走进 General Catalyst 的15亿美元 AI Roll-up 机器

GC is one of the 3 largest venture players. We put about $1.5 billion to work on the creation strategy, and first and foremost, it’s really around incubating companies. There’s a ton of potential right now, especially to incubate applied AI companies.

Hmm.

As the models get better, whether it’s OpenAI, Anthropic, Google, xAI, or others, we really do think there will be a lot of new companies that can be built on top of this technology. There’s a huge opportunity on creation to incubate applied AI companies. When we do that, we go one of 2 ways: either we incubate the applied AI company and have it scale organically, and then give it capital for marketing and growth, or we give it capital to go buy its distribution, client list, and the data that comes with it.

That second part has begun to be called the AI-enabled roll-up. Obviously, we are extremely involved in companies like Long Lake, Udia, Titan MSP, Beacon, Crescendo, and others that are now kind of the leaders in this field. At GC, we’ve been investing between $100 million and $150 million, but over 3 or 4 rounds of funding. We like to lead at least the first 2, and then we start bringing in other investors in the 3rd or 4th.

We think that in this much larger TAM than software, you can have software-like margins, because once you take out 30%, 40%, or 50% of what people do and free them up to do the hard stuff, you can double the revenue. Unlike private equity, we screen really hard for whether the company wants to change and whether it wants to implement AI. Then, of course, we want to hold these companies for 7 to 10 years, take them public, and not necessarily add debt and cut costs. Some of the companies in the roll-up will hit $100 million of EBITDA, and they’re less than 2 years old.

Oh my God. Marc Bhargava, welcome to Sorcery.

Thanks for having me.

1. General Catalyst Becomes A Company

Well, this is fun. This is great. I’ve been waiting to do this for a while now. You are the managing director at General Catalyst. I don’t think many people know the breadth of General Catalyst’s asset management function. Could you just break down how it’s structured?

Yeah, sure. I can talk a little bit about GC and how we’re set up and what we’re all working on. We were a venture capital firm, but we’re evolving to be really a company. People ask me, “Well, what does that mean?” There are 2 big parts to it.

As a company, we’re actually building companies for the long term. People have seen the announcement about us buying a hospital system in Ohio, for example. We did that through an organization called HatCo, which we incubated and then used to buy a hospital, and we plan to hold that for a really long period of time. We have a wealth management business that we’ve set up to help founders manage their money. That’s another company we’ve built in-house that we plan to hold for a very long time.

Finally, we now have an AI consulting business called Percepta that comes in and changes Fortune 100 companies, helps implement AI, and really leverages everything we’ve learned from AI investing. That’s another company we’ve built in-house that we will hold for a really long time, maybe forever in these cases.

So GC as a company is building in-house what we call transformation companies, with these 3 being examples of businesses we think should exist and are good for society. We’re not planning on taking them public. We want to hold them for a really long time. In addition to these 3 transformation companies, we have another bucket that we’re probably better known for, which is our funds.

There, we have the traditional venture funds: Ignition for the early stage and Endurance for the later stage. We also have a Customer Value Fund for helping companies scale really quickly, which we can talk about. Lastly, where I’m focused is on creation, which is part of our funds business.

Creation has been a very fast-growing fund for us because it’s all about manufacturing outliers. While our venture funds are always looking for the next great company or person, on the creation side we’re incubating companies that we would want to take public in 7 to 10 years. They’re not necessarily companies we would majority-own or hold forever, but they’re businesses we think really should exist—businesses we’re creating through creation that we want to take public in 7 to 10 years.

Most of the time, they don’t exist because there’s not enough capital for them, they’re multidisciplinary and require a lot of different expertise, or they’re a bit more traditional or outside the box—things like the AI roll-ups we’re working on, these new thesis spaces that just haven’t really happened yet. Creation is where I spend most of my time. But GC as an organization, we truly are a company now, building companies for the long term and creating and incubating companies that we want to take public in the future.

So how large is this? With the company as a new structure, will you be weighing it with AUM? How does this differentiate?

We’re kind of one of the 3 largest venture players. The way we really view it is through the returns for our investors. Our venture funds have to outperform and do really well to attract more LPs and, obviously, to attract more founders.

The nice thing is that all of these pieces do tie together with 1 question: What are the best ways to work with the most founders and help them the most? In answer to that question, we try to plug and match both the companies and the venture funds.

For example, if you’re a founder with a tech business and you’re growing quickly, maybe you use our wealth management transformation company to manage your money. Maybe you use Percepta to help you create your own mini-model or inference, or to improve your technology stack. Percepta can come in and really help with that.

