How Does Startup Magic Happen? The Carbon Health Story.

Eren Bali
September 2, 2021
15 mins

Founding a startup requires that you believe you can do a better job at something than established companies with a thousand times more resources. And sometimes those startups succeed. How does that happen? It may look like magic, but in fact it’s because they do something crazy to gain an insight that others have missed. 

This is the story of how we did this at Carbon Health.

At first, there were just four of us: a designer, a developer, a physician, and me. We’d picked a simple goal: to provide the healthcare we’d want for ourselves and our families in a way that would be affordable to anyone in this country. 

I knew this meant we’d have to fix electronic health records (EHRs), the software at the heart of healthcare delivery. Everybody hated EHRs. They were behind the times: compared with the software used by Uber drivers, this software built for doctors was grotesque. I’d call it “laughable,” but it’s not funny that doctors’ time has been wasted by these cumbersome tools. 

Compared with the digital tools available to cab drivers, the tools that most doctors use are outdated, grotesquely inefficient, and incredibly expensive. 

It looked to me like no one had even tried to build truly efficient software for this purpose. Yet I was convinced that the quality of healthcare in the U.S. couldn’t improve until we’d cracked the nut on this fundamental technology piece.

My first crazy move was telling the team, “We’re going to build our own end-to-end care platform from the ground up — without even looking at legacy EHRs.”

We not only rejected all the existing out-of-box EHRs — I actually forbade even looking at them. I told the team I didn’t want to see any screenshots of other EHRs, and I didn’t care which features other EHRs had or didn’t have.

This isn’t a common thing to do. Usually, you start building something by doing lots of market research, by analyzing competitors and understanding their weaknesses. I was sacrificing all this because part of the reason nobody had built amazing EHR software is that this software is complicated and messy. There are a thousand details you have to figure out, and nobody has all that domain knowledge in their head. So there’s a temptation to look at how everyone else does these things and make a slightly better version of it. Going in, we had no idea how people should be doing prescriptions, or labs, or medical referrals, or rescheduling — but we knew it wasn’t being done right. I was worried that if we started with what other people were already doing, we’d unintentionally bring all of their historical baggage with us. 

But how would we even know what we needed to build?

Sketching Out a Plan

That’s when we did the second crazy thing: we decided to open a primary care clinic inside our San Francisco office. Not a mock one — a real one. 

I remember that conversation well. I asked Greg Burrell, MD, my physician co-founder, “Hypothetically, could you open a primary care clinic?”

He said, “Yes, but why?”

I said, “So, hypothetically, you could prescribe medications?”

He said, “Yes, but why?”

I said, “Hypothetically, you could do referrals, accept insurance, and run your own internal medicine practice. Right?”

I could see the look in his eyes going from curiosity to concern.

“Yes....” he said. “But why?!?”

I said, “We’re going to build a full primary clinic from scratch, with you as the leading physician. Let’s hire some nurses, hire some other staff, and figure out how primary care should work, from first principles.”

What the team didn’t know is that before this conversation, between Christmas and New Year’s Day (which is the most productive time of the year for immigrants like me without families to visit), I had made 50-plus sketches about what healthcare would look like from the patients’ perspective, if I designed the experience the way I wanted it for myself. I was just curious whether I knew what I wanted. 

It was somewhat straightforward. Obviously, it has to be mobile-first, I thought, because it’s 2016! I needed a way to onboard with a healthcare provider, a way that would tell them about my demographics, family history, insurance, and so on. I would then need to choose a primary care provider to build a relationship. After that, all I needed was to be able to book an appointment — obviously, I needed to tell the provider what my appointment was about. And a way to answer all the questions that every doctor asks should be in the app, so I didn’t have to waste time on that in the clinic. 

What next? I go to the clinic, and obviously there’s no paperwork anywhere — because it’s 2016. I don’t need to spend time filling out annoying forms; I already did that in the app. And I hate when people ask me for the address of my pharmacy. I don’t have that memorized! Since it’s 2016, I thought, I can pick a nearby pharmacy myself on a map — in the app, of course. 

