Episode #14 - Matthew Woo, Co-Founder, President, and Chief Product Officer

Written By Omar Mousa

Listen On

 

Description

Matthew Woo is the Co-founder, President, and Chief Product Officer at Summer Health, where he heads the development of a tech-enabled, text-first pediatric care platform. Summer Health provides parents with fast, convenient access to pediatricians through text, offering responses within minutes. Matthew has been instrumental in integrating AI into their workflows, ensuring that busy parents can manage their children's health needs quickly and effectively while aiming to reduce potentially unnecessary emergency room visits or lengthy waiting room stays. In our conversation, Matthew shares his insights on:

  • AI Basics for Healthcare: Matthew walks us through the fundamentals of AI, offering a clear explanation of how AI is being applied in today’s healthcare landscape and what makes it such a powerful tool for clinicians.

  • Operationalizing AI in Healthcare: Insights into how Summer Health has deployed AI in low-risk, high-reward areas, the importance of rapid iteration, and how they ensure data privacy and security in their AI deployments.

  • AI Integration at Summer Health: Matthew details how AI has been integrated into clinical workflows at Summer Health, addressing inefficiencies and improving patient care through tools like medical note generation and patient history summarization.

  • Scaling AI and Key Metrics: Practical strategies for scaling AI across teams and the key metrics that Summer Health tracks to measure the success and impact of their AI initiatives.

  • Ethics and AI in Healthcare: How Summer Health ensures that AI tools are safe, fair, and bias-free, with clinicians involved in every critical decision made by AI.

  • Building an AI-First Culture: The steps Summer Health has taken to foster an AI-first culture, including training and empowering teams, aligning AI with strategic goals, and fostering a culture of experimentation and rapid iteration.

In this episode of Concept to Care, Matthew provides a deep dive into how AI is transforming healthcare at Summer Health. His insights on the practical application of AI, scaling its use, and building a culture that supports innovation and collaboration are invaluable for healthcare product leaders looking to leverage AI in their own organizations.

www.concepttocare.com

Some takeaways:

  1. AI Basics for Healthcare: Matthew Woo starts by walking the audience through the basics of AI, defining key concepts to lay a foundation for understanding its role in healthcare.

    1. Defining AI: Matthew describes AI, or artificial intelligence, as the use of machines or computers to simulate human intelligence. This includes decision-making, reasoning, and problem-solving, allowing AI to take on tasks that typically require human cognitive abilities. AI is not a new concept—it was first introduced in the 1950s—but its capabilities have evolved significantly due to advances in computing power and data availability.

    2. How AI Works: At its core, AI systems are built to analyze large amounts of data, recognize patterns, and make predictions or recommendations based on that data. For example, in healthcare, AI can be used to analyze medical images or predict patient outcomes based on historical data.

    3. Evolution of AI: Matthew highlights the progression of AI from its early stages to modern-day applications. Initially, AI was used to create "expert systems," which encoded human expertise into rule-based systems for tasks like diagnostics. However, with the advent of technologies like deep learning and neural networks, AI can now learn from vast datasets and adapt its responses based on new information.

    4. AI in Today’s Healthcare: In the healthcare sector, AI plays an increasingly critical role. It can assist in diagnostics, suggest treatment plans, and even predict patient outcomes. For example, AI can help radiologists interpret medical images or help clinicians identify patterns in patient data that may not be immediately obvious. AI’s growing presence in healthcare allows for faster, more accurate decision-making and the ability to provide care at scale.

  2. Different Types of AI

    1. Machine Learning (ML): Matthew describes machine learning as a branch of AI that uses large amounts of data to make predictions. ML has been traditionally applied in healthcare for tasks like classification and ranking, such as helping radiologists detect issues in scans. However, this form of AI requires substantial investments in data and infrastructure, and its use cases have typically been narrow and specialized.

    2. Generative AI: Matthew explains that generative AI has gained prominence with models like OpenAI’s GPT. Unlike traditional machine learning, which focuses on specific tasks, generative AI is capable of producing outputs such as medical notes or care plans. At Summer Health, generative AI is used to assist clinicians by drafting medical documentation, saving time and improving workflow efficiency.

    3. Transformers and GPT: Matthew references the development of transformer models, particularly those like GPT, which became more powerful with larger datasets. These models are able to generate coherent responses and assist in healthcare contexts, such as summarizing patient information or generating care plans, allowing healthcare professionals to focus more on patient care.

  3. Operationalizing AI in Healthcare

    1. Start Small with Low-Risk, High-Impact Tasks: Matthew shared insights on how Summer Health operationalizes AI by focusing initially on low-risk, high-impact areas such as drafting medical notes. Rather than applying AI to high-stakes clinical decisions, they began with administrative tasks where mistakes are less critical, allowing for faster deployment and measurable efficiency gains. He also emphasized the importance of ensuring data privacy by working with safeguards like having a Business Associate Agreement (BAA) with OpenAI, ensuring patient data security while using AI tools.

    2. Rapid Iteration and Clinician Feedback: Another key strategy was rapidly testing AI models in real-world workflows and collecting clinician feedback quickly. By tracking how often clinicians use AI-generated outputs and what adjustments they make, Summer Health continuously fine-tunes its models to improve both accuracy and user satisfaction, ensuring AI adds value without disrupting clinician workflows.

  4. AI Integration at Summer Health

    1. AI Integration into the Platform: Matthew explained that Summer Health started by integrating AI into low-risk, administrative tasks before expanding its use to clinical workflows. AI is deployed in non-real-time applications initially, such as drafting medical notes after patient interactions. This integration has streamlined processes and enhanced operational efficiency by reducing the time spent on documentation.

    2. Problems AI Addresses: AI at Summer Health primarily addresses the inefficiency in clinical workflows and repetitive tasks. For example, AI-generated medical notes allow clinicians to spend less time on administrative work and more time on patient care. It also helps with summarizing patient histories for specialists, reducing the need for patients to repeat themselves, and improving the continuity of care.

    3. Some of the Most Impactful Use Cases at Summer Health Where AI Was Deployed:

      1. Medical Note Generation: AI drafts medical notes based on patient-doctor interactions. This has cut the turnaround time from 48 hours to 15 minutes, allowing patients to receive their care plans more quickly and feel more supported.

      2. Patient History Summarization: AI summarizes a patient’s past visits and interactions, providing clinicians with a clearer picture of the patient’s medical history. This is especially useful when specialists or new doctors are brought into a case, improving continuity of care.

      3. Provider Efficiency: AI tools at Summer Health help clinicians by reducing administrative burdens, such as preparing reminders, care plans, and even helping with coding for billing. This frees up time for clinicians to focus on complex, high-touch patient interactions.

  5. Scaling AI and Key Metrics at Summer Health

    1. Strategies for Scaling AI: Matthew explained that scaling AI at Summer Health involves making AI tools accessible across teams and continuously improving them through feedback loops. AI applications are designed to be used by a wide range of team members, from engineers to clinicians, allowing rapid iteration and deployment. They’ve invested in creating a platform that enables quick integration of new AI use cases and encourages team members to explore new AI applications. This scalability comes from developing tools that can be easily applied and adapted for different workflows.

    2. Key Metrics Driven by AI: Summer Health uses several metrics to measure the impact of AI. These include:

      1. Clinician Utilization: Tracking how often clinicians use AI-generated outputs, such as medical notes or care plans, helps assess whether the tools are improving efficiency. They also measure how much clinicians modify AI-generated content.

      2. Time Efficiency: AI is tracked for its impact on turnaround times, such as reducing the time from patient interaction to the delivery of a care plan.