Maybe if you’re in healthcare, you decide to sell into our hospital system in Ohio and test out your technology. So the transformation companies that we build in-house at GC are ultimately also helping our founders while delivering value for our investors in GC.

The same is true on the venture side. Not every founder has the perfect idea and is scaling. Some founders are second-time founders thinking about their next thing. So with creation, we help give them the idea. We introduce them to their co-founder. We help them even before the capital formation.

On the other hand, if you’re already growing really quickly and want a way to scale without as much dilution, our Customer Value Fund helps with that. So GC is doing a lot of things at once. We’ve hired incredible people for each transformation company and each fund product.

Ultimately, we now feel like if you’re successfully growing and running a company, there are more ways to partner with GC than with any other firm out there. Because of that, we’re really seeing our brand rise. We’re seeing the founders we back be extremely high quality, and their companies do well because there are just so many more ways to plug in and work with GC than with an individual solo venture fund.

2. The Creation Strategy Takes Shape

With your focus on the creation side of things, how much have you allocated to creation?

In our last fund cycle, we put about $1.5 billion to work on the creation strategy. First and foremost, it’s really around incubating companies. We think there’s a ton of potential right now, especially to incubate applied AI companies.

Hmm.

As the models get better, whether it’s OpenAI, Anthropic, Google, xAI, or others, we really do think there will be a lot of new companies that can be built on top of this technology. Every 3 to 6 months, the models are getting better and better, especially recently at things like reasoning, logic, coding, and math. So we think there’s a huge opportunity on creation to incubate applied AI companies.

When we do that, we go one of 2 ways. Either we incubate the applied AI company and have it scale organically, and then give it capital for marketing and growth, or, in about half the cases after incubating an applied AI company, we give it capital to go buy its distribution, buy its client list, and buy the data that comes with it. That second part has begun to be called the AI-enabled roll-up.

Obviously, we are extremely involved in companies like Long Lake, Udia, Titan MSP, Beacon, Crescendo, and others that are now kind of the leaders in this field.

3. The AI Roll Up Thesis Emerges

Wow. I think what would be really interesting would be to take a step back, because General Catalyst really leaned into this moment. People have forgotten about this, but there was an existential moment for SaaS in early 2023 or something like that. Everyone thought venture was dead, SaaS was dead. It was the end of everything.

No more venture capital. It’s over. That was the whole idea. But then some people thought, “Oh, no, this is an opportunity to roll up companies, to apply AI, and to use these efficiencies.” One of those funds was General Catalyst. So you guys actually really leaned in hard for this. What was the decision-making process for it?

Yeah. For us, it was a both-and strategy. We actually think 3 approaches can win. One, some folks view it as the model companies will end up being the winners.

They’re going to keep iterating, make better and better models, and go direct to the consumer. We’re obviously investors in Anthropic—large investors from the last round and from this current round. We absolutely think companies like Anthropic, OpenAI, and Google can benefit, and the model layer can benefit, so we agree with that.

We also agree with folks who say, “Well, it’s been overestimated. SaaS is not dead.” If you look at public market comps, they’ve continued to grow and do well. There are all kinds of new, interesting SaaS companies that are being started and growing.

But we also think there is this third bucket, so we’re kind of an all-of-the-above. We think there’ll be winners in each category, and we obviously want to be part of them. But that third bucket is for really fragmented industries where it’s very hard to sell into AI-native services and products.

Can we actually build the AI-native service platform and software and then go buy our distribution? A few examples: One that was very early on was Crescendo, where we led several rounds of funding to build AI-native software for call centers. We teamed up with Andy Lee, who ran Alorica for over 30 years, a call center chain, with 2 amazing CTOs. Someone from GC went and joined full-time, and we built out this software that automates 50% to 70% of what a call center does.

So, after proving that out in a year and having 10 or so pilot clients, we gave them the money to actually go buy a call center, and they’re well on their way now to taking 10% EBITDA margins and turning them into 40% EBITDA margins. That’s because when half the tasks can be automated, people can start to focus on cross-selling, revenue opportunities, and the harder tasks.

So Crescendo is just one example. With Long Lake, a second example, they’ve gone after the HOA management space, PO services, and others. They’re also scaling quickly, doubling the EBITDA margins of the companies they’ve been buying, and they have the benefit of having all of the data and the ability to do change management and change practices.

