đ§đ How US Healthcare Actually Works | Nikhil Krishnan, Out of Pocket
The WWII tax loophole that broke it forever, where AI is seeing real adoption in healthcare today, why most medical software sucks, and the most common bad healthcare startup ideas
Nikhil Krishnan is the Founder of Out of Pocket, a media company that makes understanding healthcare more entertaining and accessible.
We spend 100 minutes talking about how the US healthcare system actually works. Nikhil details how a change to the tax code during World War II changed it forever, all the ways AI is seeing real adoption in healthcare, why physician burnout and independence is one of the biggest problems in the industry, and most common bad startup ideas that come up over and over in healthcare.
Support this Episodeâs Sponsors
Numeral: The end-to-end platform for sales tax and compliance.
Flex: Sign-up for Flex Elite with code TURNER, get $1,000 here.
Amplitude: AI analytics, all you have to do is ask.
Inquire about sponsoring future episodes here.
đ Stream on Spotify and Apple
Timestamps to jump in:
1:23 How the US healthcare system works
4:26 Why US healthcare is different from the rest of the world
12:01 Why healthcare costs keep going up
15:58 Core problem: is healthcare a marketplace or not?
21:04 How money flows + Two-way price negotiation
27:34 Why payments are seeing early AI adoption
30:08 How AI could change healthcare delivery
35:40 Doctorâs are trapped on a productivity hamster wheel
39:28 How incentives shape healthcare delivery
43:53 Healthcare is an implicit jobs program in the US
48:45 Areas AI is overhyped, worst healthcare startups
55:30 Consumerization of healthcare
1:01:58 Rise of Peptides, understanding risks and downsides
1:09:35 Why all medical software is so bad
1:11:58 How to do enterprise sales in healthcare
1:14:51 The battle forming between Scribes, Search, and EMRs
1:18:33 Why we need more physician independence
1:26:51 Starting Out of Pocket in February of 2020
1:29:12 Write to meet your audience
1:37:54 Using AI as a content creator
1:42:13 How to get started writing on the internet
Referenced:
Find Nikhil on X / Twitter and LinkedIn
đ Stream on YouTube, Spotify, and Apple
Transcript
Find transcripts of all prior episodes here.
Turner Novak:
Nikhil, welcome to the show.
Nikhil Krishnan:
Whatâs up? Excited to be here.
Turner Novak:
Iâm excited to have you too. Weâre going to talk healthcare more broadly, but also tactically, and then some content creation stuff. I guess weâre both content creators and people always want to know about that.
Nikhil Krishnan:
This feels like when Twitch streamers collab on something and the internet freaks out. This is the Avengers of people that are too online, me and you.
Turner Novak:
Yeah. The comments for this, people are going to be absolutely losing their minds about two heavy hitters from the internet just coming together.
Nikhil Krishnan:
From the extremely specific niche of the internet.
Turner Novak:
Yes. But really quick, for people who have never come across you on the internet before, how do you describe yourself and Out of Pocket more generally?
Nikhil Krishnan:
You just described both of us.
I try to make healthcare more entertaining and accessible. Thatâs the short version. Weâre trying to teach people how healthcare works. I write a newsletter, we do a bunch of courses, we have events, etc. The general idea is we want it to be easier for people to understand how healthcare works, and we hope that people will build more interesting things if it makes more sense to them.
Turner Novak:
That makes sense. How would you describe the healthcare industry to someone who has no idea how it works?
Nikhil Krishnan:
I would just step on your foot really hard and be like, âHow does that feel?â Thatâs the entire healthcare system in a nutshell.
Turner Novak:
Thatâs it. And then you hand me a bill after you step on my foot.
Nikhil Krishnan:
Yeah, exactly. This is for your own good.
Turner Novak:
But you send it three months later, I canât access it online, and itâs an envelope that gets lost in the mail.
Nikhil Krishnan:
As it should be.
The US healthcare system is messed up. Itâs very confusing. The reality is that US healthcare was never a planned system, the way a lot of other countries actually designed theirs. It just appeared and congealed over many, many layers. Because of that, weâve created bespoke rules for different slices of the system.
What we now have is more like 50, $100 billion micro systems in a trench coat underneath what we call a $5 or $6 trillion healthcare industry. Each one has different regulations, different people who pay for it, different services theyâre allowed to offer.
One of the reasons healthcare has so much administrative bloat is that you now have to track all of these different rules. If someone jumps from one system to another, you have to understand what that means and go chase down a bunch of information. Itâs a very messed up system. But I really think of the US healthcare system as a bunch of small micro healthcare systems within a larger umbrella.
Turner Novak:
And you said 50. Is that because each state is different, or is it more about different aspects of the system?
Nikhil Krishnan:
Yeah, I shouldnât have used 50. Thatâs a little too on the nose for states. The categorization is really more about who pays. It could be a state if itâs Medicaid, or the federal government if itâs Medicare. But now the federal government also contracts with private insurance companies for Medicare Advantage, so thatâs a different set of rules. If you get your health insurance through your job, thatâs another set. And sometimes an employer will pay medical bills directly rather than go through an insurance company, which is its own set of rules.
Itâs actually probably more than 50 if you really think about it, but itâs not state-based. Itâs really about whoeverâs footing the bill at the end of the day. Compare that to other countries. If you hear âsingle payer system,â thereâs one payer, and itâs the government. In the US, we have a multi-payer system, and each of those payers has different rules.
Turner Novak:
You mentioned this idea of planned versus unplanned. In other countries, was there a moment where the government decided, âWeâre taking this overâ? And how did it evolve differently here?
Nikhil Krishnan:
For a lot of other countries, the government basically said, âHereâs the role weâre going to play in the healthcare system.â Whether thatâs setting prices, running hospitals directly, or creating a marketplace where insurance companies compete, the government made a clear decision about its role.
The US is very different. The weird quirk here is that employers choose health insurance on behalf of their employees. This is one of the original sins of US healthcare, and it wasnât born out of any planned healthcare decision. It came from a tweak in the tax code.
Everyone was off fighting World War II and the government was worried about wage inflation. The labor force at home had a lot of leverage, so the government said, âWeâre going to cap wage growth, but in exchange, you can use tax-exempt dollars for other things.â One of those things was health insurance. So this was essentially a footnote, you know what I mean?
Turner Novak:
It wasnât really meant to be that big of a deal back in the â40s?