      3. Patient Engagement: They monitor patient satisfaction scores to ensure that AI-driven interactions improve the patient experience. Increased patient engagement and faster care delivery are key signs of success.

      4. Content Accuracy: By measuring the accuracy and helpfulness of AI-generated outputs, particularly in clinical settings, they ensure that AI supports quality care.

  6. Ethics and AI in Healthcare

    1. Human-in-the-Loop Model: Matthew emphasized that Summer Health maintains a human-in-the-loop approach to ensure AI does not make critical decisions without clinician oversight. AI-generated outputs, such as medical notes or care plans, are always reviewed and approved by clinicians before being finalized, ensuring that AI augments, rather than replaces, human judgment in high-risk medical situations.

    2. Bias and Fairness: Summer Health works to mitigate bias in AI models by carefully monitoring outputs for any unintended biases, particularly in clinical decision-making. They continuously refine AI algorithms based on feedback and results to ensure they work equitably across different patient populations.

    3. Data Privacy and Security: To protect patient data, Matthew highlighted the importance of strict privacy measures, including the use of Business Associate Agreements (BAAs) with AI service providers like OpenAI. This ensures that any patient data processed by AI models is handled in compliance with HIPAA regulations, protecting patient confidentiality.

    4. Transparent Accountability: Summer Health encourages transparency in AI usage by keeping clinicians informed about how AI tools are developed and deployed. This fosters trust and allows clinicians to understand and control how AI tools are applied in their workflows.

    5. Monitoring and Feedback Loops: The organization actively tracks how AI-generated recommendations are used and modified by clinicians. This monitoring and feedback system allows them to identify potential risks, continuously improve AI models, and ensure safety is maintained throughout patient interactions.

  7. Building an AI-First Culture: Summer Health has implemented several key strategies to foster an AI-first culture, ensuring AI is embedded in their core operations and workflows:

    1. Training and Empowering Teams: One of the first steps Matthew and his team took to build an AI-first culture was ensuring that all teams—clinicians, engineers, and even non-technical staff—understood the potential of AI and how it could be applied in their work. Matthew emphasized upskilling teams through internal training sessions on AI tools and technologies, which empowered them to identify areas where AI could be most impactful.

    2. Aligning AI with Strategic Goals: AI initiatives were always aligned with Summer Health’s larger strategic goals of improving patient care and increasing efficiency. Every AI project was vetted for its potential to drive these outcomes, ensuring that AI was a core part of the company’s strategy rather than a standalone initiative.

    3. Fostering an Experimental Mindset: Matthew mentioned that Summer Health encouraged a culture of experimentation where teams were not afraid to try new AI tools and approaches. By normalizing experimentation and learning from failure, they created an environment where innovation around AI was continuous.

    4. Rapid Prototyping and Iteration: To keep the AI-first momentum, Summer Health adopted a rapid prototyping and iteration process. Matthew mentioned that AI tools were quickly built, tested in live workflows, and refined based on user feedback. This iterative approach helped teams see immediate results, making them more likely to continue exploring AI-driven improvements.

  8. Pragmatic Approach to AI

    1. Start Small with Proven Use Cases: Matthew emphasized the importance of starting with low-risk, high-reward applications of AI. Teams should focus on areas where AI can demonstrate quick wins, such as automating administrative tasks or generating medical notes. By starting with simple, well-defined use cases, teams can build confidence in AI’s value before moving on to more complex applications.

    2. Leverage Existing Expertise: Rather than hiring large AI teams from the start, Matthew encourages teams to leverage existing in-house talent. Engineers and clinicians who understand the company’s needs are well-positioned to identify where AI can add value. AI tools today are accessible, making it easier for non-AI specialists to integrate AI into workflows with minimal training.

    3. Rapid Experimentation and Feedback: Teams should adopt a rapid experimentation mindset when deploying AI. Matthew recommends running small pilot programs, gathering feedback quickly from end users, and iterating based on real-world usage. This approach allows teams to refine AI models without making large, upfront investments, and ensures the AI tools align with user needs.

Show Notes

Where to find Matthew Woo:

—

Where to find Angela and Omar:

—

Referenced:

Check out our website: https://www.concepttocare.com

Subscribe to our newsletter: https://concepttocare.substack.com/p/episode-14-matthew-woo

Transcript

[00:00:00] Matthew Woo: And let me first define what AI First Culture means to us, which is a bias towards constantly asking for every decision that you make on a daily basis. Can AI replace, augment, or teach me something about what I'm trying to accomplish today? 

[00:00:18] Angela Suthrave: Welcome to Concept2Care, where we hear candid stories of success and failure, discuss strategy, and dive into the details that offer advice on what to do and what not to do in health tech.

[00:00:29] Omar Mousa: Whether you're a seasoned pro, growing your career, or just starting out, our aim for this podcast is to be relevant, real world, and tactical. We're dedicated to not only entertaining you all, but also empowering you with actionable insights that can be applied beyond the podcast, one concept at a time.

[00:00:45] Angela Suthrave: This is Angela 

[00:00:46] Omar Mousa: and this is Omar. 

[00:00:47] Angela Suthrave: Welcome to concept to care concept to care. Welcomes Matthew Wu today. He is the co founder and chief product officer of summer health. Summer health is the fastest way for a parent to get connected to a pediatrician through text message based care. They're on a mission to raise the healthiest generation.

[00:01:06] Angela Suthrave: And in this episode, we're excited to talk about summer health model. And because summer health has an AI first culture, we're going to go deep into AI. And we're also going to get his advice on how to build an AI first culture. We hope you enjoy this episode.

[00:01:26] Angela Suthrave: Matthew Wu, hello. 

[00:01:28] Matthew Woo: Hey, hey, it's been a minute. 

[00:01:29] Angela Suthrave: Yeah, welcome to Concept2Care. 

[00:01:31] Matthew Woo: Yeah, excited to be here. 

[00:01:33] Angela Suthrave: Why don't we start with you telling us a little bit about yourself? 

[00:01:36] Matthew Woo: Yeah, I'd like to say I started my healthcare career, uh, 30 years ago when I was three years old, sleeping underneath my mom's desk, cause she was actually running IT at the Toronto Sick Kids Hospital.

[00:01:50] Matthew Woo: Um, but in all seriousness, you know, growing up, I actually think I have a different story than a lot of, you know, founders, which is I actually tried not to do anything related to technology. My dad was a software engineer. My mom was in healthcare IT. Um, but with everything happening around the internet and smartphones and eventually the iPhone, that was kind of the breaking point.

[00:02:10] Matthew Woo: And so in college, after, you know, unfortunately getting a full time offer, a consulting firm decided, Hey, you know what, let me try to build a mobile application. I think it was really cool at the time and fell in love with the idea of talking to users, working with engineers, working with designer, and just building something.

[00:02:28] Matthew Woo: It was called shore up. It helped businesses identify people that prefer the business. Made all the mistakes, didn't know what I was doing, but really fell in love with the process. So, you know, launched got about 30, 000 users and then shut it down. Um, went to consulting, realized it wasn't for me. Um, and started to pitch to startups that could help them with their pricing.

[00:02:47] Matthew Woo: Cause that's what my consulting firm did and landed at meetup. Um, and they're like, Hey, do you want to do this thing called an associate product manager? Again, still didn't know what product management was interviewed with it. Not knowing what the role really entailed, except for the fact that, uh, I'd be working with engineers and got the job.