Then, of course, third and fourth, we have companies like Titan MSP in the MSP space. They went out, got 6 pilot clients, and showed us they could automate 38% of what an MSP does, which is an outsourced IT services firm. Now they’ve bought RFA, which is a well-known MSP in New York, and they also are well on their way to doubling EBITDA margins.

We definitely think that there’s an opportunity for software, but there are a lot of industries that are highly fragmented. They’re split across the country. They don’t really buy technology products. When they do buy tech products, they call it IT, and they cap it at 2% or 3% of revenue.

At the same time, there’s this ability with AI to go in and automate 20%, 30%, even 50% of the tasks at those businesses. To really make the most out of AI technology, you want to buy these companies and be able to free up people to perform the higher-earning tasks, really grow the company, and grow the margin story. That’s what we’re starting to see proven out across all of these case studies and examples. We’ve been getting a lot of interest from crossover funds and others as we’ve been doing this, and we’ve really been proving the model.

So these are traditionally not sexy industries at all.

Definitely not. Yeah.

They’re very boring.

Definitely.

But you came from a really fun industry. You came from crypto and everything. From my understanding and research, you looked at 70 different industries, and you landed on 10 of them for this fund. What was your process for that, and how did you evaluate the opportunity set for those 10 industries?

4. Seventy Industries Become Ten

Well, I’m still very bullish on crypto and got into the space in 2016. I was an early investor in amazing companies. I started to go meet with Jen and Greg. We built it and sold it to Coinbase. Today, it’s the heart of Coinbase Prime, which Greg Tusar, my co-founder, still runs.

After the Coinbase exit, I found myself doing a lot of angel investing, a lot of it in great fintech companies like Ramp and Zip and others. But I was also looking at NFTs and other things in crypto that didn’t really play out. At the same time, I started meeting these really smart AI founders, and I felt like if I was giving money to people putting monkeys on the internet, I probably should support these PhDs in AI from Stanford who are working really hard on their companies and pay it forward.

So I got to be in the angel round of Harvey, Together, Windsurf, and a lot of these names. Seeing their progress was really what inspired me, then, at General Catalyst, to say, “I still want to invest in crypto,” which is obviously growing really well in terms of Bitcoin and stablecoin adoption and things being built on Ethereum, “but I also want to be part of this AI movement.”

As I started to think about how I wanted to be part of this AI transformation, I started talking to a lot of second-time founders like myself. Most of them were identifying that the hard part was go-to-market. I can build AI software to automate a lot of what a call center does, or a lot of what a law firm, an accounting firm, an IT firm, HOA management, or property management does, like with Dwelli out in the UK.

But it’s going to be really hard for me to go to market in these unsexy industries where I don’t have a network, where it’s really fragmented, and where they don’t really trust tech. It’s not the same as a Ramp or Brex selling into all the hot startups.

It was really brainstorming with those second-time founders, thinking about how we take this AI technology and really scale and grow, that we together conceived of this AI roll-up concept very early on, especially with folks like Long Lake being a great example, or Crescendo. Those were really 2 of the first in this space, and we got very lucky to work with them.

As we started seeing their success, we stepped back and said, “We should double down here.” We were also obviously getting a ton of interest from other investors, LPs, and others. That’s when we became more systematic, about 1.5 to 2 years ago, and said, “Okay, we have these success stories. Maybe they’re a bit more one-off.”

It was about these amazing founders we wanted to back and about applied AI, but let’s actually look at 70 services-based industries and try to figure out which of those 70 we could automate 30% of the tasks happening there.

And we were able to figure that out by first asking, “Well, what can AI automate? What’s it good at?” There were 4 pretty clear buckets. 1 was customer success, support, and service. Basically, AI agents can do that task pretty well.

2 was data entry and evaluation. If you’re filling out the same forms and adding the same tables, that was the 2nd bucket. It could really do that well. 3rd was creating content and copy and marketing, so an FAQ, an investor presentation, or a list of dos and don’ts. That sort of content was really good, as was responding to email, doing translation, or creating an NDA contract.

Then a 4th bucket, which has only emerged in the last 9 months, is basic logic and reasoning. In insurance, for example: “Should I underwrite this person? They’re a super-risky driver. Here’s their background.” It’s not so clear, but you can do almost as good a job as a human saying, “Yeah, here are the risks, here are the rewards, and here’s my recommendation.”

So this 4th bucket of basic logic and reasoning, which has really been pushed forward by all the automation in things like code and other areas, was 1 of the 4 buckets we identified. We mapped those forms of automation against these 70 industries, and we came up with 10 where we had really high conviction that at least 20% to 30% of tasks—not people, but tasks—can be automated.

