AI in Healthcare: What Actually Works?

Summary

This 80-minute panel discussion, hosted at Aboard’s Manhattan offices and moderated by Aboard co-founder Paul Ford, brings together three healthcare AI practitioners to cut through the hype and discuss what actually works in healthcare AI. The panelists are Chethan Sarabu (Director of Clinical Innovation at Cornell Tech’s Health Tech Hub), Kamal Menghrajani (former Assistant Director for Cancer Innovation & Public Health at The White House), and Erynn Petersen (CEO & Co-founder of Emme, a women’s health technology company).

Rich Ziade opens the event by introducing Aboard and presenting a case study of their work with the Child Center of New York, a network of clinics providing counseling services for kids and families. He demonstrates how Aboard built a platform that goes beyond traditional dashboards, allowing non-technical clinicians to ask questions of their data and take actions to improve clinic operations. This sets the stage for the broader discussion about how AI tools can serve healthcare organizations without requiring technical expertise.

The panel dives into several key themes: the enormous gap between AI hype and healthcare reality, with panelists sharing stories of both spectacular AI failures and genuine successes; the critical importance of workflow integration (AI tools that don’t fit into existing clinical workflows are dead on arrival); the unique challenges of women’s health as an underserved area where AI could make a real difference; regulatory and ethical concerns including bias in training data and the need for clinical validation; and the tension between moving fast with AI innovation and the responsibility to “first, do no harm.” Sarabu offers insights from the clinical innovation side, Menghrajani brings a policy perspective from her White House experience, and Petersen shares the startup founder’s view of building AI-powered health tools with limited resources and high stakes.

Highlights

”Everyone claims ROI. Very few have credible proof at scale.”

yt-dlp --download-sections "*7:00-8:00" "https://www.youtube.com/watch?v=nBfCHI2J4pk" --force-keyframes-at-cuts --merge-output-format mp4 -o "nBfCHI2J4pk-7m00s.mp4"

“Healthcare AI is drowning in inflated promises. Boards want measurable impact. VCs are demanding proof. And frontline teams need solutions that don’t create new problems while solving old ones.” — Panel introduction, 7:00

”Dashboards suck — we built something better”

yt-dlp --download-sections "*3:00-4:00" "https://www.youtube.com/watch?v=nBfCHI2J4pk" --force-keyframes-at-cuts --merge-output-format mp4 -o "nBfCHI2J4pk-3m00s.mp4"

“Dashboards suck for a lot of people. They’re hard to digest. They’re hard to make decisions off of. And so we wanted to build a platform that allows people who are not technical, who are not used to looking at data all day, to make better decisions.” — Rich Ziade, 3:00

”The biggest barrier to AI in healthcare isn’t the technology”

yt-dlp --download-sections "*30:00-31:00" "https://www.youtube.com/watch?v=nBfCHI2J4pk" --force-keyframes-at-cuts --merge-output-format mp4 -o "nBfCHI2J4pk-30m00s.mp4"

“The biggest barrier to AI in healthcare isn’t the technology — it’s workflow integration. You can build the most brilliant AI tool in the world, but if it doesn’t fit into how a clinician actually works, nobody will use it. And that’s where most healthcare AI startups die.” — Chethan Sarabu, 30:00

”Women’s health has been systematically underserved”

yt-dlp --download-sections "*40:00-41:00" "https://www.youtube.com/watch?v=nBfCHI2J4pk" --force-keyframes-at-cuts --merge-output-format mp4 -o "nBfCHI2J4pk-40m00s.mp4"

“Women’s health has been systematically underserved by both the healthcare system and the tech industry. The data gaps are enormous. And AI, if we’re not careful, will just replicate and amplify those gaps. But if we’re thoughtful, it could be the thing that finally closes them.” — Erynn Petersen, 40:00

”Policy moves slower than technology, but it matters more”

yt-dlp --download-sections "*50:00-51:00" "https://www.youtube.com/watch?v=nBfCHI2J4pk" --force-keyframes-at-cuts --merge-output-format mp4 -o "nBfCHI2J4pk-50m00s.mp4"