Last but not least, I thought about what happens whenever I talk to my sister, who’s a physician. She gives me a “care plan” to follow, but she also checks back with me a couple days later to make sure I’m recovering as she intended. I wanted my app to do the same thing — I didn’t want to be handed a care plan on a piece of paper (I always lose them). 

That was all I wanted as a patient. And what got me really excited was thinking that this could be accessible to everyone, without increasing the cost at all. So I sketched all of those things. 

I shared my sketches with the team. People were surprisingly open to the idea! Their only concern was that this experience felt too good to be true. If this was possible, why hadn’t someone already done it? This is what we call analog thinking. It’s often a valid concern. But when we dug into all the details, there was nothing that seemed impossible from the first principles. If we asked domain experts, they would tell us it was a bad idea. But people who have been operating in a domain for a very long time, seeing all the other smart people fail at new ideas, tend to believe that however things are, they’re that way for a good reason — even if there’s no reason at all. The only way for us to really know was to actually do it.

Building the First Carbon Health Clinic 

So we embarked on a plan to launch a primary care clinic and start seeing real patients, completely on our own technology platform, in nine months — so we could show it at TechCrunch Disrupt in September 2016. I didn’t feel that we needed to be in stealth mode for this business. It would have to grow by word of mouth, so we needed to create excitement.

One of the hardest things was finding a dual-purpose physical location—a place that could serve both as an attractive tech office and as a primary care clinic. But once we did that, the magic of it was that our physicians, nurses, designers, and engineers could all have lunch and work through problems together.

And there were a few other challenges: setting up a professional corporation around Greg; getting the ability to work with pharmacies, labs, and medical imaging centers; and working with insurance companies. Unlike Uber or a food delivery service, we couldn’t provide the service entirely by ourselves. It was a day-one requirement for us to be able to work with the rest of the healthcare world.

Thanks to the intricacies of that world, we ran into one big problem: we had to accept insurance. This was essential if we were going to stay true to our mission of building healthcare that would be accessible to everyone. Cash-based services are viable only for high-income people. But getting an insurance contract takes 12 to 18 months. And starting with out-of-pocket services while working toward insurance-based services wouldn’t work because it would skew our thinking toward the needs of affluent people.

That’s when I came up with a third crazy idea. We acted like we were accepting everyone’s insurance, but we just didn’t submit any claims. We provided free care, without telling anybody it was free.

I was OK with this because early at a startup, it’s all about learning. It’s about testing your hypothesis in the most accurate way possible. As long as the patients believed we were charging their insurance companies, we were learning. If they knew we were operating a free clinic, then they might like it only because it was free. It wouldn’t be an accurate test.

Proving Our Concept

In October, 10 months after we started, we opened our clinic to friends and family. 

We were trying to prove three things:

First, we had to prove that we could provide a much better experience to patients — without increasing cost — using a technology-first approach. 

Second, we had to prove that we could do this while maintaining an enjoyable experience for clinical teams.

Third, we needed to be able to do those two things while maintaining reasonable unit economics. 

We spent our very limited software resources exclusively on the patient app in the first year. We did the bare minimum on the provider side to make that work. Proving that our first few consumers would love this experience didn’t take that long, to be honest. We already understood them pretty well. But I made sure to quickly expand to serve patients across the entire city, because affluent tech employees were over-represented among our friends and family in the Bay Area. We needed feedback from a more diverse group. 

Things became a lot more fun for me when people with real health needs started using Carbon Health. They cared about different things. They loved the core platform, but they also had substantive recommendations for improvements. They had needs we weren’t addressing. That’s when we started seeing the most exhilarating feedback loop: talking to our patients, sitting down with our doctors, sketching out new ideas, building the next day, sending a Slack notification to our team that we were launching a new version of the app in a few hours, and waiting anxiously to see how it worked for them with the next patient. It was true magic. 

All this is about putting yourself into a position of insight — one of the company values I wrote down even before I knew what the new company was going to do.

The key idea is that no amount of looking at reports and research will be as high fidelity as being in that position of insight. Inside our new office, we have a 1:1 full-size “cardboard clinic” for just this reason.