Then, if you work at an HOA management company, a call center, or in legal, with the 30% free time you have, we now ask you to do more work: take on more clients, cross-sell them, or do harder tasks. With the same cost basis, but adding all these AI tools for automation in these 4 buckets, folks can now take on 20%, then 50%, and, over 3 or 4 years, even 100% more tasks with the same cost basis.

We’re already seeing in many of the case studies that, if you look at the first couple of companies they acquired, even if they’ve only grown revenue 20%, if they’ve kept the cost basis flat using AI and then freed up people to do 20% more tasks, businesses with around 10% or 15% EBITDA margins can become businesses with 30% EBITDA margins. You can roughly double the free cash flow, and we’re seeing that happen in many of our businesses in a 1-year period.

For us, though, at the heart of that is: Is the AI automation enough? Obviously, we decided not to go forward with 60 of the 70. So we held ourselves to a really high bar: Is this an industry where AI automation can actually make a meaningful difference? Then can we actually go and buy and roll up these companies, get their data, improve the models, improve our software, automate 20%-plus of tasks, free people up to cross-sell, and get more revenue?

The result is a higher revenue base with a very similar cost basis, which generally translates to a really large improvement in EBITDA margin.

5. Services Businesses Gain Software Margins

This market, some estimate, is between $20 trillion and $30 trillion. How do you view this type of private company in the grand scheme of knowledge work?

Yeah. We have a slightly more conservative estimate of $16 trillion for services industries globally, which is really massive, to your point. On the software side, software globally is like a $1 trillion opportunity, so it is 16 times larger.

Why hasn’t everyone in VC rushed to invest in services businesses like accounting, legal, IT, and insurance? Well, historically, they have not been that profitable, so they generally break even in many cases. Obviously, as you get to scale that $1 trillion of revenue in software, the marginal cost is really low, and companies can become very profitable and have amazing cash flow.

Companies like CrowdStrike and ServiceNow, as well as the hyperscalers like Amazon, Google, and Microsoft, show this model. There are all these amazing, proven examples in the software field of having very strong free cash flow and high margins.

Our thesis is that these unsexy services industries that were break-even or had 15% to 20% EBITDA margins, with net income obviously much lower, can look like software from a margin profile. Once you take out the repetitive tasks and take out 30%, 40%, or 50% of what people do, freeing them up to do the hard stuff with the same cost basis, you can grow revenue by 50% or double the revenue.

Your margin profile now looks much more similar to software. So we think that in this much larger TAM than software, you can have software-like margins. That’s only happening now because of AI, both in the form of LLMs and creating software to automate tasks, but also in the form of AI agents that can actually handle tasks from soup to nuts.

We think there’ll be a mix of AI-native software, agentic workforces, people working for agents, agents working for people, and kind of mixing all of that up—a totally different margin profile for some of these companies.

6. The AI Roll Up Playbook

I’m so curious how you assembled a team to do this, because it’s fairly novel. You’re doing it within a traditional venture capital fund, but you’re bringing traditional private equity into it as well, and also tech. You need operations. So how did you assemble the team, and what’s the composition?

Yeah. So the Creation strategy at GC has really existed since the beginning. We’ve really focused on backing second-time founders and operators, and stories like Kayak and Livongo were built out of our offices.

In 2021, we formalized it a little more, had a dedicated sleeve around it, and said, “We’re gonna commit this much capital.” Then we doubled that and grew it last year as well. But the mantra at GC—how do we get amazing ideas and co-founders, put them together, build companies, and leverage our network—has been around for a long time. Many times, it’s multidisciplinary: one’s in healthcare, and one knows technology. That’s been the case for a long time.

So we’re building off a great foundation at GC, but we’ve really changed it now by also adding in a lot of the applied AI piece. At our heart, we have 3 things. First, we definitely want second-time founders like myself, Hemant Taneja, who built out Livongo, or others. Folks on the team like Chris Kaufman have had operational roles. He was an interim CFO at a rug company marketplace, so we definitely want people who have founder or operating roles.

Then we also want people who have some VC or angel investing experience. Lastly, we’ve been mixing in people who have private equity experience. I started my career at McKinsey and then worked in private equity, so we have a lot of folks on our team like Cade, Chris, Sarah, and Mark Crane.

We’re looking for this—maybe it’s more difficult to hire for—but people who have PE experience, operational experience, and VC experience. We’ve connected this motley crew. What’s been really helpful to us is that our founders are selling us.

Mm-hmm.