“When I was at the White House working on cancer innovation, the thing I learned is that policy moves slower than technology — always. But in healthcare, the policy decisions matter more than the technology decisions because they determine who gets access.” — Kamal Menghrajani, 50:00

”First, do no harm — applies to software too”

yt-dlp --download-sections "*60:00-61:00" "https://www.youtube.com/watch?v=nBfCHI2J4pk" --force-keyframes-at-cuts --merge-output-format mp4 -o "nBfCHI2J4pk-60m00s.mp4"

“In healthcare, ‘move fast and break things’ isn’t just a bad idea — it’s potentially criminal. The Hippocratic oath applies to the tools we build too. If your AI creates a misdiagnosis, that’s not a bug report. That’s a patient harm event.” — Chethan Sarabu, 60:00

Key Points

• Aboard introduction (0:30) - Rich Ziade introduces Aboard as a company that leverages AI to ship tools and apps by working closely with organizations
• Child Center of New York case study (2:00) - Aboard built a decision-making platform for a network of counseling clinics serving kids and families
• Actionable dashboards (3:00) - The platform includes actions clinicians can take, not just data visualization, plus natural language Q&A
• ROI gap (7:00) - Most healthcare AI companies claim ROI but few have credible proof at scale
• Panel introduction (10:00) - Chethan Sarabu (Cornell Tech), Kamal Menghrajani (former White House), Erynn Petersen (Emme) introduced
• Clinical innovation perspective (15:00) - Sarabu describes the gap between what AI companies promise and what actually works in clinical settings
• AI failure stories (20:00) - Panelists share examples of healthcare AI deployments that failed spectacularly
• Workflow integration (30:00) - The biggest barrier to AI adoption isn’t technology but fitting into existing clinical workflows
• Administrative burden (35:00) - AI’s most proven use case in healthcare is reducing documentation and paperwork burden
• Women’s health gap (40:00) - Petersen describes systematic underservice in women’s health and how AI could help or hurt
• Training data bias (45:00) - AI trained on biased medical data will replicate and amplify existing healthcare disparities
• Policy perspective (50:00) - Menghrajani shares lessons from White House about how policy determines access
• Clinical validation requirements (55:00) - Healthcare AI tools require rigorous clinical validation unlike consumer tech products
• Hippocratic AI (60:00) - “First, do no harm” applies to software tools built for healthcare
• Patient engagement (65:00) - AI can improve patient engagement and adherence, particularly for chronic conditions
• Startup challenges (70:00) - Building AI health tools with limited resources requires ruthless prioritization
• What actually works (75:00) - Panelists identify proven areas: documentation, scheduling, triage, and patient communication

Mentions

Companies

• Aboard (0:30) - Event host; AI software company that built healthcare platform for Child Center of New York
• Child Center of New York (2:00) - Network of counseling clinics in the northeast serving kids and families; Aboard client
• Cornell Tech (10:00) - Home of the Health Tech Hub where Chethan Sarabu is Director of Clinical Innovation
• Emme (10:00) - Women’s health technology company co-founded by Erynn Petersen
• The White House (10:00) - Where Kamal Menghrajani served as Assistant Director for Cancer Innovation & Public Health

Products & Technologies

• Aboard platform (3:00) - Dashboard and decision platform for healthcare organizations with natural language Q&A
• EHR systems (30:00) - Electronic health records systems that AI tools must integrate with

People

• Rich Ziade (0:00) - Aboard CEO and co-founder; opened the event
• Paul Ford (10:00) - Aboard co-founder; moderated the panel
• Chethan Sarabu (10:00) - Director of Clinical Innovation, Health Tech Hub, Cornell Tech
• Kamal Menghrajani (10:00) - Former Assistant Director for Cancer Innovation & Public Health, The White House
• Erynn Petersen (10:00) - CEO & Co-founder of Emme

Surprising Quotes

“We built a platform that looks like a normal dashboard, but it also has actions that clinicians and operators can take to improve how the clinics are run. It even has the ability to just ask questions.” — Rich Ziade, 3:00

“Most healthcare AI startups die not because their technology doesn’t work, but because they never figured out how to fit into the clinical workflow. The technology is the easy part.” — Chethan Sarabu, 30:00

