My AI Healthcare Company Failed, Here Is Why

Some of the biggest challenges a healthcare company can face are sales distribution and having a product that is too closely tied to a single feature.

There is no right answer to solving these issues. Even former executives who were in the C-suite at major health plans and health systems struggle with them if they become founders.

You will most likely struggle with this too if you are a founder.

Yet, here are some lessons I’ve learned that will hopefully be helpful.

A disclaimer: If you are working on the idea I will be writing about or are involved in it, this is just my story, and you may have a completely different experience.

Over the past few years, I’ve iterated through a number of healthcare company ideas that I’ve bootstrapped. The one I’m writing about now is about AI language interpretation in healthcare.

The solution was to build an AI system that translates speech in real-time between patients and clinicians, as well as text for necessary notes for patients.

When building the company, I followed all the recommended steps I was advised to take.

I talked with dozens of different stakeholders. And made it a priority to find a pilot study at a health system.

All of this failed.

And when interviewing stakeholders who worked with interpreters, a clear problem emerged. They conveyed their frustration with having conversations with patients that take four times longer, as well as with the administrative challenges of finding an interpreter.

Yet, a potentially good product does not mean it will sell well in healthcare.

I also encountered a double-edged sword: although I benefited from the first-mover advantage, I was navigating an environment where regulations and guardrails were not yet established.

More generally, on a micro-level, here is why it did not work:

1. AI scribe companies were becoming prevalent, making the solution I was working on too close to a feature.

2. Communicating the nuances of fine-tuning a model, which differed from Google Translate, was challenging both from a sales perspective and in terms of convincing clinicians to use a specialized solution instead of a non-specialized product.

3. The product most likely requires an expensive validation study to prove its validity.

4. Translation companies are the ones with large datasets that allow for fine-tuning to create the best models. I tried engaging in a CVC partnership but found it tricky and encountered little interest.

5. These healthcare language translation companies already have the sales channels to sell their AI products. Additionally, there is a significant cultural shift for translators, who are concerned that LLMs might take over their jobs.

6. While the technology is needed, the total addressable market (TAM) for the product is not huge.

On a macro level, here is why it did not work:

There are a lot of "hair on fire" problems in healthcare; I should have been focusing on the "house on fire" problem.

Jessica Chao does a fantastic job articulating similar ideas. Even though I read her work back in early 2023, I learned these lessons the hard way.

This particularly stands out to me:

Thinking Bigger

So, I decided to focus on the translation product as a way to address broader challenges.

The concept was to target "cost concerns, mistrust, and lack of access." Theoretically, my approach made sense: by addressing one issue from the customer's perspective, you could build trust and eventually tackle more comprehensive changes.

Unfortunately, I've found that this approach doesn't align well with how things often work or with the challenges involved in addressing core problems.

How I think about this is in terms of local minimum vs. global minimum.

What I want you to take away from graph below is that you might think you've identified the root cause of the problem, but it could just be a local minimum.

The true root cause is the global minimum—the deeper, more fundamental issue.

In a non-D2C model, as you move closer to addressing the global minimum or root causes, the solutions tend to become more complicated and expensive based on my experience.

This makes raising money challenging, as you need more substantial proof points to attract investors.

However, this creates a chicken-and-egg problem: you need the product proof points to raise the money, but you need the money to develop and demonstrate those proof points.

Giving Up on the Idea and Not Myself

As a founder, you pitch your ideas to colleagues, friends, and family. You are excited, and they are excited.

You take leaps of faith. You spend your own money. You give your time and energy.

You feel like a wunderkind.

That this is your life’s work.

That this is a key milestone of why you were put on this planet.

Then the company just doesn’t click.

Unfortunately, I think there is a harmful theme in the startup world that you should always be resilient and never give up. Yet, one of the most important lessons I’ve learned is to know when to give up.

Thankfully, I never raised any capital for this company, so I was able to move on from this idea fairly easily.

And while I don’t agree with everything he says, Scott Galloway wrote in his great piece Quitting Time about how the more companies someone starts, the greater the likelihood of being successful.

It’s really only a failure if you don’t take the time to learn, reflect, and grow.

The journey to success often involves many hits to your ego. To truly succeed—not in terms of venture dollars but in terms of patient impact—you can’t let your ego be your goal.

What I've been realizing is that finding a balance between seeking hard truths and maintaining confidence in yourself is crucial.

One of the most important things I’m focusing on now is finding and maintaining relationships with people who are there for me in both good times and bad.

I’m optimistic that these learnings will enable me—and hopefully others—to improve our healthcare system and the means to get there.