We’ve made about 9 offers in this AI-enabled software roll-up idea, or AI-enabled software that turns into roll-ups. We’ve made 9 offers, and we’ve won 8 out of the 9.

GC has certainly created a brand around the idea that, if this is a strategy that resonates with you and you’re a founder, there are a lot of ways we can help you: setting up the team, looking at targets, looking at industries together, thinking about how you should be using debt funding, and thinking about whether CVF makes sense for your business.

We’ve really created a flywheel and playbook around this. What I’m most proud of is how it’s resonated with founders. We’re clearly the top choice in this space.

That’s so fascinating.

Yeah.

Can you talk through the funding mechanism part of it? How do you structure it?

Yeah. We normally do 1 or 2 rounds to build the software piece. With Titan MSP, for example, we gave them a traditional seed/Series A round to build out AI automation for IT services for MSPs, and they went out and got 6 pilot clients in over 3 or 4 months.

They showed us, “Hey, look at all the tickets that happen in an MSP. We can now automate 38% of them.” When that happens, we start talking to them about a 2nd round of funding. We say, “Well, let’s actually go look for targets, because if we can buy 1 of these companies and automate 38%, we’re certainly gonna increase the margin and be able to reinvest a lot of that free cash flow into growth and create a really successful company.”

We look at the pilot clients that they’re working with and ask who really wants this AI technology. Unlike private equity, we screen really hard for whether the company wants to change and implement AI. Of course, we want to hold them for 7 to 10 years and go public, rather than necessarily add debt and cut costs. That’s a really different model than private equity.

In their case, we gave them a 2nd round of funding when they found their target RFA. Now that they’ve found that target and are doing really well on the AI transformation, they say, “Okay, now we want a 3rd round of funding to go and do a lot of small tuck-ins or buy another platform.”

At GC, we’ve been investing between $100 million and $150 million in each of these projects, but over 3 or 4 rounds of funding. We like to lead at least the first 2, and then we start bringing in other investors in the 3rd or 4th. We get to work with some really great crossover funds and other VCs as well on these projects.

Some, like Elad, whom you've had on the podcast, are great collaborators of ours, among others. So we've definitely been seeing a lot of success with that.

You said you plan on holding these for 7 to 10 years and hopefully going public or having some sort of exit scenario. That's a little bit different from the fast turnaround of a traditional private equity fund and everything that goes on with that. Call it—

Yeah.

—bad practices. But I'm curious about that philosophy. Why hold it for that long?

Yeah. A lot of it is because we want these to be compounders. The end vision for most of the founders is, “I want to be the next-generation TransDigm or Danaher or Constellation Software.” These are companies in the public markets with over $100 billion in market cap that have outperformed the Nasdaq and the S&P. What they do is go buy companies, improve their operations, generate more free cash flow from the company they bought, and with that extra free cash flow, go buy even more companies and rinse and repeat.

They've been termed “compounders” by public investors, and they've done really well. It's a great model. Our view is that they're going to be these AI-native compounders—

Hmm.

—that go buy companies, improve them, and generate more free cash flow. Then they reinvest that to continue to grow quickly, but the driving improvement is more AI. Historically, it's been things like pricing, sourcing materials better, or maybe improving some of the go-to-market. Here, a lot of the improvement is in this AI automation piece, and so we think there will be plenty of these $100 billion AI-native compounders in the public markets 10 years from now.

The founders who are starting these companies generally have very strong tech backgrounds, and they want to build a business that's VC-backed. They want to take it public, and then many of them want to run it in the public market for 10 years. So they really want ownership of this for the next 20 years. That's very different from private equity, where there's a management team that works for a private equity firm and is highly incentivized to get the company sold in 3 to 5 years, maybe to another PE firm or maybe to a strategic buyer.

The focus there is adding debt, cutting costs, and improving margins that way. It's not really about investing in technology for the long run. With some of this AI technology, you add costs, right? So you can't go in, add debt, and cut costs. We're actually going in and adding costs. We're saying, “Hire engineers.” We're subscribing to LLM-based software. We're going in, investing, and adding costs, but we think it's worth it because we're adding a lot less cost than the amount of revenue we're adding as we're freeing people up to do more and more revenue-producing tasks.

So our mindset is longer term, and the founders who actually own these companies—majority owners—have a longer-term mindset. It's just a different strategy. PE has obviously done really well. There are a lot of efficiencies in many markets. There's also not enough automation in many markets to do what we're doing.

