Anmol Madan, Livongo, on using data science to transform care delivery

Stay updated on our releases: subscribe on Twitter, Spotify or Apple.

In this episode, we sit down with Anmol Madan, Chief Data Scientist at Livongo, where he is responsible for machine learning and data science across all of Livongo’s products and services, to help people be the best versions of themselves.

Anmol previously co-founded Ginger.io, an AI-driven behavioral health system that provides access to 24/7 coaching, therapy and psychiatry, and served as its Chief Executive Officer for 7 years. Under his leadership, Ginger.io built an AI-driven member and clinical product, established a high-growth distribution model with employers, became broadly available as an in-network health-plan benefit, and raised $35 million in venture funding from top-tier Silicon Valley VCs.

Anmol received a Ph.D. in machine learning applied to human behavior from the MIT Media Lab, and has authored 20+ scientific publications (in Science, AAAI, Ubicomp, NIPS), and over a dozen patents related to machine learning in healthcare. He has been recognized as one of the 100 Most Creative people, and as a World Economic Forum Technology Pioneer.

START — 6:00 Early career path

• “What did you want to be when you grew up?” As a child, Anmol watched a ton of action movies, like James Bond or the Matrix. He never wanted to be the main star — instead he longed to be the “hacker guy” who breaks into systems and controls all the networks.
• How hacker dreams led Anmol to where he is now: Anmol spent the past few decades training computers, smartphones, and other devices to better understand humans. He completed his PhD at the MIT Media Lab, where he worked on machine learning. Some of his projects included mathematical tone of voice models, how fast we are speaking, and tying these things into different contextual settings. His thesis explored using mobile phones and sensor data to understand how ideas, music, opinions spread across our society.
• Founding Ginger.io: His work at MIT generated meaningful insights in the healthcare space, too: when people are sick, their behaviors change, and phones actually detect these behavioral changes. With these findings, Anmol started a company called Ginger.io, where he was CEO for seven years. The company focused on applying machine learning to behavioral health, stress, anxiety, and depression. Of all of the traits that characterize depression in a clinical setting, most can be measured using a smartphone, such as movement patterns, or the time you went to school or work. Ginger.io raised approximately $35 million during Anmol’s tenure.

6:00–9:35 Transition to Livongo

• Leveraging learnings from Ginger.io: Anmol started as Livongo’s Chief Data Scientist about one year ago. This opportunity is a great chance for him to double down on his work from Ginger.io, now applying his expertise in machine learning and AI to the “whole person” and using this to improve chronic conditions. Livongo has products in diabetes, hypertension, weight management, and a growing presence in behavioral health.
• From CEO to Chief Data Scientist: Anmol describes himself as a very product-driven CEO, as opposed to a sales-driven CEO, given his experience on the product and technical side of things. He describes his role as Chief Data Scientist as not so dissimilar from that of a product-driven CEO; however, the biggest shift was the actual scale and stage of the company. Obviously joining a near-IPO company is much different than building a startup out of your bedroom.
• Joining Livongo pre-IPO: Anmol joined Livongo right as the company was about to go public. Livongo was the first digital health IPO in three years and the largest digital health IPO in history, which made for a really interesting and exciting time to join. It wasn’t just about doing great work for members and clients (which was also critical), but also about validating an entire industry. Digital Health has seen multiple billions of dollars in investment over the past few years, and Livongo’s IPO put them front and center of this industry-wide validation.

10:00–14:00 What exactly is data science? What does it look like at Livongo?