Nikhil Krishnan:
Right, because healthcare was also very cheap at the time. It was house calls and people would get cocaine for a migraine and move on with their day.
Turner Novak:
Hereâs some hard drugs. Because we didnât really have a pharmaceutical industry. People werenât really getting surgeries. You got sick and you died.
Nikhil Krishnan:
Yeah. We didnât have the level of complex hospital care we have today. A lot of that is a post-World War II development. So because of that, healthcare was a blip in terms of cost and people didnât think much about it. But that weird tax quirk essentially created, or really poured gas on, what we know as the third-party private insurance industry.
As healthcare got more expensive and more complicated, it had all the factors needed to become runaway costs. Employers had tax-exempt dollars, people werenât paying out of their own pockets, and healthcare itself kept getting more intensive. Costs started ballooning. Then in 1965, we created Medicare, a whole new set of rules, with the government getting much more involved in payment. You start layering these things on top of each other, putting band-aids on problems, and it was never a fully planned healthcare system.
Turner Novak:
Okay. And when you say tax-exempt, just so everyone knows, when you give someone a salary, the company and the employee both pay taxes on it. But with health benefits, itâs compensation you receive without taxes. The company doesnât pay taxes on it either?
Nikhil Krishnan:
Yeah. The clearest way to explain it: if youâre self-employed, youâll probably buy health insurance off the individual exchanges, the marketplace. When you do that, youâre paying with post-tax dollars. When an employer does it, theyâre paying with pre-tax dollars. That alone is a huge boost.
If youâve ever left a job and had to pay COBRA, continuing your old employerâs health insurance, youâll notice the cost spikes a lot even though itâs the same plan. Two reasons. One, your employer was probably subsidizing a large portion of your premium. Two, youâre now paying it in post-tax dollars. Double whammy.
There are some newer tools getting a lot of traction too, like ICRAs, which are a popular startup idea in VC circles. The concept is employers create a wallet to give you the tax-exempt dollars theyâd normally spend on health benefits, but now you can go shop around for health insurance or other healthcare services yourself. Same tax treatment, but more consumer-oriented.
Turner Novak:
And one interesting topic. You mentioned healthcare is five or six trillion dollars, something like 20% of GDP, and itâs always going up. Why do healthcare costs keep rising? Canât someone just say, âThis is ridiculous, letâs cut costsâ?
Nikhil Krishnan:
Yeah. A couple things. First, as a percent of GDP, healthcare actually hasnât gone up as much as people think over the last 10 to 12 years. Thereâs a lot of debate about why. Some say people are rationing care, thereâve been drug breakthroughs for heart disease that brought costs down, others say the government just pays less now. But it actually hasnât risen as much as we think.
The caveat is that who pays has been shifting a lot. Within that 17 or 18% of GDP, a lot more cost has shifted onto individuals, through higher deductibles, higher premiums, or wages that havenât gone up because the money went to healthcare costs instead.
Second, at a macro level, our population is getting older, so people are consuming more healthcare services. And there are new popular categories, like weight loss drugs, that are expensive and in high demand. Which brings you to an ideological question: what should the mechanism be for reducing prices or consumption? Itâs not simple.
If you just say âweâre not paying for this anymore,â someone has to ration care, and that gets very political. Healthcare is also the number one employer in most states. If you cut spending, a lot of people lose jobs. Look at the jobs report from the last year. The only sector with consistent positive growth is healthcare. Itâs an implicit jobs program.
On the price side, one of the cardinal sins of the US is that we donât have a centralized price negotiation system the way other countries do. In other countries, the government sets prices or caps how price growth should be. In the US, we donât have that. So when the government says it wonât pay as much for something, hospitals just raise prices on the private insurance side. Itâs like squeezing a balloon. Reduce it in one place and it grows somewhere else.
Turner Novak:
Yeah, thatâs fair. You could argue America is a capitalist market, there should be competition that brings costs down naturally.
Nikhil Krishnan:
I think the core problem with US healthcare is that we donât know which one we want. Do we want a government-intervened system, or a free-market one? Thereâs no agreement. Each administration goes one way, the next might go the other.
If you ask 50 different people what the ideal healthcare system looks like, you get 50 different answers. And because thereâs no unifying theory of what we want, we canât build toward it. So weâre just regulated enough to prevent competition, and just free-market enough to enable price gouging.
Turner Novak:
Yeah. Depending on who you are, thatâs either an incredible opportunity or terrible at the same time.
Nikhil Krishnan:
Totally. The marketplaces are a good example. As part of the Affordable Care Act, the idea was everyone would buy health insurance on an exchange, a real marketplace with competition, where you pick based on customer experience, coverage, deductible, etc.
But for a marketplace to work, everyone has to be in it. If only relatively sick people come to the marketplace, or if most people stay on employer insurance because itâs more attractive due to the tax benefits we talked about, itâs not a real marketplace. It kind of looks like one, but itâs not quite marketplace enough to actually work.
Turner Novak:
Have we ever tried to move off the employer-tied healthcare model, or has it never really been attempted?
Nikhil Krishnan:
The Affordable Care Act was really trying to do that. It had two prongs: make it more attractive to buy your own health insurance on the exchange, and make it less attractive for employers to offer rich benefits. There was something called the âCadillac tax.â If you offered health coverage that was too good, youâd get extra taxed. The idea was to push people out of the employer pool into the individual exchange pool. But that part of the deal got killed.
Also, employers actually like this system. If youâre a big employer, offering health benefits is an excellent talent recruiting and retention tactic. You probably know people who keep a job purely for the health insurance. A lot of employers are quite pro-status quo.
Turner Novak:
Yeah. The classic startup founder setup. Youâve got your startup, not making money, taking on tons of risk. And then you have a partner who works at a big company with a nice salary and really good benefits. Two-pronged approach.
Nikhil Krishnan:
If anyoneâs listening, the startup idea is a dating app between founders and people with Fortune 500 benefits.
Turner Novak:
Thatâs a big opportunity. So how does money actually flow in US healthcare? How does it all work?
Nikhil Krishnan:
This is a five-hour podcast. Thereâs not enough time. But at a high level: most people get health insurance through their employer. Your employer picks a handful of plans, you pay a portion of the premiums, the employer subsidizes the rest. When you go see a doctor, the insurance company pays the majority of the bill, plus most of your prescription drugs.