[00:03:03] Matthew Woo: And so that's kind of how I started my product career. Been in product for more or less the last 10 years at meetup. I built out their notification system. Sounds crazy. Um, and I feel really old, but they were like figuring out how to transition from desktop to mobile. And so we had to figure out how to build a notification system, um, from there, you know, worked with some small startups like, yo, high growth companies like intercom and eventually landed at Facebook, I guess now called meta working on their marketplace product for a bit.

[00:03:30] Matthew Woo: And then eventually back to my messaging, uh, roots at what's up, laying out their business messaging team. And while I was there, I always knew I wanted to be a founder, uh, as a boy, mainly because my father was an entrepreneur as well. And it was just something ingrained in me. And so when I got introduced to Ellen and started to learn about her mission to simplify the way in which people access care and ran a few early MVPs and talk to parents, it was very evident from the beginning that messaging would be a great way to help parents get that instant access to care, especially late at night.

[00:04:03] Matthew Woo: I think that's like the worst feeling as a parent feeling like you can't Help your child, whether it be a high fever, or maybe they fell down, or, you know, maybe there's a rash that's, you know, getting out of hand. And so just being able to, uh, help parents in their need was something that attracted me to join Summer Health as one of the co founders.

[00:04:23] Matthew Woo: Two and a half years later, uh, you know, the team is now, you know, 17 people, um, were available in all 50 States. Um, and yeah, we just raised our series eight and super excited to continue this mission. And we can talk a little bit more actually about what, how the mission has evolved, but that's a little bit about me and, uh, how I came to be at summer health.

[00:04:43] Omar Mousa: Thanks Matthew. It's a great background. We want to go a little bit more deeper into summer health. So could you give us like an overview of the mission, the business model and maybe a little bit about the platform? 

[00:04:52] Matthew Woo: Yeah, so summer health is the fastest way for a parent to get connected to a pediatrician through text message based care on average.

[00:05:03] Matthew Woo: The response time is actually 3 minutes. We have an SLA of 98 percent of the time people get a response within 15 minutes and we. We started off with our mission, which was to simplify the way in which people access care, and that's why we picked this SMS medium. We've learned since then that because it's so convenient for people to use Summer Health, they started to use this for a variety of things.

[00:05:26] Matthew Woo: Not just for urgent care, but also to ask things about how to sleep train, you know, uh, lactation support. Uh, behavioral questions. You know, if their child is falling behind on developmental milestones. And so we started to add specialists to our platform. We called it everyday care. Um, around that time, we started to ask our questions.

[00:05:44] Matthew Woo: Well, is there more that summer health should be doing? Is it really just simplifying access to health care? Um, or do we have a bigger mission in mind? And not surprisingly, we raised the stakes. No longer is a mission just to simplify the way in which people access health care, but it's really, really to help raise the healthiest generation.

[00:06:01] Matthew Woo: And that is the mission that led us to raise our Series A to really add fuel to what we've built and also tackle new segments of the population. Not just people that can afford our subscription model, which allows for unlimited sessions, but actually starting to work with both commercial and Medicaid plans.

[00:06:19] Matthew Woo: For people that need it and so that's the next chapter of summer health over the next year. 

[00:06:24] Angela Suthrave: I love your entrepreneurial spirit and, uh, as a mom of two boys, I definitely, uh, relate to those late nights when you need help. Um, so I can definitely see a compelling problem that you all are trying to solve.

[00:06:38] Angela Suthrave: One of the things that Omar and I were most drawn to about summer health is your use of A. I. So what we'd like to do is cover some A. I. Basics and then we can go into how summer health uses A. I. So if you're talking to someone like your grandmother, how would you explain A. I. In the simplest terms, 

[00:06:58] Matthew Woo: I would say that A.

[00:07:00] Matthew Woo: I. Being artificial intelligence is the idea that we can use Machines or I think my grandma would know what a computer is to basically simulate the intelligence and thinking and reasoning that a human has. And so that would probably be the most simple way to describe it. And artificial intelligence is in fact, not a new thing.

[00:07:27] Matthew Woo: You can think of the most basic. form of artificial intelligence is writing a set of procedures and decision trees inside of a book that maybe an electrician or someone else is using to diagnose a particular situation. It was actually an idea introduced all the way back in the 1950s. And when my dad was, you know, an entrepreneur, he was basically trying to encode expert opinions on mining operations to create expert systems that was effectively artificial intelligence in that age.

[00:08:00] Matthew Woo: And so it's really been around for a long time. It's just that each phase of both the boom and the winters, there's always been limitations on what artificial intelligence could do. Um, but I think we're obviously very clearly with the rise of chat JVT and transformers, an interesting turning point of what artificial intelligence is and what You need both, uh, in terms of data capabilities to actually put it into practice.

[00:08:30] Matthew Woo: And so, you know, that's a lot of why Summer Health has been so early in embracing AI. 

[00:08:35] Omar Mousa: Matt, I'm sure you get this a ton and I hope you don't, you don't mind me being so direct, but I often feel like in exec meetings or even like exposure to border investors, I hear like, People describe AI in the same way how you would describe software.

[00:08:52] Omar Mousa: Like AI just does everything and nothing at the same time. And so I know there are very specific use cases and there are different types of ais. So could you, you know, contextualize for us, like what are some different types of ais and then, you know, how is that relevant to healthcare? 

[00:09:07] Matthew Woo: I would say that AI is probably like the broader term or, and maybe the systematics people might argue the inverse.

[00:09:18] Matthew Woo: But within AI, there is one branch, which is the, you know, what people have traditionally thought of as machine learning and machine learning is basically this idea that you can capture enough information, um, both in terms of the inputs and the outcomes and basically create a model that can help predict certain outcomes.

[00:09:37] Matthew Woo: And it was mainly focused on, you know, the realm of. Classification. Um, and maybe ranking. Um, and so these are things that were used to do things like helping radiologists help, you know, identify if there's, uh, you know, issues with with a particular scan or x ray or, you know, Facebook in particular machine learning for ranking.

[00:10:00] Matthew Woo: What was always challenging was that form of AI when it comes to machine learning, you know, it wasn't generalizable. It was very narrow in the types of use cases, and it required a huge investment in terms of the data that you would need to capture, clean, process. You would need really sophisticated teams that really understood the statistics and mathematics behind how to fine tune these algorithms.

[00:10:26] Matthew Woo: Um, and so, you know, for example, if you look on Facebook, a lot of their machine learning engineers have PhDs. And so the, you know, capital investment for you to make a justification to invest in the I was so high in the past and what has transformed recently is this idea and it came with the transformer, which is a paper that Google wrote a while back and at the beginning was actually not very good.

[00:10:52] Matthew Woo: at what it can do now, which is generative AI, until it amasses a certain amount of data. And obviously the breaking moment was in 2021, when OpenAI introduced Chachabit and people really saw the magic that here is this embedded model that you could just ask questions and it could generate at that time, just logical, you know, outputs that people got really excited about.

[00:11:14] Matthew Woo: We can obviously talk about where it's gone from then, but that really opened up people's eyes and lowered the amount of investment, both in terms of data, capital, and skill, um, and really opening it up for many others and companies to basically leverage AI. 

[00:11:30] Angela Suthrave: We hear about A. I. And people have concerns about hallucinations and data privacy and other things.

[00:11:37] Angela Suthrave: How do you think about key challenges and considerations when you think about operationalizing A. I. In a health care setting? 

[00:11:45] Matthew Woo: Yeah. So I think that there is a couple of different areas that are maybe not unique, um, but definitely important to take into consideration in the setting of health care. One is the safety of where AI is being deployed.

[00:12:05] Matthew Woo: I think right now we are still, you know, not at the benchmarks necessary where AI can truly, you know, be responsible for, Uh, clinical decisions, especially in settings that are not as clear cut as in radiology, where, you know, there are some studies that have proven that it can be as accurate or in some cases more accurate than radiologists.