• Livongo’s vision: Livongo considers itself to be an “applied health signals” company. They think about themselves in the context of “AI+AI”, which stands for: aggregate, interpret, apply, and iterate. Their vision is to understand members at a level of richness that doesn’t really exist anywhere else in healthcare. Getting really, really good at understanding members allows Livongo to personalize how to support their members so they know interventions are likely to work for them, from support to coaches to medication management to new kinds of sensors.
• Deeply understanding the member: Livongo’s vision is derived by a desire to understand their members better than anyone else in healthcare, using this knowledge to build a core product and technology experience. By tying these things together, Livongo is able to drive strong clinical outcomes while also satisfying their customers: they have a net promoter score of 65, on par with consumer tech companies like Apple and Netflix.
• Why this matters: Chronic condition management for diseases like diabetes and heart disease poses a significant cost on society: 90% of US healthcare spend is related to chronic condition management. Over 147 million individuals in the United States have one or more chronic conditions. We all know people — friends or family members — for whom this is relevant, and it poses a huge economic burden. It’s important that we support these people, yet this support really needs take place outside of the doctor’s office, when people living with diabetes and hypertension make everyday decisions. It’s really challenging to power through these decisions when presented with day-to-day stresses and anxieties, which are only exacerbated in the current climate.
• Tying this back to data science: Drawing this back to Anmol’s role in the context of the company’s mission, data science and machine learning clearly play a huge role in pushing Livongo forward. It makes his job “fun and very interesting” because he gets to work on problems across different parts of the business in support of this overarching goal. Today, Anmol’s Data Science org at Livongo comprises six different teams all working on machine learning, AI or data science and how these things can be applied to different parts of the business, such as personalizing hardware devices, experience with software devices, applied health signals and sensor data.

14:00–16:15 Using data science in practice

• Changing the service model: Livongo’s data teams closely with their business operations, coaching operations, and business forecasting units. Anmol recently published a blog post in which he unpacks the existing service model in the context of healthcare: member services and coaching, primarily. Instead of waiting for people to metaphorically “walk in the door,” what if there were systems that could predict who would need help and when, flipping the interaction in the other direction? Livongo is building predictive systems to determine who needs help and what type of help is best for them.
• Recent partnerships: Livongo recently announced a partnership with Dexcom, who is a leading continuous glucose monitor (“CGM”) vendor, which enables Livongo to bring CGM data into their system. They also recently announced a partnership with leading labs company Prognos Health, which allows them to bring in clinical labs data. Both of these are ways in which Livongo can better understand their members in a way that marks a significant improvement over how the system currently does.
• Applications in marketing: Anmol is also excited to work with Livongo’s marketing team, too. There is a huge opportunity to experiment and learn about how Livongo talks to different types of members, different demographics, and how Livongo messages and communicates potential campaigns, from retention to value to enrollment.

16:15–20:35 Building foundations to gather data while earning and maintaining member trust

• Foundation 1 — be member-centric: Start with a member-centric view. For example, thinking about how someone on the Livongo team might want to use a service, how they would want a product to work with them, or how marketing would best interact. This includes member privacy and thinking about how individuals grant permission for their data to be used. To support someone as a person and potential member, they need to have visibility and transparency, knowing that their data is private and secure.
• Foundation 2 — maintain clinical rigor: Social media can send many messages through consumer-targeted marketing. However, in healthcare, it’s not enough if a person simply clicks on a message: you also have to understand if you are nudging people towards the right clinical action, which is a level deeper than converting a click. Did it move people in the right direction? Did it have efficacy? There is a clinical validation piece, such as publishing at conferences (among other avenues) to ensure they get this right. To date, Livongo has had approximately 32 publications and abstracts, something Anmol works closely on with the company’s Chief Medical Officer, Dr. Bimal Shah.
• Keep the goal in mind: In addition to the two foundations, Anmol says you still need to focus on a clear goal. Machine learning systems today are very sophisticated, and there are a lot of off the shelf tools in which you can throw in a bunch of data and have it start predicting things for you: the bar is quite low. The focus needs to be on the business/product use case. For example, if the use case is “let’s get our members to engage with actions or behaviors”, then it becomes very easy to start personalizing their experience. On top of this, you need to embed measures and metrics that can tell you if the systems are working in the right direction. Think of it as how Netflix makes recommendations and then validates their recommendations by whether you rated the show or movie with a thumbs up or a thumbs down. It’s just as important to validate that the recommendations you made were appropriate as it is to make them at all.

