Super Staffing Healthcare, Codifying Compliance, & Scaling Services
52:09 Share

Upwards of 300,000 clinicians left the workforce in the year 2021. The large banks right now, up to 15% of their staff is just in compliance. You go from having 5% net margins to 30% net margins. We've just never seen the rate of technology adoption that we're seeing now ever in the history of healthcare. If you don't do this right and you don't figure out like a very repeatable motion for these smaller acquisitions, it's going to be really challenging to build a big company.

How long are we willing to wait to solve this problem? The everyday person should really care about this.

Hello everyone and welcome back to our Big Ideas series. We're sheer hours away from 2025, which will mark nearly 22 years since the Human Genome Project was completed, 21 years of Gmail, 19 years of Twitter, 15 years of Instagram, and even 17 years of the App Store, which released 500 apps when it first came out and now has millions of apps that have generated developers over a trillion dollars.

It's been a wild ride, and that ride is far from over. And that's why our partners, who meet daily with the folks who are building our future, are sharing their big ideas for 2025.

Last year, we predicted... A new age of maritime exploration. Programming medicine's final frontier. AI post-schemes that never end. Democratizing miracle drugs. And so far in this year's series, in part one, we covered the growing hardware-software intersection, and in part two, how large companies and startups are fighting for AI dominance, but also how countries are thinking about its impact on their sovereignty.

Today, we talk applied AI and how this platform shift changes the game in healthcare, fintech, and even opens up a new opportunity to rethink growth in traditional service industries. Now, if that wasn't enough, you can check out the full list of 50 big ideas over at a16z.com slash big ideas.

As a reminder, the content here is for informational purposes only, should not be taken as legal, business, tax, or investment advice, or be used to evaluate any investment or security, and is not directed at any investors or potential investors in any A16Z fund. Please note that A16Z and its affiliates may also maintain investments in the companies discussed in this podcast. For more details, including a link to our investments, please see a16z.com slash disclosures.

First up, AI applied to healthcare, an industry fraught with dissatisfaction. Too slow, too expensive, too clunky, but also too few staff. Healthcare is dealing with the mother of all clinical staffing crises. We're short hundreds of thousands of doctors and nurses relative to the level of rapidly growing demand for clinical services that is projected in the next five years.

That was Julie Yu. I'm a general partner on the bio and health team here at Andreessen Horowitz. So can AI in the form of specialist models realistically turn this around? Here is Julie's big idea.

My big idea is what we call super staffing for health care. And what that means is we've talked obviously about just the rampant challenges that exist in our health care industry. So much is broken about it. But really one of the sort of underlying fundamental issues that we face is this supply demand mismatch between the amount of clinical supply that we have in the forms of doctors and nurses and other clinical service providers,

versus the level of demand that we as patients represent in the industry. And it's not just that the demand itself is high. It's that the nature of the demand keeps getting more and more complex. We've talked a lot about how the chronic disease burden in America is increasing over time. The number of diseases that any individual has is growing and growing. The number of drugs that we're all taking is growing and growing. And so, again, not only is it that there's just a shortage of labor relative to the number of

humans who need it, but also the nature of the work is also getting more complex over time.

So let's dive into that. Just how short are we in this clinical staff? There's a lot of different reports and stats that have a wide range of numbers here. But I think generally speaking, when you look at, for instance, the number of doctors that we have estimated a shortage in, it's anywhere between, let's call it 60 to 100K doctors today in terms of the relative demand that we have in our system and roughly about 75K to 150K nurses in a commensurate sense.

But I think probably one of the more interesting issues to unpack here is what do we actually mean by shortage? So there's largely two schools of thought of how people define this. Simply, the number of human clinicians that we have is one way to look at it. The other way to look at it, though, is given the number of nurses and doctors we do have, are they actually fully performing to the full extent of productivity that number of laborers actually represents?

And I think it's probably at the end of the day, a combination of both. But those are two different ways that people have cut that pie. And how did we get here? What is leading to this shortage? If you zoom out the lens entirely on this highly regulated industry, there's actually a structural and regulatory constraint on the amount of supply that we allow into the industry on any given time basis. Through education, you're saying? Education is one. Certainly, the number of accredited medical schools is constrained, both in terms of actual accreditation, but also in terms of funding.

Standing up of medical schools is actually funded by federal dollars and there's just a physical limit on that. The other version of that is licensure. So once you get through your training, you have to be licensed at the state level to be able to practice medicine. And again, there's only a certain number of licensures that are approved on a given basis. You have to renew that every few years.

So there's structural constraints in the industry level that constrain the supply side. In addition to that, once you are a doctor, there's all sorts of challenges, some of which you alluded to, of the job itself that have led to constraints. One is certainly the aging population. There's a ridiculous stat that in the last couple of years, the percentage of doctors that are over the age of 60 has reached like 45 percent, which is crazy.

crazy. And so there's this huge silver tsunami dynamic of what happens when all those doctors retire. There's going to be a huge gap just purely on the basis of age. There's also the burnout issue. Just given everything that happened during the pandemic in particular, we put so much burden on our frontline staff and what we were asking of them to do was obviously very challenging. So the

Doctors and nurses have left the workforce actually in droves because of that burnout issue. There's lots of surveys that show that upwards of 80% of doctors and nurses will say the number one issue in their lives is burnout due to their jobs. And again, another ridiculous stat, upwards of 300,000...

clinicians left the workforce in the year 2021. This is in America specifically? In America alone. And even in the last couple of years, about 7% of the country's total active physician workforce left their day jobs as being clinicians.