[00:12:25] Matthew Woo: That being said, when you're talking about a primary care relationship and basically having information that is being interpreted. You know, a lot of that still relies to have a human in the loop. And so, you know, judging the safety in terms of, you know, how much potential harm could it cause to a patient if it hallucinated is it's something that we definitely take a lot of talk a little bit more, but how some health we handle that.

[00:12:50] Matthew Woo: I think the second thing is we've seen what happens when data is misused. Um, from a social media perspective, and so obviously health care, that's even more important because it's related to people's lives. And so making sure that, you know, that the vendors and partners that you work with actually handle the data in a way that protects the privacy.

[00:13:13] Matthew Woo: Um, and interest of your patient is incredibly important. And then I think the third thing to realize is that the use of AI has different levels of risk depending on who the audience is. So whether that be the clinician, um, a support person or a patient, and these are factors that you should be weighing as you think about your use of AI.

[00:13:36] Omar Mousa: Matthew, thank you. I think. You know, all that information gives our audience hopefully a basic understanding of A. I. And generally how it's used. Let's get a little bit more into summer health and the application of A. I. Within summer health. So how is A. I. Being integrated into summer health platform and what specific problems are being addressed?

[00:13:55] Omar Mousa: For clinicians and patients through AI, I 

[00:13:58] Matthew Woo: think that, you know, at summer health, we really kind of, as we shared, the first thing that we did, um, and we kind of talked about is, you know, the handling of data. And so we didn't go and rush out to use whatever, uh, foundational models that were out there at that time, mainly being.

[00:14:19] Matthew Woo: Googles and OpenAIs until we were sure that they actually had the right data policies in place. And so it was when OpenAI introduced their BEA, and what that really meant was it had zero data retention. So the minute that you hit it with a particular prompt or query, it would drop the data instead of including it in the training data.

[00:14:39] Matthew Woo: Um, from there, we evaluated, well, again, kind of using the framework that I just shared, what is the level of risk to a patient, and where is their low risk but potentially high impact in terms of the efficiency. And from there, we, uh, identified a variety of provider use cases in which we could help improve both, uh, efficiency in terms of providers being able to get through the administrative workflow, but also in their quality of care, and we can talk a little bit more about that.

[00:15:06] Matthew Woo: Thank you. I think what's kind of amazing, though, was just how quick it was to get started. And I think that some people in the past have thought that, you know, you need to hire an AI engineer, you need to get all your data in a row. But I would bet that every startup probably has one or two team members already tinkering with AI.

[00:15:30] Matthew Woo: And truthfully, what I did was I spoke with our, uh, hopefully this is kosher to share, but I, uh, spoke with Ellen. I'm like, Hey, can you give me your medical records? And are you comfortable if I share it with the foundation model, even though I don't have a BA? And let me just test. What I could produce, right?

[00:15:48] Matthew Woo: Could I actually do a proper, you know, medical note? Could I actually, you know, help identify some CPD codes? Could I actually, you know, come up with a care plan and a reminder, um, for you? And the answer was more or less, yes, obviously varying degrees of accuracy. Um, and that gave us a lot of confidence to invest more in it.

[00:16:09] Matthew Woo: And that was a simple. You know, two, three hour investigation with the permission, obviously, of someone to share the information with, uh, these, these models, 

[00:16:18] Omar Mousa: I had a random follow up. You mentioned the B. A. A. And kind of the, uh, no data retention. It sort of drops a query. I'm wondering from a health care perspective, like the model trains on the information and effectively gets better and supporting the use case.

[00:16:35] Omar Mousa: Given that it's not retaining the information, what sort of expectation around the model improving can we expect? Or like, is that even a concern for you all, um, given the business problem you're solving? 

[00:16:47] Matthew Woo: Yeah, I think that we talked a little bit about how, you know, similar health is approached differently, but we believe the foundational models are just going to get better.

[00:16:57] Matthew Woo: And that's clearly not a space where we have the data, the infrastructure and the expertise to continue to improve the foundation models. The fact is, you know, since November of 2021, we've seen new models like Anthropic. Come out, Nistral, um, all of these different players. And we just continue to believe that it's going to get better.

[00:17:16] Matthew Woo: And basically, not once they commoditize, each of them are going to have their own thing, but there's only going to be a few players left at the end where the, um, I believe the magic happens is your ability to actually rapidly, uh, Iterate and apply these models in the way we're actually capturing information about the output and whether it is used and the quality of it in the context window that is going to really help and a lot of that infrastructure to capture, you know, if you were the draft message that doctor is going to send, did they use the message?

[00:17:53] Matthew Woo: How did they change it? Um, did the, uh, patient actually like the message? Did they, um, respond to it? Did they not respond to it? All that information is information that we capture on our side of the house so that we can continue to fine tune the examples of prompts. Um, and teach the model to basically fine tune its style to each provider to optimize for a particular outcome.

[00:18:18] Matthew Woo: So that's kind of where we think the magic is happening, um, in terms of the infrastructure we built. And then the second thing I'll say is that since the time of, you know, when GPT first came onto the scene, the context window was, I don't know, 10, 000 tokens. That's maybe 30 pages. Google has announced a million token size window is probably going to go to 10 million.

[00:18:42] Matthew Woo: That's like 3000 pages of information that you can ground. The model with and I think that's where we're obviously seeing a lot of, uh, innovation in terms of how you architect that rack structure to make it most contextual and funny enough, it's now pulling in some of the more traditional machine learning tactics to basically be able to retrieve the right set of information.

[00:19:02] Matthew Woo: What they call documents and in the infrastructure, but you know, information into the context window that can help ground whatever prompt that you are using. 

[00:19:10] Angela Suthrave: I love the way that you describe the building of the infrastructure. Can you help us to understand? So summer health does 24 7 chat based support by providers to parents?

[00:19:22] Angela Suthrave: How do you deploy the AI model to help with that workflow? 

[00:19:25] Matthew Woo: Yeah, so I'll say that where we started out was nothing real time. It was just, Hey, here's a chat that has happened with the patient. Um, can you draft a medical note? And the reason why we focus on that first one is pretty low risk to, you know, we were like, Hey, you know, a bridge is doing this.

[00:19:44] Matthew Woo: Everyone's doing this. I'm curious. Like, how difficult is it really to do this? Um, and we did it during a hackathon and whenever engineers basically put it together in 24 hours, I'm like, wow, this is really good. And he actually worked with a clinician whenever I had a clinical to really find to the prompt, provide examples, and we were so compelled by it that a week later, we actually launched it into production.

[00:20:07] Matthew Woo: And so. What happened was we saw that the time it took for a patient to get their medical note and their recommended care plan so that they would know what to do for the next couple of days went from a turnaround of 48 hours to less than 15 minutes. Um, so it's almost instant in that they would actually get their medical note.

[00:20:27] Matthew Woo: And I think that's really important and for a parent to feel cared for. Since then, we've started to, uh, investigate and understand, um, how can we start to integrate AI in more places. And so the first thing that we actually had to build was a platform that allowed our engineers to, And clinicians to start coming up with more applications.

[00:20:45] Matthew Woo: So the next one is also pretty low risk is hey, you're talking to a doctor. How many times have you talked to a doctor and realized you had to repeat yourself or maybe you're talking to a specialist and you had to repeat the past medical history? Um, we have that problem because we have specialists. So where we use AI then is, well, when the specialist is talking to a patient that they haven't seen before, we can summarize the past three visits for them to get, so that they can have a quick understanding of what's been happening, and they would be able to potentially call and say, Hey, maybe the reason why they're not sleeping well is because they had this rash.