20:35–26:30 COVID’s impact on digital health

• Industry shifts: The last six weeks have changed our industry in a big — and potentially permanent — way. The impact of the pandemic on our society and on people’s lives is incredibly sad, and a side effect of this is an acceleration of recognition for digital health companies, telemedicine and remote monitoring. This is true across the commercial sector, healthcare systems, and the government, too.
• Next set of challenges: Lots of clinical care has moved to a digital interface, such as via smartphone or texting. However, the resource constraints (e.g. the number of medical professionals) haven’t changed. There’s a huge opportunity for AI and machine learning to meet this need — why can’t many of the initial assessments happen on digital platforms by algorithms or algorithms working interactions with humans, or medical doctors as appropriate?
• How Livongo fits into this new world: Livongo has a very large member base, particularly large for a digital health company. They also have a large set of sensors out in the world collecting data. There are good risk models that people have built from the clinical side, given what is known about the factors that put people at higher risk for COVID-19, such as chronic conditions. As society gets better at testing and surveillance, Livongo can bring these external insights into their own model, which allows them to do dynamic risk modeling, which then allows them to understand who is at the highest risk. Livongo can work with their healthcare partners to make sure that individuals get the support that they need.
• Livongo and behavioral health: Another pathway Livongo is seeing is from members who come to them on the behavioral health side. For example, a lot of members managing their chronic conditions are also experiencing behavioral health needs: people who have kids, who have been hit harder than others, economically or through their health care. Many of these members, who already work with and trust Livongo, are seeking help through the Livongo platform they already use, which saw a 140% increase in utilization via unique web or mobile logins between September and March.

26:30 —33:10 On leading during COVID

• Emphasize focus & prioritization: Livongo is a high growth company, so there is a ton to get done. During COVID-19, though, people might come to work on Monday and only be able to give 70% because they’re also taking care of kids. Perhaps people have other constraints, like looking after a family member who isn’t staying with them. In these situations, knowing what you actually need to get done becomes that much more important. This requires a heavy operational focus, too.
• Overcommunicate: Anmol says effective communication is key, and this often entails getting creative with format. If you want to do a team brainstorm, for example, you might have to turn it into a written format, as opposed to in person. It can be worth doing that bit of extra work to make things a little more manageable for people. Another example is just holding office hours for his team, where anyone can openly dial-in. Without the hallway interaction, you have to figure out a way to move to a digital format.
• Encourage self-care: It’s also really important to encourage people to take time for themselves. Many people need to help their kids with school every day or go on a grocery run, which limits the amount of time you actually have to do the things you’re committed to doing. As a manager, Anmol helps his team figure out how to scope work down. “This is a marathon — it’s not a couple of weeks.” Anmol says that giving ourselves that space, protecting our mental state, making sure we’re taking time out for ourselves for things that rewarding and fulfill us, are critical.
• Working in healthcare, now: Anmol describes the energy and passion at Livongo right now, for example the speed at which they rolled out a behavioral health offering to their entire member base. The FDA’s recent announcement that Livongo can be used in inpatient settings. People at Livongo remain super passionate because they recognize there is a huge opportunity to have an impact, to help at risk people and even to save lives. This is an important guiding force at the company.

33:10–END On roles at Liovngo & career advice

• Growth!: Livongo plans to grow aggressively, especially over the next six months. There are positions across the board in Business Development, Comercial, Sales/Marketing, Product Management, Data Science, and Engineering — job board can be found here! There are additional roles in Anmol’s org related to data science and analytics, including managing positions. It’s an awesome time to work in digital health & applied health signals — there’s a chance to shape the industry at this crucial juncture.
• Joining a startup versus pre-IPO company: When he has explored roles previously, Anmol has a mental list of hypotheses about the direction the world I going in, and he makes his decisions based on this rubric. Hypothesis #1: healthcare is making a shift; consumers are changing how they want to interact with healthcare and wellness in general. Hypothesis #2: We're only scratching the surface when it comes to the role of machine learning & AI. A lot of companies are thinking about both of these things, yet only a few have the mindset, the DNA, the culture to bring it to market. This is an interesting thing to consider if you’re on the job market.
• On starting a company now: Lots of great companies are started in times such as now! Anmol shares that when he wrapped up grad school in 2008, everyone else was getting a job at a large tech company or an investment bank — he was going to a startup. If you have an idea and believe in yourself and your vision, there’s never a bad time to start a company. You might actually see less competition (though also more constrained capital). Having a good hypothesis about the core reason this company will be successful and why it should exist but doesn’t yet serve a founder well.
• From the technologist angle: Lastly, especially as a technologist/entrepreneur, it’s important to think about how the business model will work in healthcare. Healthcare is generally more obscure when it comes to the role that different entities play, which make their business incentives just as complex.
• Final words of advice: Our society needs people who want to build businesses or join companies that will make the world a better place. Healthcare is a great place to do that.