We have a lot of respect for private equity and the work folks like Thoma Bravo, Blackstone, and others have done. They've been incredibly successful. We're playing our own game around, “Hey, how do we help these founders create AI-native compounders that'll be in the public market with a much longer-term view, and with a view that we should take these companies public?” Maybe even GC holds the stock after they go public and keeps compounding.

Efficiencies, synergies, change management—all the good buzzwords.

Exactly. We pull some of those in and leave others out. On the change-management side, we tend to screen really hard for the right founders.

Hmm.

Our view is that when our founders create these holdcos and then want to buy companies, they generally don't do the change-management part. We screen really hard for people who want AI-driven change in the businesses from the get-go.

This is something I've noticed, and there's data around it too, but the private markets are swelling. It extends from private equity to private credit. I think this would be—

Totally.

—part of it. Carta has seen this as well, and they've extended their products into private credit and private equity. They're now doing LP management. How do you think those markets have actually changed operationally with AI, with what's going on, and with competition? Even when you're looking at buying companies for some of these, is there competition between all of them? How does that work?

7. AI Reshapes The Workforce

We haven't seen a lot of people implement AI effectively to date, and we're generally of the view, along with that MIT study, that maybe 95% of these Fortune 100 companies have tried it, said it was kind of a waste, and found that it didn't really work and was janky and that kind of thing.

So we believe that it's really hard to actually turn AI-native. Our worldview is: Let's create an AI-native team. Maybe they were heads of applied AI at Rippling, Scale, or Stripe, or they worked at places like Figma, and they've really seen AI applied within the enterprise setting.

Let's put together these teams, let's incubate AI-native teams and AI-native software, and then let's go out and buy these companies. We've found that when traditional companies are trying to implement AI, they get some benefits, for sure, but there's a lot left to be desired in terms of the amount of automation they can achieve.

At least for us, we think it's pretty difficult to just take models out of the box and implement them. You need a lot of context, you need an AI-native team, and you need a team that knows how to build software around the models. All of that is kind of difficult, including which model to use.

So we agree with the MIT study that says AI is kind of a buzzword that's hard to implement right now, and we don't necessarily think that the world is going straight to AGI, where AI agents will immediately be able to do everything. I think we sit, at least in the creation strategy, somewhere in the middle. If you have an AI-native team, but you also have the traditional clients, customer base, and services folks, and you mix it together, that's where you can have the best outcome, the best revenue growth, and the best margin profile.

We really believe in this hybrid approach, where AI in the right hands will be able to automate a good deal of work in many industries, but you'll also need people for certain tasks. This mix of people and software agents is going to be unique to each industry, and we can create the category-defining company in each industry to go after that opportunity with the full toolset in the toolkit.

I recently had Michael Barton, who's a sector head at Coatue, on the podcast, and he framed this as there being 2 camps. The first is that AI will drive—and this has to do with layoffs—efficiencies for teams, and teams will be smaller. Then there's another camp that says, no, this will make your teams much more powerful, and you'll be able to build bigger teams that are just much more effective. Where do you sit between these 2 camps?

We're pretty squarely in the second camp.

Okay.

I'll give you an example. We incubated a company, Hippocratic AI, which is an AI nurse. It's AI nurse software, and there's just such a massive shortage of nurses in this country and in hospital systems.

Our view is that AI is not coming in to replace nurses, but now 1 nurse could manage a team of 5 AIs that can do a lot of the basic tasks, like checking in on a patient or getting information. For us, we think that AI will create a lot of volume in the economy. There will be a lot more new jobs.

Every time this happens in technology, it's the internet coming in and replacing jobs, or it's locomotives coming in and replacing jobs. The truth is, it's very granular. Sadly, yes, some jobs might not be there anymore, but at the same time, it's creating a lot of economic opportunity and new types of jobs, or even letting 1 person do a lot more work.

That's what we see the most in the services industry. If someone can now do 3 times the amount of work because they have orchestrated AI agents and use LLM software, are you getting rid of someone who's now 3 times as efficient and only needs a slight salary increase? You're probably not. You probably are hiring 3 or 4 more of them.

That's what we see on the ground. In many of these spaces, there's a huge shortage of nurses, accountants, and, yes, even lawyers. It'd be nice if everyone could have a lawyer on their phone, right? In many of these industries, more and more people would love to have that.

We believe in this idea of abundance, and abundance is going to come from that mix of people and AI services. It won't be purely AI. It won't necessarily be purely replacing people. We do think that there will be a transition period, but there's some happy medium.

The second nuance is that the first things to go are a lot of the outsourcing businesses. There was a lot of outsourcing in legal. There was a lot of outsourcing in call centers. There's a lot of outsourcing in insurance, like looking over claims.

