<aside> 💡 In 2022, I moved to the US to improve the healthcare system. Today, I’m building Superpower.com.

But it wasn’t a linear path here.

This document shares my learnings from the journey, testing several key hypotheses, and throughout the process, investing in 11 healthcare companies, chatting with 138 healthcare operators, looking into 362 healthcare companies, building several health apps, and even working in a doctor’s clinic.

This document was written in July 2023. It’s now July 2024. I disagree with around 50% of what I wrote given the knowledge I’ve gained since.

More about me 👉 maxmarchione.com

</aside>

<aside> 💡 My core premises

In June 2022, I set out to fix the problems I had experienced in healthcare, wanting to dedicate the next 50 years of my life to these problems. This resulted in several premises:

1. I must build something I want for myself and my loved ones. My personal mission is to build beautiful things that change the world.
2. The company must have an accumulating advantage with scale. Growing at 200% over 5 years is a 2,000x smaller outcome than growing at 30% for 50 years.
3. The company must have a logical path to $100 billion, even if the odds of getting there are low. Valuation is a barometer for utility impact. </aside>

<aside> 💡 Five stages of hypothesis testing to decide what to work on

• Pre-work: understand the market
• Test Hypothesis 1
• Test Hypothesis 2
• Test Hypothesis 3
• Test Hypothesis 4 </aside>

Pre-work

<aside> 💡 I made it my job to know more about healthtech than industry insiders.

I’ve always been obsessed with wellness, reading papers in my spare time, getting a CGM and EEG at age 16 for fun to run ML models on my data, compounding my own supplements at home, interpreting my own blood tests, and helping my family when doctors didn’t give them answers. But I’m not a doctor.

The outsider’s perspective is a blessing and a curse.

On the positive side, I come at things from a technologist’s mindset. I’ve spent my time exploring tech, capital allocation, wellness, and design. I haven’t internalised the dogma of the industry, which makes it easier to think from first principles.

On the negative side, I have less clinical intuition and industry knowledge.

So, I made it my job to learn more about healthtech than just about anyone else. I worked at a doctor’s clinic, looked into 362 healthcare companies, and spoke to 138 healthtech operators.

</aside>

Here’s part of my Notion.

Hypothesis 1: Expertise is a scarce resource. Primary care can be so much better. There’s latent demand for more aspirational healthcare. For the first time, AI makes it possible to create ‘Peter Attia in your pocket’

<aside> 💡 Key takeaways

• AI diagnosis is not the answer [true in June 2022; now it’s different]
• Apps do a better job at productizing healthcare than general-purpose AIs
• Behaviour change is everything
• Reimbursement matters more than distribution which matters more than product
• Convenience sells more than quality

[Reading this back, I disagree with every single one of them, 2 years later. These were the ‘expert’ opinions rather than the first principles truths. Be wary of experts]

</aside>

<aside> ❓ Core hypothesis

Expertise is the single scarcest resource in healthcare; the world’s wealthiest people monopolize access to ‘10x’ doctors. These are doctors who are exceptional at what they do, akin to 10x engineers. I asked, “What would it take for the whole world to have the same quality of care as Jeff Bezos?”

</aside>

<aside> 🛠 What I did & learned

I tested three key things:

• Technical feasibility: Was it technically feasible to scale 10x doctors?
• Adoption: Did ‘quality’ sell to patients?
• Reimbursement: What was the right reimbursement mechanism for ‘quality’?

Technical feasibility

I went through three stages of working out how to ‘codify’ 10x doctors.

• Digital twin (v1) as the data architecture for personalized medicine
  • I explored building a digital twin of the human body that could detect early disease onset and provide patients with precision recommendations
  • Digital Twin prelim research [SHARED]
• Clinical decision support systems are a better way to technologize care than a digital twin
  • I explored statistical and algorithmic clinical decision support systems. I discovered a way to use a combination of the two
  • Clinical Decision Support Systems / Expert systems / CDSS / DDSS [SHARED]
  • Curai Literature Review [SHARED]
  • [Note that this was before LLMs were mainstream. Foundational LLMs completely change this]
• Behaviour change matters more than clinical interventions
  • I realised that diagnosis doesn’t matter as much as treatment and behaviour change
  • Care programs and apps do a better job of productizing healthcare than statistical AI systems
    • [True in June 2022; now it’s different]
  • ‣

Adoption & reimbursement

I chatted with 50 customers, ran Facebook ads, and launched landing pages to understand how customers thought about healthcare.

• Many customers have habituated away the pain of ‘low quality’. Unsurprisingly, most people don’t want Peter Attia. The minority who do have a high willingness to pay. As expected, optimizer healthcare is a good $100m business but a lousy $10b business
• Customers primarily perceive quality via long consults, doctors who listen, short wait times, and compliant doctors. These are challenging vectors to compete on if you also care about cost containment and are operating in a FFS-reimbursed paradigm. Actual quality is different from perceived quality. Customers, by definition, only perceive quality
• ‘Convenience’ drives far higher click-through than ‘quality’
• Reimbursement matters more than distribution which matters more than product
• Several companies had developed playbooks for supposedly higher ‘quality’ medicine: </aside>

Hypothesis 2: Personalized medicine is the future. Only now is it possible to execute on this at scale. Early adopters will pay for a high-end product that generates the data to predict and prevent disease