AI, Data, and the Art of Failing Fast in Healthcare Innovation
If you’ve ever tried to teach a toddler how to use a spoon, you know the process involves a lot of trial and error—and a surprising amount of yogurt in your hair. That’s essentially what we’re doing in healthcare innovation right now: trying to figure out how to feed the future with the most advanced spoon (AI) we’ve ever held, but we keep flinging data all over the place. As a Pediatrician, Father, and a closet full healthcare devices (including StarTac, Nokia + Blackberry + palm pilot phones) - I've learned alot about toddlers, human development, and technology.
I’ve been on both sides of the healthcare coin—provider and innovator—and I can tell you this: we’re not failing fast enough. Or maybe we’re failing too slowly. Let me explain.
The Data Dilemma
Healthcare generates massive amounts of data—EHR notes, imaging results, patient portals, wearables, and even Aunt Margie’s health-tracking app that measures steps and stress. Theoretically, AI should thrive in this environment. The more data we feed it, the smarter it gets, right? Well, here’s the catch: we’re feeding it stale crackers instead of a banquet.
Consumer-grade data, the kind that comes from wearables or apps, is immensely valuable. It can tell us what the patient’s life looks like between visits. Are they walking? Breathing? Daily temperature changes, heart rate variability, and SLEEP - sleep data is one of the most important missed vital signs. (We humans spend about one-third of their lives sleeping - more on Sleep in a future post). But healthcare is still slow to incorporate these sources. Why? Because we’re afraid of the unknown—or worse, afraid of being wrong. We do not need just FDA cleared device data to capture all of this rich immensely valuable data..
Fail Fast, Fix Faster
Here’s the thing about healthcare: it’s cautious. And for good reason! No one wants to be the person who accidentally breaks a system that’s supposed to save lives. But innovation thrives on speed, and that requires a cultural shift.
I remember the first time I suggested to a colleague that do a virtual telemedicine visit with a a small group of super stable post-operative patients. They looked at me like I’d just asked them to let a toddler perform surgery. “What if it doesn’t work?” they asked.
“That’s the point!” I replied.
Failing faster doesn’t mean failing recklessly. It means piloting, iterating, and learning quickly. It means not waiting five years to publish a retrospective study before deciding telemedicine might have potential. The truth is, every failure teaches us something, but only if we’re willing to listen.During the Pandemic - we ALL were forced to provide virtual visits (telemedicine), we didn't care if was on a platform or a video-conferencing solution. We were forced to innovate and..... it worked - We learned, we improvised rapidly, and it worked!!!
Recommended by LinkedIn
Learning by Doing (and Failing)
Let’s go back to the toddler with the spoon. What if, after one messy attempt, you declared, “Clearly, spoons don’t work for feeding children.” You’d probably be laughed out of parenthood. Yet in healthcare, we sometimes have the same reaction. If an AI model doesn’t perform perfectly right out of the gate, we question its entire existence.
Instead, we need to treat AI like a toddler. We need to:
The Human Factor
Here’s where I admit a hard truth: AI won’t replace human judgment (and today should not be used in Diagnosing or Prescribing) in healthcare. At least, not entirely. And that’s a good thing. Patients don’t want to feel like they’re being cared for completely by an algorithm. They want empathy, connection, and understanding. AI should connect with humans and keep them in the loop.
AI can be a tool to enhance human care, but it won’t succeed if we don’t set it up to succeed. That means feeding it the right data, training it with diverse perspectives, and constantly tweaking the process. It also means accepting that we’ll sometimes be wrong—and learning faster when we are.
A Spoonful of Humility
So, where do we go from here? We embrace the mess. We learn from it. And we remember that every yogurt-covered wall is a step closer to a world where healthcare innovation moves at the speed of life.
AI won’t fix everything overnight, and it won’t always save the day. But with the right data (the right partnerships), a willingness to fail fast, and a bit of toddler-like determination, it just might help us build a healthier future.
Now, if you’ll excuse me, I have a spoon to clean.
Homeopathic Psychiatrist
2wThis was an incredible read!👌
Innovation Strategist| Clinical Researcher|Digital Health | AI| Informatics| User Experience | Clinical Operations, Quality, & Patient Safety| VR | Wearables|Tech Innovation| Implementation and Scale|Process Improvement
2wGreat article! The crux of human-centered design lies in the philosophy of failing fast and failing often. By rapidly testing prototypes and gathering continuous input from end users, we ensure the technology not only meets the needs of clinicians and patients, but also solves problems in a way that is seen as value-added, safe, and efficient within clinical workflows. This iterative process drives innovation, fosters trust, and results in solutions that are both impactful and seamlessly integrated into practice. Thanks for highlighting this important aspect!