[00:21:18] Matthew Woo: Um, and so we can really improve the continuity of care. And so again, it's not, necessarily replacing a doctor, but it's definitely giving them much more important context for them to do it. And then the last thing that we've done is, you know, maybe this is what every healthcare tech company ends up doing is we built our own EHR.

[00:21:37] Matthew Woo: And that's really powerful because it allows us to rapidly deploy these new AI use cases in the right places and actually capture the signal of Hey, did they like that? Uh, summary was not helpful. If it's not helpful, we capture all the bad examples. We analyze it, figure out what was wrong with it, and then we improve, uh, the prompts and the context that we bring into the prompt on by Iraq.

[00:21:59] Matthew Woo: And so I think this is. You know, what's unique about the way in which summer health is thinking about it. 

[00:22:04] Angela Suthrave: There is uh, this running theme on our podcast where it feels like every single episode Um people give a love note to med plum. So I think that you all are on med plum, right? 

[00:22:15] Matthew Woo: Yeah Yeah, net plum is great Rushman and their team are great.

[00:22:20] Matthew Woo: I try to When we are looking for new offices, I try to get them to move in with us so that they can help build out some of the features we've been requesting. They, they politely declined, but no, I really love that team. And everything we've built is on fire, uh, including the data that we pull into our prompts.

[00:22:36] Matthew Woo: And so it's. Uh, pretty incredible because that means that we'll be interruptible, especially as we move into this world where we're starting to be more integrated to health systems and requesting medical records and actually pushing information back. 

[00:22:47] Angela Suthrave: Yeah, they're on fire. Um, no pun intended. Matthew, you touched on something really important, which is that you had a hackathon and you were able to figure out that you could build the AI model yourselves.

[00:22:59] Angela Suthrave: I think a lot of healthcare delivery organizations are at that buy versus sell. Um, build decision node. And so was curious if you could talk through that. Um, Decision making process. Obviously, summer health has invested heavily in, uh, these types of engineers. Um, and so was that always part of the strategies that you would build it yourself?

[00:23:26] Angela Suthrave: Or have you considered, um, buying a I models? 

[00:23:31] Matthew Woo: Yeah, I think that it really, I think the first thing to say is that you can't not be investing in this space, right? It's just, it's, it's inevitable. Um, it's no longer a decision. It's just something that you have to do. The second question then is where in the stack do you actually invest?

[00:23:50] Matthew Woo: Your time. Um, as I mentioned, we're not going to be investing in the foundational model. That's not where we think our unique, uh, capabilities allows us to leverage the full case of AI. Um, and then the second, the third thing that we consider is, well, how core is AI to what we do? Um, and how are we planning to deploy it?

[00:24:10] Matthew Woo: Is it something that we want breadth, uh, or, or depth and breadth and being that like, it's going to be used by a lot of different people. Um, Or maybe maybe it's like the people hasn't been deployed to do we actually have any, um, you know, tight feedback loops with them. And so for us, the answer was, you know, one is core to what we do.

[00:24:31] Matthew Woo: We want to continue to be the fastest and ideally become the most affordable way in which people can access care. Um, and three, we have, we're vertically integrated. You know, we recruit our own providers. We have built our own EHR, we can get feedback from them. And so when we look with all these AI tools out there for healthcare, whether it be, you know, Bridge or Nabla or, um, some of these other ones, it just didn't make sense for us to use those services and be limited in terms of the application, especially when we needed to get really smart and apply a thousand use cases across our EHR.

[00:25:07] Matthew Woo: music ends Um, that being said, one area where we probably make better buying is probably some AI that's associated with the billing infrastructure, because that's less important to us, right? That's not where we're going to make a differentiation. All we want is to be able to reliably get putting claims and get reimbursement.

[00:25:25] Matthew Woo: So I think that's probably an area. That we would maybe look for a vendor. That being said, we have a very unique model. It's chat based, uh, we had to create CPT codes based on chat conversations. And there might not be something that exists, so we might be creating something entirely new for that reason.

[00:25:38] Matthew Woo: And so we've been exploring that as well in terms of. You know, can we leverage our platform to actually reliably, uh, code our conversations for claims? 

[00:25:46] Omar Mousa: I like the energy you have when you describe how you're solving some of the problems. I can tell you really appreciate the building of the actual, like, software and, Platform itself.

[00:25:59] Omar Mousa: Um, and you're head of products, obviously makes sense. Um, a bit of a bias. Yeah. Yeah. And so, uh, you know, you talked a little bit about early stages of leveraging AI. What strategies have you found to be most effective in scaling AI, um, across the organization and as you've grown, 

[00:26:19] Matthew Woo: I think that there are two areas I would talk about scale.

[00:26:24] Matthew Woo: Um, one is just, You know, for the tools that you do build, how quickly and how accessible can you make it to the rest of the team? Who on your team can actually contribute in terms of continuing to push new applications and use cases based on the foundational model that we put in? We can talk a little bit about what we call Carrier AI and a little bit about how it works.

[00:26:44] Matthew Woo: But the second thing that's really important is that if you're constantly, you know, coming up with new use cases, fine tuning based on the outcomes of whether people are. You know, found the outputs valuable or not, you're going to have to invest into, uh, basically evaluation tools. One, for example, like has come up that we, we haven't decided to use that because, uh, we, I mean, maybe it's my engineering team's fault.

[00:27:09] Matthew Woo: They'd love to build their own things is autoblocks, um, AI. And so what they do is that when you come up with a new use case, you can ship it. They'll help you. They'll help you track. You know what? A variety of models would come out with the output. You can actually have a human on your team that can evaluate whether that was a good or bad response.

[00:27:28] Matthew Woo: And that way you can run statistical tests to see like, hey, if you were to run this. Uh, new prompt or deploy this new prompt, is it actually going to be better than the previous one based on the context that you added into the window? And so I think this is really helpful. And then there's also a lot of open source, um, frameworks out there.

[00:27:44] Matthew Woo: I can follow up with you actually, I should have pulled that up. Um, where you can basically just run like basic reasoning tests, grammar, tone, and the like. And this is really helpful to understand, hey, what are the, Implications of these problems because offensive effectively, it's like deploying code is going to have a direct impact on the actual users because users are seeing the outputs of these, these models, 

[00:28:06] Angela Suthrave: Matthew, you've talked a lot about the user feedback, which I love.

[00:28:12] Angela Suthrave: Can you talk a little bit more about what metrics are you looking at to drive improvement? And how are you really leveraging your clinicians and other types of users to continue to improve these models? 

[00:28:25] Matthew Woo: I think right now. Truthfully, we want to eventually move towards getting direct feedback from patients, um, as we have more use cases that directly interface with them, but many of it, um, truthfully, right now it happens with clinicians.

[00:28:36] Matthew Woo: And so every time we deploy a new use case, we'll first deploy, we have over 40 providers on our platform now, um, we'll deploy it to, Five to 10, depending on, you know, what the use case is. And so we'll actually just get feedback from them. We'll be like, Hey, was this drafted follow up message that we have created for you?

[00:28:55] Matthew Woo: Was it helpful or not? And it'll be qualitative. We have a Slack channel. They'll give us a feedback. They'd be like, Hey, I actually never like capital case, the names, or I don't use the last name, things like that. Very, um, you know, explicit feedback, but I think what's more important is to actually capture the implicit feedback.