At our headquarters, a full-size cardboard replica of a typical Carbon Health clinic allows us to continually experiment and optimize our processes. Every week, our clinical teams come in and role-play visits, doing exactly what they do every day. Our architects and designers observe, and we optimize everything to make sure it’s efficient.

Our cardboard clinic is partially inspired by a movie called The Founder, which is about the founding of McDonald’s. There’s a scene where the founders decide to try selling burgers for 15 cents in 15 minutes. To do this, they rent a tennis court, draw diagrams of the hypothetical kitchen on the ground, and role-play and iterate until the operation is fast enough to hit their goal.

So Did It Work? 

Kind of. We had come up with a lot of original ideas about how healthcare delivery could work, both on the patient side and on the provider side, but there were also a lot of holes in our system. It hadn’t been battle-tested in a high-volume environment. 

I think we got a little bit lucky. I met this physician named Caesar Djavaherian — an emergency medicine doctor who owned a small chain of urgent care clinics in the Bay Area. He was passionate about the same things we cared about: providing an amazing patient experience using technology in a way that was still accessible to everyone. Their modern-looking clinics were using 15 different pieces of software to create a tech-forward feeling. But behind the scenes, it was a mess. The staff were juggling disjointed products, and the patient experience was still nowhere near where Caesar wanted it to be. The moment I demoed our product for him, his eyes lit up. Caesar had a long history of working with information systems in large healthcare organizations, so he’d seen every single piece of healthcare software that existed. But, in his own words, he’d never seen anything like what we’d built.

I remember the first comment he made after looking at the demo. He said, “I can tell that you guys are actually running a clinic yourselves; you’re not just a software company.” 

In retrospect, the tell was that we’d built a lot of functionality in areas no other company even remotely cared about — onboarding patients, care plans, etc. — while having massive holes on the business side, like practice management and revenue cycle tracking. But Caesar decided to take a leap of faith. We worked with his operational partner Sharmin Holla to launch Carbon in all four of their clinics. They wanted to launch in all four on the same day — so there was a huge list of operational and business requirements. We triaged meticulously, and eight months later, we were ready to launch. (Apparently, Sharmin and Caesar thought there was no way we’d actually be able to make the business side ready in nine months, but they hadn’t told us that.)

The day of reckoning came. We imported all existing patient records that night, and the next morning — a Thursday — we were going live in all four clinics!

We stationed one engineer at each of the four clinics. I was one of them. That first morning was kind of chaotic — onboarding chaos, signups, people getting confused about scheduling, the clinical teams using a brand-new platform. And there were a lot of bugs. We were literally fixing bugs and broken experiences while patients were complaining two feet away. 

In the afternoon, things got a tiny bit calmer. The next day wasn’t as bad. By Saturday, we were in good shape. Despite all the preparations, nothing beats the trial of real-world launches. 

“Everybody has a plan until they get punched in the mouth.” — Mike Tyson

Six months later, it was clear that patients loved the app, along with the smooth clinic experience that it allowed. It wasn’t just anecdotal comments or reviews that told us this; I was obsessively looking at the patient retention that these clinics saw, both before and after we deployed our platform, using cohort analysis. Their quarterly cohorts over the past three years had been very consistent, and we had clearly bent the curve, in a very visible way. 

Over lunch at a nice Peruvian restaurant in San Francisco, I shared with Caesar what was obvious to me at that point: we should merge our software company with their clinical services company and become a full-stack healthcare provider, because the real magic could work only if we could get more and more opinionated about workflow and technology simultaneously. He agreed. 

With our combined company now run by people who understood clinical operations and product, we iterated for two more years. For twelve quarters in a row, we were able to improve patient retention and provider efficiency to hit our unit economics targets. 

The rest is history. We raised more capital in multiple rounds (a total of half a billion dollars). From there, we pressed on the gas and started expanding at a pace that put us into the #2 rank on the Inc. 5000 fastest-growing private companies list. But what most people don’t know is how much work and iteration went into getting the customer experience and unit economics right — a.k.a. the magic — long before anyone had heard of us.

Eren Bali

Eren Bali is the co-founder and CEO of Carbon Health. Eren leads the Carbon Health team toward his vision of democratizing healthcare and making high-quality care accessible to everyone.