But if I were to draw out the entire flow, it would look like the Charlie conspiracy meme. And itâs very different depending on whether youâre on a government plan like Medicare or Medicaid versus a private plan. Each has a completely different set of rules and payment methodology.
Turner Novak:
You mentioned something about a two-way price negotiation. What is that for someone whoâs never heard of it?
Nikhil Krishnan:
This isnât unique to healthcare. It exists in a lot of industries. The general idea is that a lot of value in healthcare comes from being large enough to negotiate against a counterpart. Hospitals want to get bigger to negotiate with insurance companies. Insurance companies want to get bigger to negotiate with hospitals or pharma. Everyoneâs trying to build leverage.
So you end up with a lot of entities that say, âWe aggregate a bunch of people together and negotiate on their behalf.â If youâre a newcomer or an individual, you probably want to join one of these aggregate pools. An explicit example: if youâre an insurance company, you probably work with a pharmacy benefits manager who aggregates all the patient lives that different insurers represent to then negotiate with pharma companies on drug prices. Better to negotiate together than separately.
The problem is that a lot of those aggregating entities, who are supposed to be negotiating against someone on your behalf, go to the very companies theyâre negotiating against and say, âIf we choose you, we want a cut.â They have all the leverage. This shows up everywhere.
Say youâre buying health insurance individually and you use a broker. The broker offers the service for free but gets paid commissions from the insurance carrier. So the person whoâs supposed to be helping you pick the best plan is actually getting paid by the carrier. They may only recommend the plans with the best commission structure.
This is true in any industry where a service is âfreeâ for you but paid by the seller. Real estate, financial services, whatever. You think youâre not paying, but you are, in the form of not getting the full range of options.
Turner Novak:
Yeah, the broker thing is classic in finance more broadly. A financial advisor recommending mutual funds where they get paid 8% of the deal value upfront. Theyâre going to recommend whatever makes them the most money.
Nikhil Krishnan:
Of course. Or getting an apartment or buying a house. Brokers are involved and you think youâre not paying, but itâs baked into the selling fee. Youâre paying; itâs just convoluted.
Turner Novak:
My favorite hack for buying residential real estate. Iâve done this twice. The realtor always gets 3%, one from the buyer, one from the seller. I donât have a realtor. I just find a house I want and say, âI want to buy this. You can be my realtor. Just tell me what I need to pay and I want to win this.â The listing agent gets double commission. So youâll win, theyâll push your offer harder.
Nikhil Krishnan:
And yeah, there are lots of ways to game the broker stuff. But at the end of the day, itâs just a bad transaction mechanism for everybody.
Especially when you think about AI tools. This is a great use case. A broker exists to reduce information asymmetry between you and the seller. Buying a house, buying health insurance, whatever. âThis is complicated, I wish someone would guide me through it.â But now with AI, you can have a copilot that doesnât have those same incentive problems. Much more straightforward.
Turner Novak:
Maybe thatâs an interesting segue into AI and healthcare more broadly. Whatâs actually getting adopted and actually working right now?
Nikhil Krishnan:
Most of the stuff thatâs actually getting adopted touches the payment rails, whatâs called revenue cycle management in healthcare. Thatâs the process by which a provider takes medical documentation, turns it into a bill, sends it to the insurance company to get reimbursed, and then the insurance company pays them, argues with them, or whatever. Itâs incredibly convoluted. Itâs a lot of transforming text from one format into another, so itâs a good LLM use case.
But simultaneously, it doesnât solve the core problem, which is that payers and doctors fundamentally donât want to pay each other. Itâs a Tom and Jerry fight. You can add all the tech you want to it, but theyâre still fighting. Now you have doctors using AI bots to call payers, and payers creating AI bots to stop the doctor bots. We kind of end up back in the same place.
That said, itâs an area where you can demonstrate ROI very quickly. You go to a hospital and say, âYour accounts receivable are way lower than they should be. Thereâs uncaptured revenue, there are codes you should be adding, and weâll get you more money.â Then you go to the payer and say, âYouâre overpaying for certain things, and our bots will prevent over-billing.â AI scribing is part of this too, recording a doctorâs visit and transforming the audio into documentation or billing codes.
But Iâm not as personally interested in that space. I think thereâs a much bigger opportunity in totally rethinking how we deliver care from the ground up. If you were to design a doctorâs office from scratch with AI at the core, what would it look like? Probably very different. The intake form would be dynamic, risk-assessing you, routing you to telemedicine if itâs not serious, or pushing you to an in-person visit only if it is.
Thereâs a company called Doctronic in Utah thatâs testing fully autonomous prescription refills. If youâre on a low-risk medication and just need a refill, the AI does it. Rewrites the script for you. Thatâs an example of where things are heading: AI autonomously doing tasks, which lets you rethink the whole workflow rather than just speeding up a human-in-the-loop process.
Turner Novak:
With something like Doctronic, youâre talking about providing actual care. Donât you need licenses? There are probably regulations around that. How does that intersect?
Nikhil Krishnan:
Yeah. One of the open questions is where liability applies if the AI system messes up. One way to work around this, and I donât have confirmation but I assume this is how Doctronic is operating, is that a doctor essentially grants the AI access under their license. They absorb the liability, but also get the financial upside. If a doctor wants to see 10x more patients and AI enables that, they take on more risk but also capture more revenue.
It then becomes the doctorâs job to evaluate the AI system and make sure itâs doing what they want. The bigger open question: should AI systems be regulated as medical devices, or more like nurse practitioners who work under a physicianâs supervision? Thatâs the debate, and itâs not settled.
Turner Novak:
Interesting. What do you think is the right framework? Because when I think about any doctorâs visit Iâve had, thereâs care staff, nurses, who do all the prep work, then the doctor comes in for a couple minutes, makes a decision based on the information collected, and then the nurse comes back to do the rest. Software could technically speed that process up, right?
Nikhil Krishnan:
Not only speed it up. You could also have a more expansive visit. If I can talk to an AI for an hour before seeing a doctor, Iâm getting more information into the system. And I can interact with it throughout my life, not just during the visit. You get very different data that way.
I donât think thereâs one right answer. One benefit of having 50 states is that different places are experimenting with different approaches. Some are running sandbox programs, try it, see what happens. Others are saying the downside risk is too high. Areas with more acute healthcare needs will probably be more willing to try AI-driven care delivery.