Even if we're able to train like 100% of the capacity of our medical system, of our licensure system, the folks are just not sticking around because the job itself has so much overhead. And the last thing I'll say is that doctors are increasingly taking on more and more administrative functions and academic functions. So another amazing stat is I used to live in Boston. Boston is actually the highest density of physicians in the entire country, but they have the worst access issues.

because the vast majority of doctors there are working in academic environments. So, you know, more than half of their time might actually be used on research, not on actual clinical activity. So I can go on and on, but these are some of the reasons why, even if you were to look at the number of doctors and nurses, there's a lot of other factors that play into the actual capacity that's represented in that pool of labor. Maybe we can also just quickly speak to what the patient feels. Is it longer wait times at the hospital? Is it more expensive procedures? Is it

not being able to even get health care in some cases. Can you speak to the second, third order effects that ladder down to the everyday person? I think the most obvious issue is wait times. And there's studies that show that the average wait time for, let's say, a specialist appointment across the country is like 50 days. And it can range from 27 days on the bottom end and up to 90 days on the high end, and even worse for certain subspecialties.

And so that alone is obviously an issue. The other related issue is that even when you're able to book an appointment, let's say a month out, there's data that shows that after the second week, like after 14 days, basically the no-show rate goes way up, right? Because your likelihood of actually committing to that date and not canceling or not just showing up goes through the roof. And so one of the exacerbating issues is that these slots are booked and therefore the doctor is saving that time.

But a lot of that just goes unused simply because the patients aren't showing up because they've waited so long for those appointments. To your second point, if I have an issue and I need to see a doctor immediately, the longer I wait, the longer the chances is that it gets worse, right? I might end up in the emergency room, which obviously can be 10, if not 100 times more expensive to our healthcare system than that initial physician visit.

Another sort of issue, especially just given how much is happening around the consumer health care landscape these days. The other challenge is that the less easy it is for you to see an actual doctor, the more likely that you're going to go online and try to self-diagnose and get advice from folks who might not be medically qualified. Exactly. And so those are some of the major issues. But I think the wait time piece is the underlying fundamental challenge that leads to all these negative outcomes.

Yeah. And I mean, you just spoke to this idea of Dr. Google, and that's where technology maybe is not playing the right role. But it sounds like technology maybe can play a better role in fixing this whole thing. And I know there's many different facets to this, but I just love to hear your take on that. Is it realistic that technology can actually come in and curb some of these issues?

I mean, I think the way to think about it is absent technology, how would we solve this problem? And I think it boils down to like, how long are we willing to wait to solve this problem through just simply growing the supply side of our market? And also at what cost to take those seven to 10 years to like fully train a doctor is not a cheap endeavor. And so that's really where I see the opportunity for technology is to say, listen, what if we just assume that there are a fixed number of clinicians in our system? And if we just have to go with the number of doctors and nurses that we have today, how can we increase supply?

supply and capacity. Well, there's two ways, and this is where this super staffing concept comes in, is how do we super staff by enabling every individual provider to be able to be more productive in their day-to-day work? So that could come in the form of things like co-pilots.

How can I ensure that the amount of time that a given doctor is taking to make any given decision at the point of care in real time can be shrunken down such that they are just far more efficient with their decision making and that the caliber of the decision that's being made can be at parity with, again, the best of their peers on the basis of great data and great decision support. So I think co-pilots is one way that you could imagine unlocking additional capacity within the current existing pool of clinicians.

And then a second vector, which is a different form of super staffing, is how do you augment the current labor pool with autonomous agents? And largely the form that's taken today is if you were to unbundle the job of a doctor or a nurse, there is a very significant portion of what they do on a day-to-day basis that has nothing to do with patient care, that has nothing to do with the use of their unique clinical judgment. So can you take those tasks, largely administrative in nature or just communications-based, and

and unbundle those and give those to an autonomous agent to then do at much better scale, with much better accuracy, with much better reliability, such that, again, the clinicians can actually focus on what they do best. And do you have a sense of what portion that is? I know it will differ based on specialty and doctor, but are we talking that this administrative portion is like 3% or is it 30%? Do you have

Do you have a sense? Yeah. There's actually studies that look at high motion of what percentage of a clinician's time is being spent on non-clinical patient-facing care. And it can be upward to 50%. Like half of their job is mired in these clerical tasks that, again, take their time away from actual care delivery. Imagine almost doubling the capacity of a given clinical labor pool simply by applying some of these unbundling tactics. Yeah.

Yeah. And are we seeing tools like this already on the market? Where are we in the trajectory of LLMs or AI in particular entering this industry and actually reshaping it? Yes. And this is what's so exciting right now is that we are truly in like a golden era of healthcare technology adoption. And one of the dynamics that we believe is driving this rapid adoption cycle of these new AI tools is what we call this kind of leapfrog dynamic that we've talked about in the past.

where healthcare historically has been a laggard with respect to software-based technology. And we've always thought that was to the detriment of our industry and everyone else has these great workflow tools, these SaaS products.

Because we didn't do that as an industry, we don't actually have the sunk cost bias that other industries are facing right now, where they say, we just spent billions of dollars laying out these workflow tools. We trained an entire generation of our workforce to use these workflow tools. And now we're being faced with the decision, okay, gosh, do I need to spend yet another billion dollars on these new AI products, completely rip and replace this last generation of technology capabilities and retrain my workforce on a whole new paradigm of how they do their work?

versus in healthcare, where we're just simply leaping from, we can't throw enough bodies at the problem to how do we actually augment our bodies and extend through these AI capabilities. And so we believe that's driving a lot of the rapid adoption cycle that we're seeing right now. There's a few examples of this. One is, so I mentioned co-pilot versus autonomous. On the co-pilot side, there's this class of products

called ambient scribe products, where it's a way for a doctor and a patient to be having a visit. And there's an ambient scribe literally listening. So the doctor can actually make eye contact with you. They're not sitting there on their laptop. And that AI scribe is specifically tuned to be able to translate that conversation into medical documentation that is compliant, that facilitates billing in the appropriate way, that completely captures the clinical context of that

particular encounter and then can be saved directly into the electronic health record. And so that is a class of products that has really taken off. Doctor friends that I talk to are all using one of these products. It's completely transformed their lives.

