type of training they’ve undergone.

Different backgrounds can contribute different insights, and each type of practitioner has a role to play. It’s not a hard-and-fast rule that you absolutely need to involve a psychiatrist, for example, but one significant issue with the system as it exists today is the volume of misdiagnosed or misprescribed patients that stems from primary care physicians that aren’t properly trained in mental health care delivery.

It’s important to understand the nuances between mental health professionals as a founder (who’s determining the direction of your products/services?), investor (does the team have the right technical expertise?), or patient (am I receiving care from the right source?).


💰 Recent Investments and IPOs

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

Investments in the space continue to pour in:


📖 Interesting Reads

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

  • Creative trends to look out for this year. If anything, it’s a fun visual scroll (Link)

  • Glastonbury’s cancellation is not a good sign for summer music festivals (Link); neither is the cancellation of Miami’s Ultra (Link)

  • Engineering ambitious ‘megascale’ structures (Link)

  • Ever wonder what it’s like to be hot? It sounds nice (Link)

  • Simple, one-stop-shop for getting your fitness plan in place (Link)

  • Spotify is working to customize what you hear based on your mood (Link)

  • Practical tips for how to read more books (Link)

  • Meditation and sleep find their way into the entertainment world (Link)

  • Want to go to space? (Link)

  • How to create your own luck (Link)

  • This detailed guide on how to make smarter playlists is worth the read for the methodology alone, even if you don’t incorporate it (Link)

  • The race for the next $1B booze brand (Link)

  • A new essay on Apple from Matthew Ball (Link)


🧠 Mindfulness Tip of the Week

Tips to improve your mental health and wellbeing.

The past few weeks I’ve been spending way too much time on my phone. It’s so easy to get sucked into endless scrolling.

There are a lot of practical approaches to reducing phone usage: set app limits, track and actively manage your screen time, move popular apps off your home screen or even delete them.

I’ve found one quick tip that has saved me from hours of distractions: setting up triple-click grayscale. There’s something about removing the colors from your display that just makes it so much easier to step away from your device.

For iPhones, you can enable this by going to Settings > Accessibility > Accessibility Shortcut > Color Filters. After enabling, just press the side button three times to turn the grayscale on/off.

Bonus: this article has some additional tips for how to better manage which notifications you receive from each app.


On Your Mind

I’d love your feedback - feel free to email me at tarockoff@berkeley.edu.

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.