My sense is there will be shifts in how we think about malpractice and liability insurance for software. And there may be a model where AI has to demonstrate a baseline level of competence through an FDA-like process, and then at the deployment level it lives under the supervision of a doctor or practice.
Turner Novak:
Yeah. And youâve mentioned before that you think doctors are kind of screwed in the current healthcare system. Youâd think itâs one of the highest-status jobs in the world. So whatâs going on?
Nikhil Krishnan:
To be clear, âscrewedâ is relative. In the grand scheme of things, they still have very secure jobs that pay quite well.
Turner Novak:
Yeah, Iâm clip farming here. Iâm trying to get some clips.
Nikhil Krishnan:
Of course. I think the issue doctors run into is theyâre essentially on a productivity hamster wheel. You go work for a hospital, the hospital controls your time and what youâre allowed to do. Youâre put on a treadmill of 15 to 30-minute visits, banging them out, hitting productivity metrics to earn your pay.
Turner Novak:
So there are quotas in the backend, like you have to see this many patients?
Nikhil Krishnan:
âQuotasâ isnât quite the right word. Thereâs a system called RVUs, relative value units, which are scores that determine how you get reimbursed. The score factors in how complex the procedure is, the liability risk, the equipment required, etc. Itâs a scoring methodology set by the government for Medicare payments, though it gets adapted for different contexts.
As a doctor, you can get bonuses based on hitting certain productivity levels. Youâre expected to do a mix of services that hits a certain threshold. So itâs not as blunt as âyou get paid per prescription.â But you are incentivized to do a mix of things that are most productive for whoever employs you. Someone else is making a lot of those decisions on your behalf, while the upside goes to the hospital rather than to you.
Thatâs combined with the fact that the social status of being a doctor has declined quite a bit. People are generally more skeptical of the healthcare apparatus. Patients come in more bitter, and theyâre often more complex medical cases. Doctors are building fewer long-term relationships with patients. Care is more episodic. The social connection that used to exist between doctors and patients has eroded.
Turner Novak:
Yeah. Iâve definitely had doctors walk in, glance at the chart, and be like, âHey... Turner, right? Howâs it going?â
Nikhil Krishnan:
Yeah, thatâs my parents. Theyâre doctors. They come in and itâs like, âHey, Nikhil, right?â
Turner Novak:
Are you a nephew or a son? I canât remember.
Nikhil Krishnan:
Iâm an only child. Iâm the only one.
But yeah, the system doesnât encourage long-term relationships anymore. Itâs a bit sad.
Turner Novak:
So procedures, is that ultimately where the profitability comes from for most health systems?
Nikhil Krishnan:
Yeah, doing procedures. But there are also newer ways hospitals make money through more convoluted schemes.
For example, there are government programs where hospitals serving mostly lower-resource patients can acquire drugs at a much lower cost through a program called 340B, and then get reimbursed at normal insurance rates. So youâre acquiring drugs cheaply and getting paid at full rates. Thereâs a huge spread in there.
And I donât think itâs malicious. Itâs more that we havenât figured out how to structure payment for care in lower-resourced areas. Weâre still debating who should foot that bill.
Turner Novak:
It makes sense though, right? If youâre choosing between building a hospital in a wealthy area where people spend a lot versus the lowest-income area in the country, youâre going to build where the money is. So you need some incentive to make care exist where people canât pay full price.
Nikhil Krishnan:
Of course. You need some program saying, âDonât just cherry-pick healthy, easy patients.â But itâs a tightrope. If you make the subsidies too profitable, people will find ways to game that too. So itâs a bit of a lose-lose situation no matter how you set it up.
Turner Novak:
Is there anything about the healthcare system that surprises most people? Like a dinner party trick, you drop it and it blows peopleâs minds?
Nikhil Krishnan:
I think a lot of people donât understand how large healthcare is as an employer. Itâs an implicit jobs program in the US. Manufacturing was essentially replaced by healthcare as the backbone jobs program, and healthcare is now the number one employer in this country and in most states.
So if we want to see the productivity gains from tech and AI, we also have to figure out what the next jobs program looks like. Something that people without advanced degrees can earn a meaningful income from. Right now thatâs healthcare. What it looks like after is not clear.
Another thing that surprises people: what percent of total US healthcare spending do you think goes to pharma?
Turner Novak:
I feel like itâs either really high or really low. Iâll guess 62%.
Nikhil Krishnan:
Itâs somewhere between 9 and 11%. Way smaller than most people assume. I think it feels bigger because expensive drugs get a lot of press, and because a lot of people canât afford them. Those stories are real and worth taking seriously. But relative to everything else we spend in healthcare, pharma is actually a small slice.
Turner Novak:
I saw a stat once, Iâll probably butcher it, but something like 40% of all cable news ad revenue comes from pharma. Which is a wild number.
Nikhil Krishnan:
The Skyrizi jingle is permanently burned into my brain.
Turner Novak:
Itâs a funny juxtaposition. Youâre watching a basketball game, peak physical specimens on the court, and then thereâs an ad for some drug with a list of terrifying side effects.
Nikhil Krishnan:
Itâs kind of interesting. I was debating this with someone recently. Of all the things we want people to consume more of, drug products are actually not that bad on the scale of things.
Turner Novak:
Oh really? How so?
Nikhil Krishnan:
I mean, itâs a debate. But getting more people into sports betting or buying certain other products is probably more net negative than some pharma advertising. There are a lot of people who are underdiagnosed with heart disease who might actually go see a doctor if a condition is put in front of them often enough.
Direct-to-consumer pharma advertising is bad in a lot of ways and itâs easy to make that case. But I think a lot of the frustration with pharma ads is really frustration with consumption culture in general in the US. Relative to other categories that get heavy ad spend, pharma might not be the worst.
Turner Novak:
Yeah. And the effect of one of these drugs might be that it cures your diabetes or meaningfully improves your life. Meanwhile you look at an alcohol ad, all parties and fun, no mention of the downsides, and alcohol can actually be really destructive.
Nikhil Krishnan:
Itâs kind of crazy to me that pharma companies have to list all their side effects in an ad, and alcohol companies donât.
The pharma party sounds way less fun because theyâre telling you about all the potential complications. The alcohol ad makes it look like youâll be cool and having a great time.