They're getting back hours of the day that are no longer being spent doing manual documentation, which is a requirement of their job. Another example is a class of autonomous communication platforms where one of the biggest ways by which the health care system communicates with us is through call centers, right?

I think certainly in senior populations, you tend to see that folks want some degree of personal connection. And obviously, they're much more used to using the phone modality. And so there's companies like Hippocratic AI in our portfolio that has built an army of super staff of folks who can actually play the role of either an outbound call center agent that is trained on very specific protocols.

that are infinitely patient and can actually execute a number of tasks that otherwise would rely on humans to do. And what we're seeing is it's remarkable. Number one, most hospitals have about 10 to 20 percent of their headcount open.

at any given time. And they simply just can't find enough people to fill those roles, namely in the call center setting. And so what they're saying is, gosh, if we can't hire people fast enough, like how amazing is it that we could spin up not just one or two, but 10 or 20 of these autonomous AI agents that are actually, again, doing the same job as these humans would do.

And what's also incredibly interesting and part of the why now for why we think this trend is going to be at a tipping point is they're starting to say, OK, why can't we just pay for this out of our labor budgets instead of our IT budgets? And one of the historical constraints with health care technology has been that the

portion of the budget of, let's say, a hospital that's spent on IT has been a tiny fraction of what it is in other industries. The average is like 2% to 5% of budget goes towards IT and healthcare versus like 15% to 30% in like banking or financial services. And so now we have this huge unlock opportunity to say, rather than be constrained by that tiny IT budget, what if we tapped into the 60%, 70% labor budget and

and actually be able to create additional leverage by deploying AI in that fashion. So these are some of the fascinating trends that we're seeing already happening in market that demonstrate the sort of rapid adoption cycle of these new AI products. Is it just that there is such clear demand to fill these roles that we're seeing something different today? Or why does it seem like there has been this adoption or like jumping to participate in this AI wave when we didn't really see that in the past wave? There is just a very high burden of integration.

At the end of the day, if you're going to be introducing a new software tool into someone's workflow, it has to deeply integrate with an EHR type system or whatever other practice management system they might have. And those systems are 40, 50 years old. They don't always have great APIs. And so just the amount of work and time and customization that's necessary to get those in has been a huge barrier to adoption in the past. The second thing I would highlight is that these are just magical products.

We have a company called Ambience Healthcare in our portfolio that creates one of these Ambience Scribe tools. And when you look at the testimonial videos that they have from their doctor users, it's unbelievable. I mean, these folks are like literally in tears saying that they've been a doctor for 30 years. It's like the first time that they've actually enjoyed the job of being a doctor because it's just allowed them to focus on the patient.

The last waves of technology products that have been given to them have just increased the amount of complexity of their jobs, whereas these are doing exactly the opposite. Are there challenges that you're seeing on the ground, whether it's the tools today or the things that you see on the horizon? We were talking about health care, so it could be regulation or some other thing that you think may be actually producing friction for us to implement this.

I think the obvious thing is just the stakes are much higher, obviously, in the healthcare setting. And the hallucination issues that you see with kind of the generalist foundation models, like that is just a non-starter if you have issues like that in this particular setting. And so what we have seen and perhaps why there was maybe a little bit of lag relative to some of the generalist foundation models in terms of adoption and healthcare, but these companies that I'm describing have built specialist models.

models in healthcare. One of the issues with the Generalist Foundation models is that, I mean, not an issue, it's obviously a huge feature that they are trained on internet data, like the entirety of all publicly available human knowledge. But the reality of healthcare is that the majority of the esoteric

protocol-specific, compliance-specific information tends to not just be online. It tends to be in these proprietary systems behind the firewall, embedded into paper-based binders and things like that. And so because of that, it is necessary for these companies to build systems

specialist models that are trained on a different set of data, then they might extend some of those journalists models, but they do need to ingest a different set of information and specifically be tuned for safety. And so people do need a specialist approach to tuning the foundation models to perform appropriately in this setting.

Another one would be exactly what you said, the regulatory compliance issues associated with the use of autonomous AI in particular, and how that bumps up against FDA approval criteria. Under what circumstances do you need to get a AI tool approved by FDA or not? Do you?

Yes. In fact, one of the benefits that healthcare has had is that we are the only industry that actually has an existing regulatory framework for approving products, especially in the clinical setting. So there are already hundreds of AI radiology models or other models that sort of help automate the assessment and diagnosis of a given set of patients based on a certain area of medicine.

But how does that apply to a Gen AI model? There's a lot of work being done right now by the FDA to really extend those models to be appropriate for these far more dynamic systems that just have a lot more uncertainty than your standard ML predictive model modality, which is really what that system was designed for.

The last thing I'll say is reimbursement. There are some early signs that insurance companies are creating actual billing codes for AI, where they now say that an AI-enabled visit or an AI-enabled diagnostic can now be reimbursed differentially. And you can get credit for using these AI systems in a different way than if you were just not. And so that's really exciting to see these new paradigms persist.

But I think there's still very much a long pool until we have the ability to say, okay, if I have a fully autonomous like AI therapist, Slingshot AI, which is one of our portfolio companies, how can we ensure that those will be systematically reimbursed, at least at parity with what a human therapist might get reimbursed for? That has not yet been defined in our industry. So...