What we're seeing a lot of is that when you add AI technology to American companies, they're now much more efficient. They don't need to outsource a lot of these more repetitive tasks.

So I think the area that will be hit the hardest is probably abroad, in places like India, the Philippines, places like that, even Mexico. So here at GC, we are committed to trying to figure out how to retrain and reskill, and whether we can invest in companies that are training people in new fields. And we think it'll be felt most, actually, outside of the United States.

After surveying all these different shifts and this transition period that you talk about, I think there's definitely going to be one, too. You don't know what the future of jobs will be. Calci had a market that hit up to, like, 86%—I haven't checked it recently—on layoffs in tech.

And that there was an 86% chance that there were going to be more layoffs this year than in 2024. But I don't know if that one's true or not, so you'd have to check the data.

Right.

But you would think that because of all of this mix, new industries might pop up. What are you thinking might come next? What could come next?

Yeah. I think it'll be a lot more of the things that right now are really limited to people with resources. So I think there'll be a lot more legal use cases and way more representation in legal. I think real-time understanding of your company, your balance sheets, and accounting—there'll be a lot more there.

So I think there'll be industries that already exist, but with much more abundance. On the nursing side, right now, if you're discharged from a hospital, maybe they follow up with you for 3 months or 6 months, but it just doesn't make economic sense to follow up with patients 9 months or a year out. But if you were truly doing the best thing for your patient, you would follow up with them in 9 months and in 12 months.

So I think a lot of these industries, like healthcare, insurance, and legal, where people have resources but then are tapped out, there'll be just a lot more around that. And so I think you'll have way more sophisticated markets in all of these places.

So, to be as unbiased as possible and objective, do you think AI roll-ups are underrated or overrated?

I think they're underrated right now. Most people you talk to haven't really heard of them. So we're just getting started on this part. I think right now people think, “Oh, it's a way to maybe put more money to work or something,” because obviously, as you buy companies, you can move quickly and put more money to work.

But I think folks don't fundamentally understand that in a $16 trillion services industry, we could have companies growing faster than software companies and at higher margins than software companies. And so this pool of $1 trillion that VC has been fishing in has actually expanded now, maybe not to the full $16 trillion, but to multiple trillions. I think very few folks really grasp that. And so I definitely think that for now they're underrated.

Yeah. I brought that up because I'm thinking about all the numbers that you've been mentioning throughout this, like increasing EBITDA, the growth, and the account count—all of it. It seems much more lucrative, and so I'm happy that we're diving deep into it. So when we look at this industry and the different players that go into it, how much of a role do consulting services play in this, and what do you think happens to McKinsey—

Mm.

—where you used to work, or Accenture, and how effective are they in this process? Because it's really—

Totally.

—difficult.

Our view is, at least on the AI roll-up side, we're more focused on these smaller, fragmented-style businesses and putting them together and doing this AI transformation. McKinsey, Accenture, and others really focus more on the Fortune 100, and I think they'll have some success in embedding AI there.

But at GC, we created a company called Percepta, one of our transformation companies that we want to hold for the long term. The thesis there is that maybe traditional consulting isn't really designed for a lot of the AI implementation piece, so they can give more strategy advice, but not necessarily the same kind of, “Hey, here's what we're seeing. It's a fast-moving industry. Here are the companies we've invested in. Let's actually go and implement kind of fast-moving, AI-changing technology in the Fortune 100.”

So I do think there's a huge opportunity for companies like Percepta. You probably will also see McKinsey and Accenture and others benefit, and the Fortune 100 also will have to evolve. Our focus more, though, on creation is incubating these companies in more fragmented industries where you don't necessarily even have the resources to go to a McKinsey or an Accenture if you want to add AI implementation.

Because of General Catalyst's very large platform that you have, how do the private market investments help determine or inform these companies? Are you partnering with any of them? How does Anthropic fit in here, and what are you learning from them?

Yeah. So with Anthropic, we're kind of learning where the models are going. There's always a new release every 3 months or 6 months. We're better understanding what tasks these models can do, especially as logic and reasoning get better.

And then we have portfolio companies like Rocks, which is taking on Salesforce, for example, and is more AI-native, or Udea, which is focused on the legal space and is also more AI-native. Serval, which came out of stealth last week, is taking on ServiceNow in a more AI-enabled way.