Turner Novak:
I might literally die from this drug. But from the alcohol? Iâll be hanging out with all the cool people having a blast.
Nikhil Krishnan:
Exactly.
Turner Novak:
So, talking about AI in healthcare. Are there areas where the attention is a little out of whack with actual effectiveness? Things that are overhyped?
Nikhil Krishnan:
Yeah. A lot of the stuff thatâs just speeding up the revenue cycle process. I donât think making how you submit a bill to insurance faster is going to change that much. The payers on the other side will just build their own systems to slow things back down. The friction is kind of the point of the system in some ways.
Iâm not excited about that space. Thereâs a lot of money going into it, but itâs not durable, and the vendors are pretty interchangeable. You could swap one out and nobody would notice. So I think that gets a lot more attention than it deserves.
Turner Novak:
Are there areas that sound like great startup ideas or healthcare business opportunities but almost never end up working? Classic traps?
Nikhil Krishnan:
So many. Medical tourism marketplaces come up constantly. âItâs all cash pay, people go abroad, what if we just built a marketplace for it?â But the good hospitals abroad already have their own programs and are pretty much full. The rest are often not great. And US health insurance doesnât cover complications if you get medical tourism treatment and come back. I hear this one all the time and I never think itâs going to work.
Another one is âWeâll analyze your medical bill, find errors, and take a percentage of savings, and it costs you nothing.â That sounds great. And I actually think you can build a small business around it, especially with AI tools today. But the real problem is that it requires fighting with the hospital, which often just doesnât play ball. It has nothing to do with the tech. And youâre catching a patient at a very specific moment in their journey. Really hard to acquire them, and theyâll only use you once. Very lumpy, non-recurring revenue.
Clinical trial patient recruitment is another one I hear constantly. These can be fine ideas, but theyâre often not defensible and arenât really venture-scale businesses. You can probably build a good services business doing it, but everyone wants to raise $10 million and try to build something huge. Thatâs just the wrong capital structure for the type of business it is.
Turner Novak:
It keeps coming back to profitable acquisition and retention of customers. Which is why primary care, even though it doesnât make money on its own, is valuable for health systems. Itâs how you acquire people for the higher-margin procedures.
Nikhil Krishnan:
Yeah. And even if a transaction is one-time, you need people to keep coming back for different things. Another problem is insurance-tied churn. You offer a service covered by someoneâs plan, they change jobs and switch insurance, and suddenly itâs not covered. Built-in churn.
And just remember, most people arenât using the healthcare system on a recurring basis. They use it relatively infrequently. So building a business with predictable, recurring revenue in that paradigm is really hard.
Turner Novak:
Yeah. It seems like the answer is always âsell to the employer,â they pay the bills for individuals. Or maybe the insurance company. But Iâm not sure how that whole ladder works.
Nikhil Krishnan:
Every level has different trade-offs. If youâre selling to employers, most of them use benefits consultants or brokers to evaluate vendors. So you have to go to the brokers, but the brokers want their cut, have their own preferred vendors, and their own commission structures. Now youâre selling to the broker who sells to the employer. And then you still have to get the actual employees to use your thing. Thatâs three different sales in one.
Turner Novak:
And then you might think, âThere are hundreds of brokers. Should I skip all that and go straight to the insurer?â
Nikhil Krishnan:
Then the insurer says, âGreat, whereâs your data?â And theyâll probably send you to one of their employer clients to pilot. And youâre basically back at square one.
Turner Novak:
Fair. So whatâs happening with the consumerization of healthcare? Is that the movement toward going directly to the individual patient?
Nikhil Krishnan:
Yeah. Weâve swung toward a pretty paternalistic healthcare system, and now thereâs a swing back toward something more consumer-oriented. AI tools play a big role. People have much more expertise in their pocket now to make healthcare decisions. Thereâs also more push to make it easier for patients to access their own health records and get context around their data.
I think people should have more agency in their healthcare decisions, including being able to take on more risk in exchange for cheaper or faster services. Like, âI want this AI to prescribe me this drug. It might get it wrong sometimes, but Iâll get a cheaper, faster service.â Thatâs generally where things are trending.
But there are downsides too. Dental is a good analog. Itâs extremely consumer-oriented, schedulable, price-transparent. But you still get constantly upsold. There are bad actors who tell you that you have cavities you may not actually have. All the pros and cons of consumerization apply.
How far do we want healthcare to move in that direction? Probably further than we are today. But you already saw the GLP-1 situation play out. Thereâs a whole spectrum of compounded weight-loss drug providers where itâs basically, âCome get your drug here, donât ask how we make it, just trust us.â
Turner Novak:
Yeah. âItâs super cheap. Donât ask why.â
Nikhil Krishnan:
Exactly. Thatâs probably too far on the consumer end. So whereâs the middle? Thatâs the real question.
Turner Novak:
The dentist thing reminds me. Every time you go to a new dentist, they tell you that you have a ton of cavities.
Nikhil Krishnan:
Of course. But now maybe I can take that same X-ray image, run it through an AI tool, and walk into the dentist as a much more informed consumer. Thatâs probably a net good for everybody.
Turner Novak:
So with people able to just ask Claude or ChatGPT healthcare questions, are the big AI companies just winning consumer healthcare? Or what happens?
Nikhil Krishnan:
A few things. Liability is still an unanswered question. If the AI gives a wrong diagnosis and something goes wrong, whoâs responsible? I think itâll tilt toward doctors or the medical establishment absorbing that liability for now. We donât have good frameworks for standalone LLM liability, so LLMs will be limited in what they can do autonomously. Prescribing a drug, for example, is a hard next step for them.
The thing is, youâll eventually hit a point where the AI can get you so far, but the next action requires the healthcare system: getting labs done, getting a drug prescribed, having someone physically examine you. LLMs alone canât close that loop. So people who can offer those real-world services have an advantage.
Where LLMs are already great is interpretability. People upload their lab results to get explanations of anomalous values, for example. But to generate net new data or trigger net new actions in the world, thereâs still no mechanism for them to do that alone.
Turner Novak:
Thereâs always a break when a real-world action is necessary. Someone touching you, something being sent to you, a physical thing happening. That will probably always require a separate product.
Nikhil Krishnan:
Probably. And it should be under the supervision of a doctor in some capacity. Whether AI is doing the ordering under that supervision is fine, but someone needs to be the orchestrator. And someone needs to be able to escalate when something needs human judgment.