I think there's a lot of good early signs that there's a willingness by the system to embrace these new paradigms, but still a lot of wood to chop in terms of defining a comprehensive set of ways to get that in. Do you think that this has the potential to even reshape the ecosystem that already exists and not just augment it?

Yeah, I think there's a lot of really amazing things to consider, like what this unlocks in terms of both the existing clinical paths and then things that we don't even do today that are now possible to do. So one really interesting one is just part of the constraint of our current clinical supply is that we do have state by state licensure. Right. So if I become a doctor, I'm basically constrained by my geography.

And if there's a poor patient who happens to need my services in Montana and I'm not licensed there, but I do have capacity, I'm not able to serve that person. I think what this whole kind of AI world unlocks is that imagine that you had national scale capacity and capacity.

the sort of the routing challenge becomes a really unique one that actually could leverage AI, right? So you have this sort of national scale routing opportunity to just simply take advantage of latent capacity in the system at national scale. There's this new concept that actually one of our portfolio companies introduced called asynchronous medicine.

which means, okay, the typical way we think about seeing a doctor is you show up at their office, it's a 30-minute visit, it's face-to-face and live, and it's highly transactional though, right? You're in and you're out, and then months might go by or years might go by between visits. Whereas asynchronous care is something that completely flips that paradigm on its head and says, what if your doctor was in the cloud?

And any time of the day, 24-7, you could text them, you could send them a message, and they would respond. But the level of ongoing engagement that you can have, like the friction to anyone encounter with your doctor goes way down because you can text them at any time of day. And that's a new paradigm that, again, is purely enabled through these AI-powered clinic models.

where you can have these doctors who are available 24-7, who are using AI as co-pilots to enable them to be hyper-responsive to any question that comes in. And so I do think that the nature of maybe how doctors do get trained and what specialties actually exist that don't exist today that are purely available in the post-LLM world, I think a lot of those contours are very dynamically changing as we speak.

Maybe just looking to 2025 in particular, since this was your big idea for next year, is there anything that you're specifically looking out for or excited to see? We've just never seen the rate of technology adoption that we're seeing now ever in the history of healthcare IT. We've been waiting for these moments to take what we know to be the problems and we know what the solutions can look like, but there's just been, again, like this sort of a barrier between the problem space and the solutions.

And I think what we're seeing now are signs that perhaps that wall has been broken and we're at a breaking point where the industry needs these solutions. And therefore, we're going to see much higher adoption than we've ever seen before. I think it will come in the form of these super staff. I think we will increasingly see adoption of these AI tools in the form of labor and coming from completely different budgets, from completely different buyers, developers.

Julie just talked AI applied to healthcare, spawning a fundamental rewrite of buyers and budgets. But AI may also have the potential to rewrite compliance.

That was Angela Strange, general partner at A16Z. Here's Angela's big idea.

In 2025, regulation is going to become code. Regulation is one of the biggest costs to existing companies, banks and insurance companies and healthcare companies specifically, and a huge burden to new entrants coming in and to innovation.

And I think in this country, we have a system where we efficiently add regulation, but we never take away regulation. And so this problem keeps getting worse. And finally, with the advent of AI, technology can help solve this problem. If you're like me, you're probably always looking for ways to improve your morning routine. Whether that's waking up early, which is certainly a battle, eating a good breakfast, or listening to a quick and informative podcast, just like the Morning Brew Daily.

Every weekday on the Morning Brew Daily, Neil Freiman and Toby Howell break down the biggest news in business, the economy, and just about everything else, all in 30 minutes or less. So why not add it to your routine and listening queue? Follow Morning Brew Daily on Apple Podcasts, Spotify, or wherever you get your podcasts.

You mentioned the fact that we add regulation every year. Just how lofty are these regulations? If you look at just federal banking regulation, right now there are 50,000 different rules and regs across different agencies, and that's up from 10,000 in 1970. Has compliance always been this burdensome, or what's actually driving that increase?

Here's how it's done today. You've got a large thousand page PDF documentation. A software company will make a fairly brittle workflow tool and then you will have to hire many people to work in said brittle workflow tool and figure out is this compliant or not compliant. So, for instance, anti-money laundering.

work in a compliance monitoring tool, it sends up alerts if they think Stephanie is laundering money. And then somebody who works in that compliance department is going to have to do a bunch of research across a bunch of web sources and other different databases and write up a small report. Yes, no. When that goes wrong, it goes really wrong. And most recently, big headline, TD Bank was fined $3 billion for having a long backlog of these compliance alerts. If you actually break down what's

what they were failing to monitor. We're talking about tens of thousands of detection alerts, thousands of investigation cases. These are really big numbers. And to your point, the kind of things that are really hard for startups to manage. So maybe we can talk about that. How does that impact the economics, whether it's the compliance officers they need to hire or just like dealing with these fines? How does this actually work out? The large banks right now, up to 15% of their staff is just in compliance.

You went into the fourth fastest growing job in the United States in the last 20 years. Compliance officers. Surprise. I led the witness there, right? So you can't hire enough people. They're completely underwater, right? I'm sure the compliance team at TD was working very hard. So we need a technology solution. So how is this going to happen?

I believe that every white collar worker eventually is going to have a co-pilot and some of those roles will be replaced by agents. But we're talking regulation. So we're not just going to let LLMs out into the wild. We're going to use them to turbocharge very hardworking compliance people. And that's exactly where some companies are entering. So, for instance, you would get an alert saying,

We think so-and-so is money laundering. They can go do the web searches. They can check the databases. They can bring all of this information together. And now, instead of a compliance officer having to do hours of research, they can look at this report generated by their friendly compliance co-pilot. Yes, no, that's a five-minute task versus a several-hour task. Why do you think LLMs in particular are uniquely suited for this?