[00:29:11] Matthew Woo: And so what we'll do is we'll track on our. Uh, backside. You know, how often are they using the output directly? If they were to adjust it, what percentage of the message was adjusted? Um, three, how does it compare in terms of the time they took to open that chat bubble to send a message? Did it actually improve the, uh, efficiency in time?

[00:29:34] Matthew Woo: And most importantly, you know, I think we've mentioned this past, like, did it actually lead to a higher engagement with the patient in which the patient felt care for? Because we actually capture CSAT as well after every conversation. So we understand, did the patient actually find this conversation helpful?

[00:29:49] Matthew Woo: And we use all these signals. Um, and I wouldn't say it's automated. We have a data scientist that will basically look at this, um, On a monthly basis and help us give us feedback to then improve and iterate on this problem. I think eventually it should be something that's live and automated. Um, but, you know, in the earlier stages, we really want to be intentional about the changes that we're making.

[00:30:07] Matthew Woo: Um, let me say, you can imagine that, um, not only are we changing the prompts at a per provider level, but it can also change based on the time of day. And one kind of funny thing that we learned was that providers are more likely to send what's generated via the, um, AI, um, make less changes when it's late at night.

[00:30:29] Matthew Woo: Not surprisingly, that's actually not a behavior we want because that's probably because the doctor is just really tired. They just want to send the message and go. So that's a behavior that we had to correct, um, and something that we wouldn't have heard via explicit feedback. They probably loved it. But implicitly, we knew that.

[00:30:43] Matthew Woo: It wasn't necessarily, they were sending it too fast. They had no, there's no possibility that they were able to read and decide whether this is the right message to send. And so those are things where we had to turn it off in those, those cases. And so these are things where we're constantly iterating and learning and making sure that patient, um, safety in terms of what's being sent is, is still very much top of mind, even with the human in the loop.

[00:31:04] Omar Mousa: Matthew. One thing I appreciate about you, like as you've been talking on the show is Uh, patient safety has been pretty much top, you know, I have heard it a couple of times come out of your mouth and how summer health is leveraging or like thinking about that. So I want to ask you, uh, let's, let's dive into AI ethics and patient safety a bit.

[00:31:24] Omar Mousa: So what, what are some ethical guidelines that summer health follows and when developing technology? 

[00:31:30] Matthew Woo: Yeah. I think that the most important thing for me is truly like when they're. Safety in terms of, you know, are we misrepresenting the care that they're getting that they're, they're, um, in some, frankly, paying for, um, and also, you know, is the advice going to cause adverse harm in the same way that, you know, we track things like prescription rates, because we want to make sure that our doctors not over prescribing, especially in the field of pediatrics.

[00:32:02] Matthew Woo: And then the second thing that is really important for us is privacy, is the information being used in a way that the patients are not aware of. And, you know, as we develop these technologies right now, much of it is on the clinician side and we're not training this data for, uh, any external purposes and whatnot.

[00:32:25] Matthew Woo: And so, you know, we have been very careful with that, but, you know, in the future, if we were to be more explicit and be like, Hey, this, Um, response was actually generated by an AI, and there was no edit, you know, that's something that we want to make sure that the patient knows. Um, so, and it was not reviewed by a human.

[00:32:43] Matthew Woo: I think that these are definitely, you know, tricky areas because, you know, how much is a human really going to trust an AI? And I think that that's something that we need to prove over time. Um, although there, I think there's different frameworks of how we eventually get to the point where people feel comfortable with it.

[00:32:56] Matthew Woo: And then same thing with privacy. If we're going to be using their data in a way that's going to be trusted. Not just benefiting them, but maybe benefits benefiting all of our users, or maybe it's something that we're using to train some diagnostics. I think that's important to make sure that they're aware that the data are being is being used in that way.

[00:33:12] Matthew Woo: I would say right now, um, that's not something that we're doing. And so we're not explicitly asking, but at some point, you know, it's definitely an opt in. I think everyone should have the ability to make that decision for 

[00:33:22] Angela Suthrave: themselves. It sounds like, um, there's tons of exciting problems for you all to solve.

[00:33:26] Angela Suthrave: Where do you see the future of AI? In healthcare heading and what types of innovations are you the most excited about? 

[00:33:35] Matthew Woo: I think right now where healthcare AI is being deployed and, you know, frankly, it's where we started out is a lot of the low risk, high effort, administrative workflows that. Doctors have had to deal with.

[00:33:50] Matthew Woo: In fact, my wife is a dermatologist, so I see this first hand when she comes home and she's still writing medical notes. I think that's great. It's going to allow them to have more time with patients. But I think where I'm really excited for is when AI is truly a co pilot with doctors, and that's something that we're starting to explore with already with us, especially a chat interface is very conducive to having a doctor.

[00:34:17] Matthew Woo: Have an AI that's going to, you know, know the entire medical history, make sure if a doctor is asking questions that isn't answered already because we have that information. Um, and so this is something that we're exploring with, but I'm really excited for that to happen, not just in a telemedicine platform, but also in real life.

[00:34:33] Matthew Woo: On the patient side, I think it's been helpful, obviously, to be able to ask AI for questions on particular topics and get answers, but I think it's going to be really exciting when AI can truly do that. Provide, you know, a deeply personal care plan based on your particular goals. And I think that it requires a little bit more than just a prompt.

[00:34:58] Matthew Woo: It requires your ability to actually index every, all the information from a medical record, be able to index, you know, what are some of the studies that have gone out there? Um, and ideally, you know, in the future, if there's a large enough sample size, I understand, Hey, for a parent that comes from a, maybe a particular ethnic Background or maybe they had a premature early birth.

[00:35:18] Matthew Woo: Like these are the things that you should take in consideration that just doesn't happen right now for physicians. Cause they don't have the time to really think through all of that. And I think the second thing is, you know, AI shouldn't just be giving you answers. They actually should help you do things on your behalf.

[00:35:35] Matthew Woo: I spoke with a patient in Medicaid and she spends. Like hours every month trying to get an appointment. And every time she gets into the front desk, you know, they're very dismissive and rude to her. Um, and so she wants to get to that point, but she can't. And she's a working mom and she works from 7 AM to 7 PM.

[00:35:53] Matthew Woo: And she's taking time out of her day. Like. Why can't I help her book an appointment? I think there's use cases that are not necessarily health care based in terms of it's giving you medical advice or assistance, but it's actually helping you get things done in the real world. And I'm also, I'm really excited for that as well.

[00:36:10] Angela Suthrave: That's great. I'm hearing you say, maybe address some social barriers that people encounter that are very real barriers to health care. I 

[00:36:18] Matthew Woo: mean, some of these, Patients have to remember if they want to get to the well child visit, they need transportation. You need to fill out these forms three days in advance.

[00:36:27] Matthew Woo: And you have all this information, uh, on you at any given time. Like, why can't we just automate all of that so that you can just show up there? And just kind of like how you go through TSA pre, you know, so these are just things that I think people think AI is just this bot that gives you answers. No, AI can do things in the physical world.

[00:36:45] Matthew Woo: We just have to enable it. 

[00:36:47] Omar Mousa: Matthew, you've been, I've seen you doing the circuit. I've seen you talking about AI and multiple different venues. Obviously, clearly you guys are leveraging it in many different ways, solving a lot of use cases. Your hiring has signaled that you guys are leveraging. Yeah, so it only makes sense to talk to you about it.

[00:37:02] Omar Mousa: So, uh, it's, it's very clear you have a culture, like a very strong culture around AI. So I want to talk to you about building kind of an AI first culture. First off, what just inspired you guys to adopt this AI first approach at SummerHealth? And like, what are the tactical things you guys did to foster, um, that culture, uh, to, to really bring it to life?