Turner Novak:
One area I keep hearing about is peptides. Whatâs happening there, and what do you think plays out over the next few years?
Nikhil Krishnan:
Oh, God. At a high level, a peptide is an amino acid chain. But the way people talk about peptides today, theyâre basically unregulated pharmaceutical products. Normally when a drug comes to market, it goes through a rigorous clinical trial process to prove safety and efficacy. That takes a long time, costs a lot, and the end product is expensive.
Turner Novak:
But the result is: âThis is safe to take. Weâve proven it.â
Nikhil Krishnan:
Thatâs what the test is supposed to establish. The alternative gaining traction now is: âWe should be able to access these compounds faster, take the risk ourselves, and test them on our own terms rather than waiting for a years-long approval process.â There are a lot of products people swear by, claiming to repair muscles, work better than Ozempic, extend lifespan, whatever.
The range is wild. Some claim to treat ADHD. Itâs basically impossible to verify because you can say almost anything without clinical data behind it. Thatâs the fundamental problem with this space.
Turner Novak:
Are they mostly injectables?
Nikhil Krishnan:
No, not all of them. There are ingestibles too. But introducing an unproven product systemically through your body, whether orally or by injection, carries different risk profiles. And since none of this is well-studied, itâs hard to know what those risks actually are.
This goes back to what I said earlier: people want agency over their healthcare decisions. This is one expression of that. Some of these compounds will probably cause serious side effects. There are efforts to standardize how theyâre procured and manufactured, but sourcing has been pretty shady.
Personally, I wouldnât do it. But Iâm not judging anyone who does. If youâre in chronic pain and thereâs been no good solution for you, and someone says this peptide might help with that knee pain, I understand why youâd explore it. I do think people should think carefully about the risk-reward theyâre making, especially if theyâre young and relatively healthy. But I understand why this trend is happening.
Turner Novak:
Yeah. I feel similarly. New options are generally good. You can make a choice. But there may always be unknown side effects we donât understand yet. It took 30 years for cigarette companies to admit the product kills you.
Nikhil Krishnan:
Totally. Itâs funny to think about the FDAâs framework here. For food, you can bring anything to market as long as itâs âgenerally recognized as safe.â For drugs, the bar is much higher, even though a specific drug affects a far smaller slice of the population than food does. Peptides arguably get regulated more like food: âThis seems safe-ish, letâs see what happens.â
Thereâs a real philosophical question about how much evidence we need before allowing something to market. We currently test safety, then efficacy, then validate both at larger population sizes. But maybe after proving safety, you just bring it to market. Let high-risk-tolerance people try it early, and let more risk-averse people wait for efficacy data to build up. We have randomized controlled trials for a reason, to avoid self-selection bias. But some people argue the time it takes to get a drug approved does more harm than good. Everything is trade-offs.
Turner Novak:
Yeah. Itâs like McDonaldâs. It wonât necessarily kill you, you can eat it every day if you want, probably not the best idea, but thatâs your call.
Nikhil Krishnan:
Exactly. You want to try the mystery Doritos flavor? You want a Baja Blast? Salute. Thatâs all you.
Turner Novak:
Do you know why all medical software is just so bad? You go to a hospital and the screen looks like it was designed in 1994.
Nikhil Krishnan:
Youâve got an MRI machine running cutting-edge physics on MS-DOS. How can these two things be in the same room?
A few reasons. First: all enterprise software sucks. This isnât just healthcare. Have you ever used SAP Concur?
Turner Novak:
Iâve used SAP. Itâs brutal.
Nikhil Krishnan:
It sucks. Any enterprise software sucks. And healthcare is very concentrated in large enterprises, so you see more of it.
Second, healthcare is rife with a principal-agent problem. The users of tools are rarely the people paying for them. The doctors and nurses using the software every day are not the hospital administrators buying it. Usability is probably much lower on the adminâs priority list. So you get software that isnât designed for the actual user.
My hope is that with new AI tools, the user interface layer changes dramatically. Your front-end experience can look completely different even if the underlying database or enterprise system stays the same. What the user interacts with doesnât have to look like the legacy system underneath. Iâm hopeful, but TBD.
Turner Novak:
What works best when navigating the sales cycle with a health system? For someone thinking about selling into healthcare?
Nikhil Krishnan:
Step one: donât do it. But if youâre going to anyway, a few things.
ROI timelines have to be much shorter. Within a year, you have to demonstrate clear ROI. A common mistake is assuming hospitals have budget for software. They generally donât. Software budgets are small and not ROI-producing. So you typically have to target a service line that has headcount budget or active financial transactions happening. Much easier to prove ROI when you can point to labor costs or payment flows.
If you say âweâll increase productivity per worker,â thatâs too roundabout. Hospitals want a direct line of sight to the return. Much easier to say: âIf we implement this correctly, you donât need to add this many new hires this year.â Then you need a champion. Someone who will walk you through the many committees it takes to get approval. Every health system is structured differently in terms of who owns budget and who makes decisions. Find the person whoâs going to run with you through all of it.
One newer motion Iâve been seeing: bottoms-up adoption in healthcare, which has historically been rare. OpenEvidence is a good example. Essentially ChatGPT for doctors. Get it in the hands of clinicians who love using it, and that usage creates leverage when talking to hospital leadership. A different playbook than top-down sales.
Turner Novak:
Is OpenEvidence going to hit a wall where the system shuts them out? Bottom-up sounds great until the powers that be say ânot allowed.â
Nikhil Krishnan:
Itâs a great question. Thereâs a triangle forming right now: scribe companies that record visits and turn them into documentation, literature review and clinical decision support tools like OpenEvidence, and the electronic medical record systems that hold all the historical patient data.
All three want to do what the other two do, and all three need each other to produce the best outcome. The EMR companies will try to roll out their own scribes and literature review features, but consumer product quality isnât their muscle. The scribe companies will try to expand, and the literature review tools will try to integrate deeper into workflows. Itâs a bit like that meme where everyone has a gun pointed at everyone else. That fight is going to be interesting to watch.
Turner Novak:
Who do you think wins? Does the system of record probably win because itâs so deeply embedded?
Nikhil Krishnan:
Itâs a good question, and itâll probably play out faster in other industries first. Think about Salesforce. Are they going to launch their own agentic tools? Just become MCP servers that charge a toll on every interaction? The healthcare version of that fight is still being figured out.