Because they're able to apply some form of judgment. And so let's take a very specific example near and dear to our hearts. Anytime we want to tweet something, we are a registered investment advisor. We have a great compliance team. We have to send our tweets over and they will tell us if we are compliant with the SEC marketing rule. Right. I imagine nobody listening has actually read that rule. It's 400 pages. Right.

So our choices are you could have a software that creates an if-then-this-that type decision tree. Or you can have a LLM ingest that rule, and they'll be able to make those judgment calls based on what's in that document, if my tweet is compliant or not, and then it could come back with specific modifications. Okay.

And there's companies, for instance, Norm AI is one that's doing exactly that. It is codifying specific regulation such that you can send it things and it'll come back and tell you the answer. Do you feel like from the types of products that you've seen that they are at the point where they can sufficiently interpret those gray areas that sometimes do exist in compliance?

So LLM's hallucinating, for my consumer colleagues, that's a feature. It comes out with great creative, fun, and enhancing things. In the world of compliance, that is definitively a bug. Smarter entrepreneurs are getting around this in two different ways. One, it's much more of a co-pilot versus a full agent. And then they're able to test enough edge cases around that the likelihood of hallucination is very, very, very low based on how they fine-tune the model to that specific regulation.

The other piece is a big part of compliance is being able to show that you've gone through all of the steps to be compliant. And so versus a black box coming back with, yes, we've onboarded this business in a specific way, it's going to show exactly all of the checks that you've done, all of the documents you've gotten, what you've pulled out of it. And so it's a very clear delineation of what's been done. But again, it saves the officer having to do all that work. They can just check the box.

That makes sense. And are you seeing these tools on the market? Give us some examples of what already exists. Notoriously, banks, insurance companies, even fintechs can be slow to sell into. I think there's a couple of things going on. One,

This has become such a burdensome problem that buyers are very open-minded to new tools. Two, companies of all sizes have woken up and said that AI can make a significant difference to our business, and so the sales cycles are moving much more quickly. They range from thinking of co-pilots that go along with your compliance officers. I mentioned a couple of examples there. The other thing that's happening is these older system of record tools, so take transaction monitoring.

that are built pre-AI, they used to be seen as too risky to rip and replace. And now what we're seeing is the new Gen AI first tools are 10x better versus 2x better. So for instance, Sardine is a company that's got a much more modern transaction monitoring system, and they are seeing customers rip and replace Actimize that's been around for 20 years. And that's just not something that we would have seen five years ago, but it's because it's just

that much better. I assume, as you said, many of these industries are a little slow to adopt technology. What stands out in terms of what may actually stop this adoption?

It is moving a lot more quickly than I would have thought. I've been investing in and around this space for the last 10 years, and the momentum and even just the ACVs that some early companies are getting is quite fast. I think the two things that come up with the blockers where the smart founders have gotten around are, one, how do I prevent this from hallucinating? And I think the two ways are, one, showing the work, and then, two, proving that you have fine-tuned your models close enough to this regulation that that actually doesn't happen, and your backup plan as a human does actually resolve it. Right.

But I think the most important can be wedging in at a pain point that is very acute with the right buyer. And personas that weren't sold into that often before have become very popular. Oh, like what? Chief compliance officers, like BSA officers. Like I imagine all of these potential buyers are now all of a sudden getting a lot of interest from very technical solutions. That's great. If we think...

maybe a little more long-term about opportunity here, right? Given the overhead that exists today, what are we unable to do because of all of that compliance? So think of just a consumer's life in financial services and how much better some of these new entrants have made your life. So trading now is free. I would argue that's in large part because Robinhood came along, made trading free, and put pressure on all the large brokerages. Yep.

For a large portion of the U.S. population, it was very difficult to get a free checking account. Time came along, enabled that to be possible, and put pressure on the incumbents. And so it is good for everyone, for new entrants to be able to come into this space. And one of the biggest barriers is just regulation.

And so the easier that we can make it for smart people of all resources to be able to start new companies, it's going to be better for consumers. It's going to be better for businesses. And as we looked at 2025, what are you looking for in particular? Yeah, I love two categories. One are old systems of record.

that you can do just infinitely better, and especially ones that have an extreme economic impact. So I mentioned before small business lending. Many businesses get loans from the SBA. It takes on average 90 days.

why. Part of that is there's a thousand plus page documentation about how do you make this loan compliance, right? So what is that LOS in that process? It's very much going to streamline that. And it is great for the economy for our small businesses to be able to get the credit that they need. So systems of record, point one. And then point two would be any area where a company needs to hire, you know,

dozen plus people to do a task where those people would rather do something a little bit more strategic. That can be a very interesting wedge in for a co-pilot type opportunity. And then you can start expanding and doing my task from there.

So as we look to these changes, it might be obvious for a big bank like TD how this can improve their operations. But why should the everyday person care about this? The everyday person should really care about this. And there's three main reasons. One, it's a big barrier to new innovative companies coming into the space. And I've mentioned examples like Robinhood, like Chime, like others that really upend some of the business models of the larger banks and provide free trading, free checking, just better consumer services. Yeah.

And then it's a hidden cost to a business's or a consumer's everyday life. It is longer for a business to get a loan because there's a lot of regulation and it's very difficult to comply with it. It is more difficult to get a loan modification that you desperately need to be able to pay your mortgage because somebody needs to be trained in the thousand plus any mortgage servicing guide so that they're able to decide, can you get that loan mod and how will they do it?