[00:37:21] Omar Mousa: Uh, to life. 

[00:37:23] Matthew Woo: I think the reason why we were so fast to move towards this opportunity was when we first started summer health. I had this hypothesis I had. Hey, we're going to deliver care by messaging is going to be incredibly affordable. We're going to have mass amount of volume. Um, and we would have all this unique data because it's not like some audio or like in person interaction with the doctor.

[00:37:48] Matthew Woo: You're not capturing any of that. Um, and we could eventually build this medical co pilot. Um, I literally called it that, you know, two and a half years ago and before GPT. But I thought it was a 10 year vision. You know, I thought that was, it was so far into the future and we just had to get the basics right.

[00:38:06] Matthew Woo: And on November, uh, you know, I think it was, is it 21st that it launched? I think roughly that was the day that chat GPT launched. And I saw what it was capable. I'm like, wow, the future is here. Like we can literally do this now. We don't have to wait. And so I have a few friends that worked at open AI and immediately called them like, Hey, when are you guys going to have a BA?

[00:38:27] Matthew Woo: Like I need this technology. Um, and so as soon as. January. Uh, they were starting to work with a few health care companies, typically larger ones. Um, but they said, Hey, you know, we would interested to hear kind of what you're planning to use it for and whatnot. And that's how we got that early BA. I think we're probably the first health care company start health care startups to have a BA with open AI.

[00:38:52] Matthew Woo: And this is before they were charging like this was it was free, which is great because you got to play around with it. Yeah. The reason why I share that, it's not a playbook that you can roll out with other companies, but I do think it points to a few tactical things that people can do to kind of kick off, you know, an AI first culture.

[00:39:10] Matthew Woo: And let me first define what an AI first culture means to us, which is a bias towards constantly asking for every decision that you make on a daily basis, can AI replace, augment, or teach me something about what I'm trying to accomplish today? And the reason why it encompasses those three things is that it's not that AI is just something that we deploy to our users.

[00:39:32] Matthew Woo: AI should fundamentally change the way in which and how we build and therefore what we build. Um, I do think that tools shape us and, and therefore we, you know, we kind of need to embody that. So tactically, the reason why I share that story, uh, first off is that there's already someone in your company that loves AI.

[00:39:52] Matthew Woo: They're probably automating like 50 percent of their job. And you just have to put out the feelers. These are the people that are posting your Slack, the, the, the tweet storms and tweet threads about the latest AI and how it's benchmarking. Like this is a person that you want. Second is, you know, get a quick win in terms of being able to prove the value.

[00:40:12] Matthew Woo: And I shared a story already earlier that I literally just asked one of our, my co founder, you can ask someone on the team, like, Hey, can I just share, I know it's sensitive information, but can I share your information? With these models and just see what's possible. Um, I think people are going to be blown away that within a couple of hours, you were able to accomplish something that they didn't think was possible.

[00:40:32] Matthew Woo: That really gives you a strong proof point. Um, and then the third thing is I do think a hackathon is a great way to get people aware of it. And so ahead of time, you know, often show them the quick proof point that you have, you know, put together some YouTube videos of explaining kind of how it's been used in other places.

[00:40:51] Matthew Woo: That's And, um, and some rough kind of themes and, and have the team kind of work and hack on a few of these, these models, give them some mock data. There's a lot of public, um, anonymized healthcare data out there that you can use. And I'm sure coming out of that, there's going to be one or two ideas where you can be, where you're confident, like, Hey, tomorrow we can go, uh, launch that.

[00:41:15] Matthew Woo: And so I think that that's a really great way to, you know, get a spark, then a flame, and then eventually something that just stays within your company. Um, and at least that's how I've done it. And I've heard a few other companies had a very similar story, um, in terms of how they got AI really going at the company.

[00:41:34] Angela Suthrave: Matthew, I think that Summer Health has a lot of really top talent and it's probably an area that you invest heavily into. How do you think about talent and what skill sets are you looking for as you continue down this AI forward path? 

[00:41:52] Matthew Woo: Truthfully, You know, kind of as we mentioned in the past, it was a strategic decision whether you should invest in the eye.

[00:41:58] Matthew Woo: You needed to have one enough data. You need to have infrastructure. You probably need to have a data engineer. Then you would need to have a machine learning engineer to get going. I think that's not the case in this world. In fact, you know, where we've been spending a lot of time talking about AI is at the application level.

[00:42:16] Matthew Woo: And so really where what you need is Product engineers. And I would hope most companies have product engineers that, you know, are obsessed with building great software. I think it's, you know, using AI is going to be the same thing as using Figma. There's really no, there's no way that you can build software without using Figma or that without using AI.

[00:42:33] Matthew Woo: And so I don't think it's a unique skillset that you need to acquire. That being said, you know, as you do scale, I do think it's important that there is some AI infrastructure that you want to build so that you can properly accelerate the deployment of different AI use cases and also do the evaluations as we mentioned.

[00:42:51] Matthew Woo: And yes, maybe if you're big enough, you know, you might want to bring back some of that traditional machine learning engineers or people that really are at the forefront. But the reason why I answer in this way is that there shouldn't be an excuse not to get started today. Uh, you don't need to bring in someone to teach you.

[00:43:06] Matthew Woo: It's something that you should be able to just get going. 

[00:43:09] Omar Mousa: I love it. I think I hear a lot of reasons why not to. And for you to say, if you have application engineers or product engineers, like you should be able to deploy something, let's so we've hit a little bit. You talked about, like, building the culture and the tactical things around that.

[00:43:24] Omar Mousa: And I think that's incredible advice. What advice would you give to other healthcare startups looking to integrate AI into their solutions? 

[00:43:31] Matthew Woo: Yeah, I think first and foremost, Make sure that, you know, depending on what the use case is, make sure you're working with the right partners. You know, and make sure that you're actually following the compliance standards necessary to protect your patient's information.

[00:43:45] Matthew Woo: At the end of the day, like that, if without that, you have none of your patient trust and therefore you're not actually delivering quality care and protecting your patients. So first, make sure that you're working with the right partners. Make sure to sign those BAs. I know they can be expensive, but it's important.

[00:44:02] Matthew Woo: And then from there, the second thing is. I mean, it sounds pretty, pretty basic, but there's a lot of, I would start off honestly just to, you know, you know, get started, watch a lot of you, like, it sounds silly, but like literally just search YouTube tutorials on, on the use cases of AI there. Andrew Ning has a great one.

[00:44:23] Matthew Woo: Um, I watch a bunch of Lang Chang. Lang Chang is a little bit difficult to deploy at scale, but it's really great to get started and testing, like just watch all the videos and you start to get a sense of what's possible. And then, you know, assuming that you can get some of that under your belt, you know, there's a, I mean, every week, there's a new paper being released.

[00:44:41] Matthew Woo: And so fortunately, we have, uh, our head of data science here that's constantly sharing these papers because I can't read them all. But, you know, start getting smart about what works and what doesn't work. And funny enough, some of these papers are not like mathematics. Literally one paper is talking about it.

[00:44:55] Matthew Woo: Chain of reasoning, uh, where you put in basically what strawberry is, but like pre strawberry, which is literally put the instructions of what you want the AI to do and, and what's the best way to format that and achieve certain results. And so I think that people are really daunted by the aspects of getting AI started, but I think again, within this day and age and all the resources out there, you know, take baby steps, but once you get a few wins, start running at it.

[00:45:21] Angela Suthrave: The, you mentioned the strawberry model, which is. Um, open A. I. S. New model code name strawberry. Maybe a month ago now, these models, I feel like have just evolved, uh, so quickly. And how do you think about those advances in the accelerated pace of the development? 