There are also major lawsuits right now around companies using agents to log in as humans and extract data from EMRs. The EMRs say itâs breaking terms of service. The startups say their users want this and the EMRs are blocking them from it. Not super clear how that lands.
My bigger concern is the consolidation dynamic. A large hospital is not switching its EMR overnight. Itâs not going to happen. So the system of record has enormous leverage. The more fundamental question is: why arenât there more new entrants? Private practices, small pharmacies, smaller insurers. If you had more competition at the small and medium business level, youâd see more experimentation with new operating systems and systems of record.
My hope is that AI-native tools make it more attractive for doctors to go independent. Imagine a product that says: âCome start your own private practice. Weâll build all the agents for you. You can see 10x more patients and capture the financial upside.â Thatâs a totally different operating system from what exists today.
It doesnât solve everything. Where do you get patients? If youâre a new insurance company, you need risk capital. But if you make it more attractive for people to start small, you chip away at the enterprise software problem over time.
Turner Novak:
It almost sounds like Shopify for doctors. Just enabling them to start their own business.
Nikhil Krishnan:
Essentially. And a million people have tried to build that company. âBusiness in a box, start your own practice.â The problem has always been that it doesnât solve the harder parts, like getting patients.
But if patient demands change, like âI really want a doctor who uses AI to refill my medications faster,â or âI want a doctor who incorporates my wearable data into my care,â then maybe the acquisition problem gets easier too. Physician independence is one of the most important things we have to solve in US healthcare. There are a lot of blockers, but there are a lot of moving pieces that could help unlock it.
Turner Novak:
Youâd probably need to build a full stack. Not just the system of record but also customer acquisition. With Shopify, that was solved through Facebook ads and content. I think you actually see this happening in therapy. Therapists building social media followings to bring in local patients, and then a telehealth back office to serve them. Maybe COVID just made it possible sooner in that space.
Nikhil Krishnan:
Itâs very cash-pay oriented, and thatâs a big part of why it works. In e-commerce, youâre paying out of pocket. With insurance reimbursement, new practices get terrible rates out of the gate. Really hard to make the economics work. But in cash-pay areas like therapy or longevity medicine, thatâs not necessarily the case.
Longevity docs are a great example. Strong social media presences, strong opinions on how care should be delivered, cash-pay model. The question is how you solve it for the insurance-covered side. Because not everyoneâs going to pay cash.
Thereâs a movement called direct primary care thatâs trying to answer this. Essentially cash-pay primary care where you pay a monthly subscription to a primary care physician and get a defined set of services. 24/7 texting, wholesale-priced medications, whatever the practice offers. The idea is to bring costs down enough that people choose it over or alongside insurance.
Turner Novak:
I feel like that ties back to physician independence. Enabling more people to start their own thing and provide care on their own terms.
Nikhil Krishnan:
Yeah. I really feel for doctors entering practice now. They come out with a lot of debt, feel like they donât have much agency, and often arenât practicing the way theyâd want to.
What surprises me, and this is anecdotal, is how many residents I meet who donât want to practice full-time. Theyâve gone through all that training, theyâre highly skilled, and theyâre asking me about health tech product jobs. Something has gone wrong if people who have done all that work are jaded enough to want out. Itâs a sad state of things.
Turner Novak:
Is it that the process of becoming a doctor is just so hard and long that people burn out? Or is it something else?
Nikhil Krishnan:
I think for some people it might be that. But more broadly, I think itâs that they look at hospital work and it doesnât bring the fulfillment they expected, or they want to be doing more interesting things. Thereâs so much happening in AI and tech, and they want to be part of it.
And thatâs great. But maybe there should also be cool, interesting things happening within the traditional medical career that make people excited to stay. Something has to make practicing medicine feel worthwhile again.
Obviously the group coming to me is self-selecting. Probably the most jaded or the most interested in health tech specifically. But I think part of it is just that they donât want medicine to be the only thing in their life, and it doesnât always deliver the level of meaning they thought it would.
Turner Novak:
Thatâs fair. So letâs shift. How did Out of Pocket get started? Whatâs the origin story?
Nikhil Krishnan:
I was at a company called CB Insights, running the research team focused on healthcare. Thatâs where I started reading deeply and started writing. Then I joined a clinical trial startup for a bit.
Turner Novak:
Oh, I forgot about that.
Nikhil Krishnan:
Yeah. I was an operator for a hot second and wasnât very good at it.
Turner Novak:
Failed operator.
Nikhil Krishnan:
Failed operator. I left in February 2020 and started writing again because I wanted to rebuild that muscle. The original idea was actually to make The Daily Show for healthcare. I was really inspired by Jon Stewart. Making boring topics genuinely interesting and entertaining. Could I do that for healthcare?
Turns out, video is really hard. We filmed some sketches and I was like, âI cannot do this consistently as a business.â
Turner Novak:
Didnât you post some of these? I feel like Iâve seen stuff over the years.
Nikhil Krishnan:
Yeah, I ended up posting a few because the footage was just sitting there. But making it a real business, doing it consistently. I have so much respect for video people now. I was just better at writing anyway, so I kept writing. And the Substack economy was starting to form around that time.
Turner Novak:
You probably got an insane boost from starting at the beginning of COVID. Everyone was home with nothing to do but read and subscribe to things.
Nikhil Krishnan:
Yeah. People were locked in their houses, healthcare was suddenly the only topic anyone could think about, and it worked in my favor. But itâs been six years since then and itâs evolved a lot.
Weâre now all-in on in-person stuff. I still do the newsletter and virtual courses, but content on the internet is changing rapidly in ways I donât love. Standing out in a sea of AI slop feels like an uphill battle. So weâve leaned hard into events. Hackathons, micro-conferences focused on builders sharing across companies, healthcare coworking spaces. I view content more as a bat signal now: âHereâs a topic I think is interesting. Letâs all coalesce around it and meet somewhere.â
Turner Novak:
So itâs more like top-of-funnel.
Nikhil Krishnan:
Yeah, top of funnel. And also a self-selection mechanism. I want people who can read to come to our things.
Turner Novak:
If youâre optimized for short-form, your brain short-circuits trying to get through a long newsletter.