So improving regulation, making it code, making it easy to be compliance with will improve everyone's lives in the economy. But if you just look at the chart, it just keeps growing and growing and growing. And so this problem is going to get worse, not better. That's just federal. Every state is following exactly the same trend line.

And the opportunity for AI to disrupt how businesses operate does not stop with compliance. AI is automating jobs across traditional service industries like insurance, law, real estate, and IT. Though many of these businesses have historically been low margin and difficult to scale, some are now leveraging LLMs, particularly to automate roles involving voice, email, or messaging, to transform into high margin, scalable models.

That was Joe Schmidt. I'm a partner on the apps team here at Andreessen Orwitz. And here is his big idea. My big idea is romanticizing organic growth. And really what that means is I see the next evolution of private equity being possible where small businesses can leverage technology to basically automate parts of their services organization and then go and use a better business model to acquire other small businesses.

Obviously, LLMs, the talk of the town, penetrating every industry. But tell me what you're seeing on the ground in terms of how these LLMs are reshaping industries. It's just amazing to see how fast adoption has been across whether it's horizontal applications or vertical applications. And just for this conversation, maybe we'll focus on the vertical. Yeah. Obviously, we recently announced our investment in 11X, which is automating the sales function.

And I don't think I've ever really seen demand like what we're seeing for basically automating the role of an SDR. And so you're already seeing it within the sales organization. If you think even on a more vertical specific basis, you know, I see it in like freight with companies like Happy Robot, basically automating the call center operations around what a freight broker does, whether it's like calling and booking a load or dealing with a new carrier that a freight broker has to deal with.

We see it in healthcare with another one of our portfolio companies, Tenor, which is automating the back office processes associated with a small healthcare practice. And so I think we're seeing a bunch of these applications pop up and frankly have growth rates on anything I've ever seen again. And I think it's just a really exciting time to be building software and to be building startups. Yeah, and I love those examples because you're touching so many different industries. And maybe if we zoom out just a little bit and think about maybe like a framework for disruption, how do you think about where maybe this could be applied?

If you're interested, go read our blog because we've got a bunch of ideas on it. And we'll link it. But I think that like the way to just zoom out and think about this opportunity is like, where are there a bunch of humans that are basically dealing with whether it's voice or paper processes, where there's a bunch of unstructured data that they need to basically synthesize and then make sense of and then generate some sort of output. Obviously, that's what LLMs are super good at.

And so like the very common way of thinking about that might be like, hey, there's a call center somewhere, right? That's like the healthcare or the insurance example or the sales example. Or maybe you have some sort of, we call it the messy inbox problem here where there's a bunch of different pieces of paper that are coming in and you need to synthesize them and put them into some database. But oftentimes that's like an internal TIPA

team. But then also you could look for wherever there might be BPO spend, like business process outsourcing spend, where maybe a consulting firm is doing this for you. So those are the hot buttons is like looking for work like that, where it's basically something that is, you know, a bit's not Adams industry, like you're dealing with paper, not physical world things and something that is conducive to be automated by LLMs. Yeah, I have to admit, when I was reading your big idea, I thought, isn't this just like private equity? So convince me otherwise, how is this different from traditional private equity?

First, like what is private equity? To oversimplify, like private equity is a buyer purchases a company using some combination of equity and debt, traditionally like a good deal of debt, and then basically tries to improve earnings in the relatively short term. Normally, private equity has like a three to five year hold period and then sell it to the next buyer. Private equity firms typically are optimizing for IRR, internal return, and they're less focused on ultimately like a multiple on money return.

basically saying they're not focused on the biggest possible version of the company that they can build. Rather, they're focused on making incremental improvements in this time period because that's like how they think about liquidity and that's how the model works. The problem with that is it creates like a little bit of a perverse incentive structure where typically they're focused on one of two ways of making money. First is like the most obvious, like when people hear like private equity, it's like, oh no, people are gonna get fired. But obviously what they're trying to say is, hey, this hasn't been run in the most lean and efficient way maybe it possibly could have.

And so, hey, the accounting department doesn't need to be this big, or maybe we can consolidate these two different teams or whatever it might be. That's the first bucket.

The second bucket normally is saying, hey, we're going to acquire other small businesses or large businesses. Doesn't matter. We're going to acquire other companies. We're going to use more equity and debt. And we're going to basically buy the next incremental company for a lower multiple, roll it into our company and have multiple expansion. So these are the two very common ways that private equity firms make money. These aren't designed to have these businesses become the best versions of themselves. It's designed to make money.

In the short term. And it does. So I think what's interesting about this model is that it's actually addressing like a third leg of the stool, in my opinion, where traditionally private equity isn't focused on like fundamentally shifting the way that the business works. What I'm talking about is actually building like a technology business and technology core to a traditional services company, right? So in the past, the private equity firm would say, hey, like we don't need that many people in accounting. I'm talking about

automating like a large swath of the company. And to do so, you need to have a very different view on returns and you need to have a very different view on short-term earnings because what will end up happening is it turns out engineers,

pretty expensive, right? And it requires like significant investment to get this business to a more automated state. And so you have to not think about IRR over the next two or three years. And instead, you need to be thinking about what's the largest long-term valuable business that I can build and how do I invest in that future? So that's like the very different way of building businesses. I think you have to structure these businesses from the start, not as private equity companies, but as like long-term permanent like

we're going to build the best version of this company ever and like a company that can go public. And frankly, I think the best versions of this company, like there's a couple of examples on the public markets that are a little bit more holdco traditional software versions of this, but

One is a very well-known company called Danaher. There's another one called Tyler Technologies based out of Texas. These are like probably the best examples of these businesses historically. But again, they're not AI native and they're basically focused on the first two buckets of this, which is like financial engineering and or like operational efficiency. Not building in AI to their businesses.

