Sep 5, 2024 · 34m · a16z
Implementation, Data, Impact of Healthcare AI with Julie and Vijay
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of a16z Bio + Health's 'Raising Health' series, Vijay Pande and Julie Yoo address inbound audience questions on how artificial intelligence can lower healthcare costs, overcome historical data limitations, and transform clinician workflows and patient experiences.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
The host as informed peer 7.3 Guest teaching 2.1 Guest disagreement 1.3 The host pushing back 1.3
speaking balance: gold is the host, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 20:16 Vijay reframes the data sharing premise
Vijay questions the underlying premise of the audience question, arguing that sharing retrospective EHR data may not be desirable or valuable compared to generating prospective data.
Hardest push from the host ▶ 23:19 Julie challenges Vijay's takeaway on incumbent data advantageWhen Vijay claims legacy companies lack a data edge, Julie immediately counters that their massive labor forces give them a unique advantage in executing prospective RLHF.
Biggest teaching moment ▶ 7:12 Vijay explains healthcare's inelastic demand curveVijay educates the host and audience on why standard economic assumptions fail in healthcare, explaining that family health crises create infinite monetary demand that distorts market pricing.
The host holds their own ▶ 2:45 Julie delivers an expert breakdown of AI unit economicsJulie takes control of the technical commentary by presenting precise operational metrics on AI SDR performance, training costs, and budget category leverage to illustrate AI's administrative impact in healthcare.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The host as informed peer | Guest teaching | Guest disagreement | The host pushing back | Why |
|---|---|---|---|---|---|---|
| AI Potential to Lower Healthcare Costs | 8 | 2 | 1 | 1 | Julie demonstrates deep domain expertise by detailing how AI sales representatives achieve 5x cost reductions, 2x effectiveness, and 24-hour training ramps, using this parallel to project administrative savings in healthcare. She also distinguishes between IT budget allocation (10%) and labor spend (60%) to explain how healthcare systems will purchase AI agents. | |
| Major Blockers to Reducing Healthcare Costs in America | 7 | 2 | 1 | 1 | Julie offers an insightful perspective on bending the cost curve through productizing bespoke human services into standardized diagnostics and drugs. She further analyzes how third-party payer contracts remain untethered from actual care delivery costs. | |
| Financial Incentives and Human Factors in AI Adoption | 8 | 2 | 1 | 1 | Julie articulates the emotional drivers of physician AI adoption, citing reduced burnout from ambient scribing tools over pure financial incentives. She also demonstrates advanced regulatory knowledge by framing the distinction between FDA diagnostic pathways and licensed clinician credentials. | |
| Data Sharing and Prospective RLHF Data Generation | 7 | 3 | 2 | 1 | Vijay reframes the question by challenging the value of retrospective EHR data compared to prospective RLHF data generation. Julie agrees, offering a sharp critique of EHR data quality as an abstract and sporadic representation of real patient journeys. | |
| Do Legacy Healthcare Companies Have a Data Advantage? | 7 | 2 | 2 | 3 | When Vijay presents a spicy take that legacy healthcare incumbents lack a data advantage, Julie pushes back gently by highlighting that legacy players possess the largest clinical workforce capable of serving as RLHF annotators. She also discusses real-world consumer behavior around AI disclosure in healthcare applications. | |
| Human Backstops and Exception Handling in Care Delivery | 6 | 2 | 1 | 1 | Vijay outlines a timeline of progressive automation from nursing to specialty care, concluding that human intervention will diminish over time. Julie synthesizes the dynamic by identifying exception handling as the permanent human backstop in clinical workflows. | |
| Personalized Preventative Health at the Individual Level | 8 | 2 | 1 | 1 | Julie critiques current healthcare delivery for relying on lowest-common-denominator population health baselines. She illustrates the potential of AI-driven precision medicine by detailing how lung cancer was subdivided into seven distinct genomic subtypes. |
Statements from this episode (18)
Pande: Biggest AI Healthcare Savings Will Come From Value-Based Care
“But the big win, as I think about it, is like in the sort of value-based care sense, like in the sort of not the sick care fee-for-service sense, but in the truly keeping us healthy sense. If AI can get out in front of things, Then hopefully, you know, we decr…”
Yoo: Healthcare Downstream Costs Are Driven by Clinical Mistakes
“And to me, the cost of mistakes piece is the thing that's, I think relatively unique in healthcare where, again, the cost of a misdiagnosis or a cost of a incorrect triage, you know, that's really, really what, you know, is ballooning so many of the downstream…”
Pande: AI Will Become Significantly Cheaper Than Human Healthcare
“Yeah, so, so AI is going in the Moore's Law direction, and healthcare is going in the other one. Sooner or later, AI's gonna be way cheaper.”