[00:45:42] Matthew Woo: I'm excited for it. I think that A lot of what we were trying to do with all the chain of thought and cyber prompts is, uh, more or less some of what Strawberry, now, you know, Strawberry doesn't include all domains and it doesn't have access to all the sets of information, especially health information for that matter.

[00:46:00] Matthew Woo: So if it can do and have that reasoning already built in, then, you know, with our unique data set, I think we're even to get better results and better capabilities. And. I think that there's this fear of AI replacing jobs and I'm sure you know what would be a lie to say it won't, but at the same time it's going to make even more personal software and delivery experiences that just wasn't possible before and that's going to require even more effort.

[00:46:27] Matthew Woo: Um, and so I'm excited for that future. And my, my wife, who's a dermatologist, always says like, Hey, if AI can help identify rashes, great. Then I can really focus on the complex patients that really need my focus and intention. 

[00:46:39] Omar Mousa: I love it. I love the perspective. All

[00:46:49] Omar Mousa: right, Matthew, we're at our very exciting concept closing call. It's meant to be fun, fast paced. Angela and I are just going to rattle off some questions. Uh, first off, are there any frameworks, methods, or processes that you've found especially useful in your work? And that others who listen to this podcast might as well.

[00:47:05] Matthew Woo: I think being a startup, the most common framework I use is one from Bezos, Type 1, Type 2 decisions. Type 1 being decisions that are big and irreversible and Type 2 decisions that are And I would say 90 percent of the time your decisions are reversible. So just make it move on, learn and get, get going. I think too many people think their decisions are type one, but it's not.

[00:47:27] Matthew Woo: And so as a startup, our only advantage is speed. So let's just get going. 

[00:47:31] Angela Suthrave: Is there a tool that is highly valuable to you that you think others may not be using and should? 

[00:47:36] Matthew Woo: Yeah, I really like this tool called voice notes. Uh, it's kind of like a voice memo app. Um, you just record whatever thoughts you have, does a really good job at accurately transcribing it.

[00:47:47] Matthew Woo: And then it has some prebuilt actions to like summarize your thoughts into bullet points or write it into an email or clean it up. And so I find it as a great way for me to just get my thoughts out quickly and share it with the team. The other fun thing that I've been doing recently, I don't know if you've seen this trend on YouTube is people buying like E Ink Android devices.

[00:48:08] Matthew Woo: And so they're kind of like Like your smartphone, but it's all in E Ink, and so it's like less addictive. I only install reader apps and, you know, YouTube and things like that. And it's been really helpful for me to focus. So when I come home, I just put my phone away, like my iPhone, and I just use this E Ink Android device to do Duolingo and, you know, read Kindle and read my Read a Later app articles.

[00:48:36] Omar Mousa: E Ink, like E dash, like E, I, N, K, like, you know, like 

[00:48:41] Matthew Woo: Kindle. Yes, I hate Kindle. The brand is called Books, B O O X. Um, I've really enjoyed it. I know that people can't see it, but this is kind of what it looks like. This is like an E Ink. Oh, 

[00:48:52] Omar Mousa: sweet. 

[00:48:54] Matthew Woo: A new toy. 

[00:48:54] Angela Suthrave: It's like how people are going, uh, back to the flip phones.

[00:48:59] Angela Suthrave: Or just phone calls and text messages. So 

[00:49:01] Matthew Woo: yeah, I got my Duolingo. Sometimes I even do my Peloton. You know, the Voice Notes app is the one right next to the Audible app. And I can listen to audio. Oh, that's 

[00:49:13] Omar Mousa: incredible. That's so cool. I do. Angela, you mentioned flip phones. I miss BlackBerrys. I felt so important with BlackBerrys and it did nothing.

[00:49:24] Omar Mousa: It's just like email, BBM. Um, nobody's probably, nobody probably feels the same way I do. Uh, I promise I'm a millennial. Okay. Um, are there any concepts in healthcare that excite you, Matthew? 

[00:49:38] Matthew Woo: I think it's been really amazing to see how fast AI is being deployed at scale. Uh, I think I'm genuinely excited about that and, and, you know, people might be like, Oh, aren't you afraid they're going to eat your lunch?

[00:49:48] Matthew Woo: Maybe, but at the same time, I think it's net positive. I think more care is better care, uh, and in this day and age, I think that, you know, I'm excited about a world where if we can capture more longitudinal data, like we can actually provide more routine checks for, uh, people in a convenient and accessible and delightful way that just wasn't possible before.

[00:50:10] Matthew Woo: I think it's kind of crazy that. You know people only see their pediatricians like once a year and this pediatrician supposed to know everything about you I'm like, how would they know anything? They've only seen you Every 12 months. And so how do we have a better touch point, um, so that we can help children, especially in this day and age where mental health and behavioral health is, is such massive issues.

[00:50:31] Matthew Woo: So I'm excited about how AI can, can do that because in the past we just wasn't possible because even if you capture that information, you could do nothing with it. Um, but this in the same age, I think you can, you can create deeply personal experiences where people feel like they're not just giving you information, but they're actually getting value.

[00:50:46] Matthew Woo: Love 

[00:50:47] Angela Suthrave: that. Do you feel like product management is a science or an art? 

[00:50:52] Matthew Woo: I think that product management is very much, um, starts off with art in that is subjective. Um, it's understanding what users need and making an hypothesis. But once you make an hypothesis and you put something out in the world, it becomes a science in which you'll very quickly see, is it going to work or not going to work?

[00:51:12] Matthew Woo: Um, and you have to take that feedback and iterate and keep on going. It can't be one or the other. It kind of has to have a combination of both and unless you're Steve Jobs. 

[00:51:22] Omar Mousa: Last question. Where can people get in contact with you if they wanted to reach out? And do you have any shameless plugs? 

[00:51:27] Matthew Woo: Yeah. Same shameless plug.

[00:51:29] Matthew Woo: Number one, we are looking for a VP of engineering. Um, we just hired a team and we're really excited about our future. And so yeah, If you're interested, reach out to me at Matthew at SummerHealth. com. I've been really bad at posting on Twitter and LinkedIn, but that's one of my, you know, half year goals.

[00:51:49] Matthew Woo: And so I'm going to start doing that more regularly. So, you know, if you enjoyed this, just follow me on Twitter at Matthew Eden, E D A N, woo, uh, or LinkedIn. It's pretty easy to find me and yeah, try out Summer Health. I think I'll keep this code available for the, like the next, next month. Month or two months.

[00:52:07] Matthew Woo: And so if you sign up, use a code summertime and then try it out. Let me know what you think, 

[00:52:11] Omar Mousa: Matthew. Thank you so much for coming on the podcast and, uh, really appreciate it. It's been fun. 

[00:52:16] Matthew Woo: Yeah, no, thanks for having me, Angela and Omar. It's a, it was fun to chat with you guys. 

[00:52:21] Omar Mousa: Awesome. 

[00:52:21] Matthew Woo: Thanks, Matthew.

[00:52:25] Omar Mousa: Hey. Thanks so much for listening to the show. If you liked this episode, don't forget to leave us a rating and a review on your podcast app of choice, and make sure to click the follow button. So you never miss a new episode. This episode was produced and edited by Marvin Yue with research help from Aditya Triyab.

[00:52:42]Omar Mousa: We're Angel and Omar, and you've been listening to concept to care.

 

Recent Episodes

Omar Mousa
Previous

Episode #15 - Ian Chiang, Partner at Flare Capital Partners
Next

Episode #13 - Ari Saft, VP of Product at Tia