That's super interesting. And I think just for the listeners, let's try to be really concrete here and paint a picture of what this could look like, this idea or opportunity of, quote, AI-powered, vertical-specific service startups. Yeah, that's a mouthful. Yeah, I was going to say, it's a lot of S's, but what would that look like? Let's try to really paint a picture here. Okay, so let's imagine that

Steph has an insurance agency in Columbia, Missouri. Great. Okay. So let's call it a small insurance agency. You do $3 or $4 million in gross written premium. That generally will mean you do a few hundred thousand, call it half a million dollars in revenue, that revenue. Let's say you have single digit margins on that revenue. So like you're living a nice life in Columbia, Missouri, but this is a small business. Yes. There are thousands, tens of thousands of these businesses across America. You're selling, imagine home, auto, life insurance, and a little bit of commercial insurance.

At your staff's insurance agency, maybe you have one or two or three people in the back office. Maybe there's a CSM that's dealing with like net new business that you're writing. There's someone in the back office that's dealing with all the paperwork, right? Because when you write an insurance policy, it's like, hey, I got to get a bunch of information from the customer. I got to go get the quote from my legacy system and I got to write all this stuff down and I got to deal with all these back office processes.

It's painful. And it's a lot of people moving paper around. That's the reason your margin is so low. And you've tried to fix this, right? You've tried to buy software, but the only software in this industry is really legacy, right? There's these two big businesses in insurance. You don't really have a great system here. And so you've just been hiring these people and you've been living this

frankly, hell for your entire career. And so what I'm talking about now is I come to you and I say, hey, Steph, wouldn't it be great if I could give you like basically automation that could take all of that paperwork that you've been paying five, 10, 15 plus people

percent of your net margin for and we can automate it and what does that require right it will require you going and getting deeply integrated from the technology side deeply integrated into the core systems of record right leveraging ai agents to basically have some of those conversations with the customer whether it's on new business on renewal or on client servicing and then actually resolving these claims in real time or like resolving these service issues in real time

Which I think is a super interesting opportunity because then you start to think about, okay, now Steph's insurance agency doesn't need four people. It might need two people. It might need one person. And you go from having 5% net margins to 30% net margins. And you go and you put a pool in or whatever it might be. Or frankly, what I think is really exciting is you use that extra cash and you buy John's insurance agency down the street. And now you see what the opportunity is to run this playbook. Yeah, because basically what you're saying is that what you put into the technology can be replicated for

insert other business here. Tell me a little bit more about why you need to buy the company though. Could I not create a tool that just sells into John's insurance company, Steph's insurance company? Yeah. And I think that's, you actually made a really good point at the end of what you just said, which is, hey, this can be applied to the next business. And if you build this software, I think in the right way, in the right industry vertical, once you've built those like core integrations in the example that we were just doing, it's into the applied Epic system or whatever core system of records

exist across the small business you're thinking about. Those systems are in many cases in these legacy industry are pretty ubiquitous. And so you can repurpose this technology across each one of these small businesses. I'm not saying it's going to be simple or easy. It'll require implementation, but that's the opportunity here is to stay vertical specific and build this like deep kind of understanding of the systems.

But it's a great question on like, why not just sell software to these companies? Frankly, if you've ever talked to a small business across middle America, like these mom and pop companies, they are not the most technology savvy customers. And it's not that they're not great businesses or super educated. It's just they just aren't accustomed to running their companies this way.

If I'm a software vendor, and frankly, I've done this, like I've gone and I've tried to sell software to a mom and pop insurance agency in Columbia, Missouri. Trying to drive adoption is super tricky. And so in buying the business, instead of trying to do business development exercises and convince someone that like, hey, the pot of gold at the end of the rainbow is worth it for you. Instead, we can now say, actually, this is what we're going to run the business. Operationally, this is our new cadence and our new workflow. And here's how we're going to get to the next stage.

So you're saying basically in purchasing the business, it gives you the leeway to actually shift things fundamentally, which is really what this is articulating. Maybe another question is just, you seem to think this is a really big opportunity. PE itself is a huge industry. So why do you think this is actually maybe a bigger opportunity than traditional private equity? So I think that this all comes down to capital efficiency. And so, for example, buying one of these small companies that I just described isn't actually like that expensive of an endeavor. And

If you can do what I'm talking about, you impact earnings. And all of a sudden, like the first one, two or three, maybe you don't immediately pay back using cash. But after the first few of these transactions, you start to throw off cash and that cash can then be used to acquire more and more of these small businesses. And so it becomes this beautiful like flywheel effect to the activity.

acquisitions that you can make and then that new business is that you acquire. And so unlike traditional private equity where you're constantly raising debt, maybe you need more equity. And again, it's much more financial engineering. Instead, you can actually just compound your investment over a long period of time and build value that way. So I think that's the opportunity here. Yeah. And I guess what you're getting at is the margin

change is so significant where you can do that. But instead, if you're doing some of the financial engineering that traditional PE relies on, you're often not changing the margins from 5% to 30% or to 50%, even in some cases. Yeah. And I think there's a nuanced point here where, again, if you do this right in the right vertical, you're

and the businesses that you end up acquiring are worth more to you than to any other buyer, that's when you can start to do really interesting things. Right. So I don't know if this makes sense, but let's say I'm doing something and Mark is doing something in the same space for those insurance businesses. We're both running the same playbook. He's a private equity firm. I'm running a technology business.

And I go to you and I say, hey, I have this wonderful technology platform. Wouldn't you like to come and grow with me? And I can take your earnings from 5% to 20% versus Mark says, hey, I'll just pay you a bunch of money and you're going to stay at 5% and maybe I'm going to fire Susan who runs your accounting, right? Whose value prop sounds better? And so in this way, you can one, like your business is actually worth more to me.