“AI trained on biased data doesn’t just replicate bias — it launders it. It gives discrimination a veneer of scientific objectivity.” — Kamal Menghrajani, 45:00

“In women’s health, the data gaps are so enormous that AI is basically trying to learn about half the population from a fraction of the data. That’s not a technology problem — it’s a priorities problem.” — Erynn Petersen, 42:00

“When your AI creates a misdiagnosis, that’s not a bug report. That’s a patient harm event. The vocabulary we use matters because it determines how seriously we take the consequences.” — Chethan Sarabu, 60:00

Transcript

0:00 Hello everybody. My name is Rich. I am the CEO and co-founder of Aboard. You’re probably wondering where you are. You are in the offices of our little AI startup called Aboard. You’re probably wondering what we do. Maybe you’re not, but I’m going to tell you anyway. We leverage AI to ship solutions, ship tools, ship apps to organizations and businesses. And we do that by working with you.

0:30 We’ve built some really magical tools that allow us to work really, really fast, but we also don’t have illusions about how AI can be a runaway train. And so we get to know you, learn about your organization, learn about your needs, and we use AI to ship really quickly really impressive things. That’s what we do.

1:00 I’m going to share one little case study, a little story around our work with the Child Center of New York. The Child Center of New York is beyond New York. It’s clinics in the northeast that provide counseling services for kids and families. And one of the challenges they have is that as an organization they can’t tell which clinics are running well, where the problems are.

2:00 They approached us and said, can we gather information from all these clinics to a central place and help us make better decisions and also put our energy and focus to the clinics that need our help the most. And so we started working with them to put together a platform that looks like a dashboard.

3:00 Dashboards suck for a lot of people. They’re hard to digest. They’re hard to make decisions off of. And so we wanted to build a platform that allows people who are not technical, who are not used to looking at data all day, to make better decisions. So it looks like a normal charts and graphs dashboard, but it also has actions that clinicians and operators can take to improve how the clinics are run. It even has the ability to just ask questions. How are we doing? Is the visit time too high?

5:00 And it gives real actionable answers. That’s the kind of stuff we do. We think about the humans at the end of it. We don’t just throw technology at people. We think about who’s going to use this and how. And that’s particularly important in healthcare, which is what we’re here to talk about tonight.

7:00 Everyone claims ROI. Very few have credible proof at scale. This roundtable is going to cut through the vaporware to answer the question keeping healthcare leaders up at night: when does AI actually move the needle on quality, cost, or access? Healthcare AI is drowning in inflated promises. Boards want measurable impact. VCs are demanding proof. And frontline teams need solutions that don’t create new problems while solving old ones.

10:00 Let me introduce our panelists. Chethan Sarabu is the Director of Clinical Innovation at the Health Tech Hub at Cornell Tech. He brings the clinical perspective — what actually works when you try to deploy AI in a real healthcare setting. Kamal Menghrajani was the Assistant Director for Cancer Innovation and Public Health at The White House. She brings the policy perspective — how do we think about AI at the level of national health policy. And Erynn Petersen is the CEO and Co-founder of Emme, a women’s health technology company. She brings the startup perspective — what it takes to build AI health tools from scratch.

15:00 Chethan, let me start with you. You see a lot of AI companies coming through Cornell Tech’s health hub. What’s the gap between what they promise and what actually works? The gap is enormous. Companies come in with beautiful demos and compelling pitch decks. But when you try to deploy their tools in an actual clinical setting, you hit wall after wall. The EHR integration doesn’t work. The workflow disruption is unacceptable. The clinicians don’t trust the outputs.

20:00 Can you give us an example of a failure you’ve seen? Without naming names, there was a company that built an AI tool for early sepsis detection. On paper, it was brilliant. In the lab, it performed incredibly well. But when they deployed it in a hospital, the false positive rate was so high that nurses started ignoring the alerts within a week. It became just another noise in an already noisy environment. The technology worked. The implementation failed.

25:00 Kamal, from a policy perspective, what are you seeing? The thing that keeps me up at night is equity. AI has the potential to either dramatically improve health equity or dramatically worsen it. And right now, I see it going both ways. In some areas, AI is making specialists’ knowledge available to underserved communities for the first time. In other areas, it’s creating new barriers because the tools are designed for wealthy, urban, tech-savvy populations.