Yoo: Labor Accounts for 60% of Healthcare Budgets, IT Only 10%
“And that could actually hit my labor spend budget, which obviously is, you know, it's like 60% of the overall budget versus the 10% on the IT side.”
Vijay Pande: Compute costs will fall a million-fold over 20 years
“The nice thing about the exponentials is that, you know, things will go down a factor of a thousand, like, every 10 years on the Moore's Law side, and we'll continue to see that. So, a thousand, maybe in 20 years, a million, you know, a thousand times a thousa…”
Yoo: Soft factors will drive near-term clinician AI adoption over money
“I personally think that that it's probably gonna be more the, these soft things in the near term. That drive adoption, at least for clinicians and then, you know, perhaps if you're able to layer on a financial element to it could just be icing on the cake.”
Pande: Healthcare AI needs a 10x improvement threshold to overcome stigma
“Yeah, so I think the bar is different. This might be the magic has to get to 10 X. Mm-hmm. Something where 10% is probably within noise and without doubt or something like that, but 10 X is kind of the thing where, like, okay, okay. We have to think of things …”
Pande: Healthcare AI will transition to real-time generated training data
“The alternative, which I think is going to happen instead, is that we're going to see data generated by for AI right now.”
Yoo: Electronic health records fail to capture real patient journeys
“EHRs are at best a highly abstract, you know, representation of patient journeys, at worst, you know, just sporadic encounter-related data that really does not capture the essence of what's happening with that individual.”
Pande: Healthcare service providers will eventually transition into AI companies
“Well, you can imagine a services company that's providing some sort of healthcare service. It's training AI with its workers. And then the easy things, which are probably the boring things anyways, AI starts to do. And then the more interesting, complicated th…”
Vijay Pande: Legacy Healthcare Data Advantage Is Overestimated
“The sort of spicy take on it is maybe not as much as they think. You know, because the temptation is the data's there, and this may be true for pharma too, may be true for pair of providers that, that data, there's a lot of it, but it might not be as good as t…”
Vijay Pande: Culture Will Prevent Healthcare Incumbents From Becoming AI Companies
“I think it's also not to underestimate the cultural sort of differences between, like, an AI company and a more traditional services company, and getting one to be the other, you know, is, is like getting a boat to be an airplane sometimes It might be pretty h…”
Pande: Medicine will transition from a human problem to a data problem
“I think we'll get to that point where medicine will be about incorporating all this data. Data about me as a patient, data about everyone else that's like me to be able to draw the best conclusions. That's something that is just a computer problem. And that mu…”
Yoo: 30 to 40% of hospital inbound calls only require basic lookups
“We used to look at, in my startup we worked a lot with call centers for hospitals, and it was something on the order of, like, 30 to 40% of all inbound calls were things that just required a lookup and didn't require a human”
Pande: AI adoption in medicine will progress from nursing to specialists
“We're gonna see it gradually, where first it's some bits of nursing, then more of nursing, then some bits of a GP, then maybe some bits of a specialist, and maybe certain specialists like neurology or something like that, where it's about prescribing drugs bas…”
Julie Yoo: Exception handling in healthcare will always require human clinicians
“So I think, at least for the foreseeable future, you know, exception handling, I think, will always require the human input.”
Yoo: Population-level healthcare inherently fails the majority of patients
“And arguably everything that's wrong with healthcare is because it is done at the population health level, because it's inherently then designed for, like, least common denominator, and therefore, like, the majority of people aren't actually being accommodated…”