Sure. Right. Because I can impact earnings in the short term based on my technology. And two, you want to work with me because you see, hey, I'm not going to have to go in and fundamentally do a bunch of these things that are really cutthroat. Obviously, maybe there are certain people that are a little redundant. So I'm not saying this won't happen. Sure. But I do think that it's a much better story and it's a growth story versus just a traditional efficiency story. Yeah. And speaking of that growth story, it sounds like

a lot of the growth is inorganic in this case. At least that's what this story is about. So tell me more about that. Why the inorganic growth strategy versus maybe a more organic approach? Yeah, so I think that the way to think about these more traditional services businesses, oftentimes they're local enterprises, right? So again, like in Columbia, Missouri, you have a reputation as being like a trusted source of insurance services. And so people want to come and work with staff for that reason. And it's actually traditionally pretty hard. It's not just insurance, but think about

real estate, law, like all of these industries, they're traditionally local. And so it's pretty hard to go add new customers and grow organically. If you're doing that, like it's pretty special. In my opinion, the best version of this inorganic growth strategy, it's like you're acquiring a very sticky, a very local geospecific customer base. And then you're improving the business that's serving them and freeing up that person who already has like all of these wonderful connections in whatever space they are to go do more of that.

Right. So now you're not focused on, hey, shoot, I've got to deal with all of this stuff in the back office. I can go to the Little League field or wherever the heck you go to sell insurance and basically get to know people and do business development. And so now not only are we impacting earnings, we're actually impacting the net new organic growth rate, which is pretty clever. The way to think about this in my mind or the way that I think about it is, hey, if you were running an enterprise software company, right, and you were in the public markets and

and you're saying, hey, I could have 18, 20, 24-month paybacks on a great enterprise customer logo. That's the way to think about these small businesses. Can you find a way to have strong paybacks on these new business acquisitions and treat them much more as just like an enterprise software sales motion rather than a traditional private equity firm acquiring some small business and doing a bunch of financial engineering on it? But obviously, this is not easy. No. This is not operationally easy. So tell me more about

that. As I sit here in my venture capital ivory tower, all these small businesses are simple to automate. Yeah. I can't wait for my tweets at me. Well, tell me more about that. So as you think through the challenges, what is most stark? What's most clear as the thing that someone pursuing this would need to solve? Yeah, I think the biggest one is just understanding that like there's going to be a bunch of human problems when it comes to this. And so you have to be like eyes wide open. You have to be a great operator and you have to deal with scaling a human driven business.

So I think that's the first one and all the implementation problems around that. There's going to be a bunch of hairy legacy systems. You're going to need to navigate like all the process mining and figuring out what do I need to build here to actually drive incremental value. So that's the second piece, human level issues. Once you get through those, then you can start to do the process mapping, which is also exciting. And then the third piece is, I've talked about this a little bit before, but a lot of these legacy services businesses, while they may be

bits oriented. Like at some point there's like an atom involved. Sure. Someone has to go do the business development. Like it's not like you can just fully automate these companies. Yeah. But it could be even more complicated. Imagine you bought saying in the insurance example, maybe you bought a third party administrator, which is someone who deals with like insurance claims. So you could deal with all the back office stuff around handling insurance claim, accepting all the documents, figuring out what's what. But AI is not going to drive the truck to the crash.

AI is not going to figure all that stuff out. And so I think that's the other tricky part of like of these spaces is like identifying where is the right kind of combination of bits and atoms, primarily, hopefully bits, and then trying to figure out how to navigate some of these human problems and implementation problems. But it's definitely going to be hairy. I think if you don't do this right and you don't figure out like a very repeatable motion for these, again, smaller acquisitions, I think it's going to be really challenging to build a big company. So that's the opportunity here.

What are you looking out for in particular? Is it a certain type of founder? Is it a certain motion that you see someone doing that you think is really intriguing and you think, okay, they're on the right track here? What are you really looking for? First things first, I'm looking for an entrepreneur or entrepreneurs that kind

that have bottoms up understanding rather than tops down understanding. And what does that mean? I want to find entrepreneurs that have earned secrets in these markets, right? So somebody who has maybe sold to this theft insurance agency or someone who has run theft insurance agency or whatever the corollary is for each one of these markets.

It's very hard to come in and know the shorthand. You have to trust me that I'm going to come in and help you bring your business and take your business to the next level if I'm going to be that acquirer. And so I think you need to be a very special person on that front. The second piece is an understanding of, hey, can this industry be automated? Is it sufficiently bits oriented to the point where AI can actually have an impact on that? Like what percentage? Right, exactly. Is it the right kind of work that is conducive to being automated by AI? That's another piece.

And then I think the third piece is hopefully there's a very large fragmented market of acquisition targets such that like you're not basically going in and having to compete with the larger mid to large size traditional private equity deals. And so if you can find one of these end markets where there's a significant number of mom and pop businesses where you can go and run that CAC minded playbook of acquiring and implementing and building a repeatable motion around that.

I think there's just insane room to run and massive businesses to be built. And so those are the combination of factors that I'm looking for. And hopefully someone out there is considering this. The way that I would run it is I would go find a design customer, start building software, and then show yourself and frankly everyone else that you can actually start to move the needle on earnings. And that's when I think, you know, you've got the tiger by the tail and it's time to go and run fast.

All right. I hope these big ideas got you geared up and ready for 2025. Stay tuned for our final segment, part four, where we dive into the big ideas straight from our A16Z crypto team, who submitted 14, yes, 14 big ideas for the new year. We'll see you then.