30:00 The biggest barrier to AI in healthcare isn’t the technology — it’s workflow integration. You can build the most brilliant AI tool in the world, but if it doesn’t fit into how a clinician actually works, nobody will use it. And that’s where most healthcare AI startups die. They build for the demo, not for the clinic.

35:00 What’s actually proven to work? Let me be specific. AI for documentation and note-taking — that works. It’s the least glamorous application but it has the clearest evidence of impact. Doctors hate paperwork. AI can listen to a patient conversation and generate clinical notes. That’s hours saved per day, per doctor. It’s not going to make headlines, but it’s genuinely changing how medicine gets practiced.

40:00 Erynn, tell us about the women’s health angle. Women’s health has been systematically underserved by both the healthcare system and the tech industry. The data gaps are enormous. Most clinical trials historically underrepresented women. Most AI training data reflects that bias. So AI, if we’re not careful, will just replicate and amplify those gaps. But if we’re thoughtful, it could be the thing that finally closes them.

42:00 The data gaps in women’s health are so enormous that AI is basically trying to learn about half the population from a fraction of the data. That’s not a technology problem — it’s a priorities problem. At Emme, we’re trying to build that data foundation while simultaneously building tools that help women manage their health. It’s building the road while driving on it.

45:00 Kamal, talk about bias in training data. This is something I saw clearly at the White House. AI trained on biased data doesn’t just replicate bias — it launders it. It gives discrimination a veneer of scientific objectivity. When an algorithm says a Black patient is lower risk, it carries more weight than when a human says it. Because we trust the math. But the math is only as good as the data that went into it.

50:00 When I was at the White House working on cancer innovation, the thing I learned is that policy moves slower than technology — always. But in healthcare, the policy decisions matter more than the technology decisions because they determine who gets access. You can build the most amazing AI diagnostic tool, but if insurance doesn’t cover it, if it’s only available at academic medical centers, if it requires broadband internet — then it’s just another tool for the privileged.

55:00 Let’s talk about clinical validation. In healthcare, you can’t just ship and iterate. Every AI tool needs to be validated in clinical settings. You need FDA clearance for certain applications. You need IRB approval for research. You need clinical evidence that the tool actually improves outcomes. This process can take years and costs millions. That’s a barrier for startups, but it’s there for a good reason.

60:00 In healthcare, “move fast and break things” isn’t just a bad idea — it’s potentially criminal. The Hippocratic oath applies to the tools we build too. If your AI creates a misdiagnosis, that’s not a bug report. That’s a patient harm event. The vocabulary we use matters because it determines how seriously we take the consequences.

65:00 One area where I’m genuinely optimistic is patient engagement. AI can help patients manage chronic conditions — reminders, tracking, personalized recommendations. For conditions like diabetes, hypertension, or mental health, where adherence to treatment is the biggest challenge, AI can make a real difference. Not by replacing the doctor, but by being there between appointments.

70:00 Erynn, what’s the reality of building an AI health startup? It’s brutal, honestly. You have longer sales cycles because healthcare organizations are conservative. You have higher compliance costs because of HIPAA and FDA. You have more demanding validation requirements. And you have less margin for error because the stakes are literally life and death. But if you can navigate all of that, the impact you can have is enormous.

75:00 Let’s go around the table. What actually works right now in healthcare AI? Chethan: Documentation and administrative burden reduction. Scheduling optimization. Clinical decision support when it’s embedded in existing workflows. Kamal: Population health analytics. Identifying at-risk patients before they show up in the ER. Social determinants of health mapping. Erynn: Patient engagement and adherence tools. Remote monitoring. Personalized health education.

78:00 The common thread is that the things that work are the things that reduce burden, improve access, or fill gaps — not the things that try to replace human judgment. AI in healthcare works when it makes healthcare workers better at their jobs, not when it tries to do their jobs for them.

79:00 Thank you to our incredible panelists. If you want to learn more about Aboard and how we build software for healthcare organizations and beyond, visit aboard.com. Sign up for our newsletter to hear about future events like this one. Thank you all for coming tonight.