Nikhil Krishnan:
You know whatâs funny? There are a lot of content creators who make content for people theyâd never want to meet. You make content for an audience you can monetize well, but youâd hate spending time with them.
I could never do that. For me, the fun is putting content out there to find my people, even if that group is much smaller. Content is not the be-all end-all. Itâs more like: let me find the people I actually want to hang out with and do interesting things with.
Turner Novak:
Iâve kind of accidentally figured that out too. My content is either really dumb and jokey, or itâs these long dialed-in conversations where Iâm just trying to learn something. And most of the people I meet through it, I end up thinking, âOh, that personâs actually way cooler than I expected.â So yeah, thatâs a good framing.
Nikhil Krishnan:
Yeah. If meeting people is literally part of your job, you naturally gravitate toward making content that attracts people youâd actually want to meet. Some creators view content as the end product. Make it, monetize it, donât interact with the audience. Thatâs a totally valid business. Itâs just not who I am.
Turner Novak:
So whatâs the business model for Out of Pocket? Revenue streams, investing. What does the whole system look like?
Nikhil Krishnan:
The newsletter has banner ads and full-page sponsorships. We co-develop courses with companies where they provide the expertise and we build the content. Those are sponsored courses. Events have sponsors and we charge tickets depending on the format.
I do some investing, but itâs typically angel investing. Honestly, investing is my least favorite hat I wear. I know itâs your whole thing, but I donât have the delayed gratification you must have. I mostly do it to vouch for people. Putting a little money behind someone is a way of sharing credibility with them in the space.
We have healthcare coworking spaces, a Slack community with some monetization, onsite training and workshops. Itâs a hodgepodge of side quests under a bigger umbrella. Purposefully diversified so weâre not too concentrated on any one revenue stream, but weâre definitely all-in on events.
Turner Novak:
On the courses. Are most of them free?
Nikhil Krishnan:
Some are paid. I teach a Healthcare 101 course where people pay directly. Mostly employers signing up their employees to get up to speed. The rest of the courses are usually free for attendees; the sponsor pays us to co-develop and co-market them.
Weâve also been doing onsite workshops with companies. Going in to teach things like claims data analysis, which is a very common data type that healthcare data teams work with. About 60 to 70% of the content is consistent. âHereâs what a claim is, hereâs how to do a data analysis.â And we customize the last portion for whoever weâre working with. Itâs just easier to keep peopleâs attention in person than over Zoom for this kind of training.
Weâve started doing some AI workshops too. Showing people the basics of Claude or ChatGPT or whatever their internal tool is, and then introducing more interesting ways to use it. Like connecting to an MCP for a specific vendor and showing them whatâs possible. A lot of people just need some handholding. Weâve been testing that model.
Turner Novak:
Yeah. A lot of people donât think about how large the L&D budget is at companies. Paying a vendor to help your employees get smarter is a real investment with real ROI.
Nikhil Krishnan:
Itâs a mixed bag, though. L&D money at larger companies tends to be earmarked for specific things, often tied to existing learning platforms that house all their training modules. If youâre not on that platform, you have to pay a take rate to get distribution. Or you need some recognized certification that gives the company cover to say, âYes, we used a credible provider.â
Itâs frustrating because we can offer training at a fraction of the cost of the big certified names, and I genuinely think our content is better. But if we donât have the right certification, some enterprises wonât consider us. I get it. Theyâve probably been burned before. But itâs still annoying to pay that price.
Turner Novak:
As youâve swung more toward in-person, how are you thinking about AI in your own work? How do you use it?
Nikhil Krishnan:
At a personal level, I use it constantly. Even just running the business. Plugging AI into our internal systems so I can get a better read on whatâs happening. When weâre doing an event, I can quickly look up who in our ecosystem has the most relevant background for a particular talk, or who I last reached out to and why. Context retrieval that used to take forever now takes seconds.
As someone who has to get up to speed fast on a lot of different healthcare topics, itâs incredibly useful. Iâve built enough pattern recognition to know when itâs bullshitting me, which helps.
One specific example: I read a lot of primary source documents. Lawsuits are actually a great way to understand how a specific corner of healthcare works. But lawsuits are extremely long and full of procedural stuff I donât care about. So I have a Claude workflow that extracts the parts relevant to me. Skip the legal proceedings, just give me the mechanics of whatâs interesting about the case. Thatâs made me dramatically faster at getting up to speed on new topics.
At the same time, Iâm worried about people using AI to pollute the internet commons, or to replace their own critical thinking. Every time I see someone just ask an AI, âTell me why I should care about this,â I think, what are we doing here?
Turner Novak:
Do you think AI-generated content is more widespread than people realize? Or do you think most people can tell and have formed an opinion?
Nikhil Krishnan:
Among people who post on the internet, itâs definitely more common. But we have to remember: most people donât post on the internet. Itâs a really small slice. And among the people who are consuming content? Most of them genuinely donât care if something was written by AI. A lot of people actually prefer it because itâs more digestible and in a format they like.
Thatâs probably frustrating to people like you and me who care about craft. You think, âThatâs AI slop, everyone can tell.â But not only can they not tell, a lot of people actually prefer it.
Itâs not that different from SparkNoting Shakespeare. âI canât read this. I donât understand the prose. Give me a simpler version.â Shakespeare would have hated that. But it worked for people. Is this really so different?
Turner Novak:
Imagine Shakespeare reading Gen Z slang. Some kid trying to explain Macbeth.
Nikhil Krishnan:
Yeah. Kind of the same thing.
Turner Novak:
So how should someone think about posting on the internet and building a following today? If Iâm starting from zero, followers, brand, reputation, whatâs your advice?
Nikhil Krishnan:
I just think people massively underrate reputation. Thatâs the most important thing.
Smart people care deeply about finding other smart people. Itâs not about rage-bait or raw eyeballs. For me, the most meaningful thing is when someone sends me a genuinely thoughtful email. No one else can see it, itâs just between us, and they wrote it because theyâre intellectually curious about the topic. I reply to every single one because those interactions matter more to me than vanity metrics.
You donât even need to post publicly. The question is: how do I find the people I actually want to meet and respect? Whatâs the best avenue for that? Maybe itâs online posting. Maybe itâs something else entirely. But your reputation compounds. Post as if your reputation is on the line, because it is.
Stream the full episode on YouTube, Spotify, or Apple.
Find transcripts of all other episodes here.