So we're working closely with Percepta to let them know about the companies in the portfolio and what they're seeing, both in terms of model improvements and product improvements. And then Percepta's getting this real-life view of everything going on, as much as it can, in AI, and then is able to help its Fortune 100 clients implement all this new technology, a lot of which our team is helping identify and pass up to them.

This was a good question from Kyle Harrison, but he wanted to ask: How do you decide whether to build an AI roll-up or invest in another SaaS company?

Yeah. It's a really good question. One answer to that is percentage of automation. So if we think a space is going to be 90% or 100% automated, then probably software is the best solution. And honestly, an incumbent like Microsoft, Amazon, or Google that has the distribution can just push the software. It can be part of your Google Workplace or your Prime subscription.

So if something's 100% automatable, it probably should be a software or agentic solution, and then most likely the winner there could be a fast-growing company, but also it's very likely someone who already has the distribution. So if coding gets fully automated, maybe it can just be pushed through by Google or Microsoft or someone else.

So one thing we're careful about in the AI-enabled roll-ups is we target at least 30% automation, but we actually don't want more than 70% automation. Because if something's approaching 80%, 90%, or 100% automation, then there's really not the people-services part of it.

The second thing is fragmentation. When the industry is really hard to sell into, imagine creating AI-native software for homeowners association, or HOA, management. That's just extremely hard to sell into. So you could create an amazing product, I'm sure, but your sales process would be very, very slow. So the second thing we look for is that it's very fragmented, and it's hard to sell a SaaS product to these businesses.

And then the third thing we look for is that it's really sticky. With HOA management, it's a 2-year contract. So it's important that it's a low-churn business where we can go in, we can do all this AI implementation, and there will be some hiccups along the way, but we can kind of smooth those over, and we have these loyal clients. Accounting is another really good space where there's very low churn that we're in, as are insurance and MSPs. And so those are the things we look for.

We do think that software services will win out in areas that are more than 80% automatable, or where it's easy to sell into your client base. You have a few large tech customers. There are plenty of areas where it really makes sense to be more of a software solution. That said, we think there's also this really big market for AI-enabled roll-ups, especially in things like services.

When you're working and building these companies, what are the metrics that you use to determine their performance? I would assume some of it's a little abstract because it's the team and that sort of thing, but how do you measure performance?

Yeah. In the first tranche of funding, what we're targeting is this 30% automation. So go out and get 10 clients or get 5 private clients, but really show us the software you can build or the agentic workforce you can build, and then come back to us and map out: If you look at all the man-hours in a company, can you automate 30% with the products that you have built or stitched together?

Then the second key metric is, when we go and actually buy companies, we're looking for how you're improving the margin. So you've automated 30% or more; how's that flowing through to margin? And we want a really detailed plan of how to double your EBITDA margin, generally from the 15% to 20% range to 30% to 40%.

So that’s the second really key metric. Once you start showing us you’re there or you’re on track to get there, that’s the second really important metric. And then the third is: now you’re growing, scaling, and moving quickly—what multiples are you buying these businesses at? How quickly do you get those multiples down as you improve the margin profile and the EBITDA profile as well?

And so then we get a bit more sensitive to entry multiples and doing things like smaller tuck-in acquisitions and using debt. And then the final one is just dilution. We want the founders of these companies to still own at least 10% of their business when they go public, ideally 20% or even 30%. That’s also a really important metric: are you compounding now with your own free cash flow and with debt financing toward the later rounds so that you can still own a big chunk of your company when you go public? And, of course, we at GC can own a big chunk.

Mm.

Ideally, also 20% to 25% when these companies go public. So, a really large, concentrated position in things that we think can be really massive outcomes. Along the way, we have these different metrics for each stage of the AI-enabled roll-up.

Oh, wow. That’s highly concentrated.

Yeah, for sure.

This is a really difficult question.

Yeah.

Where are these businesses located? Where have you traveled to?

The holding companies that we help incubate and fund are pretty much exclusively in San Francisco and New York. Out of the 8 or 9 deals that we’ve done, I think 6 are here and 2 or 3 are in New York. We have seen that talent has come back, for sure, to the Bay Area. It’s the center of applied AI and research AI, and then New York, too, for financial services especially. Some of the really great M&A specialists and people who have built and sold companies are out there.

So we’re primarily working with really experienced entrepreneurs, second-time founders, and amazing operators, and they tend to be between San Francisco and New York. Then, once we create the holding companies and they create the software, we show the automation, and now we’re actually buying businesses. Those are almost never in San Francisco or New York. They’re all across the country.

Increasingly, it’s getting more global. We invested in 2 rounds in a company called Dwelli, which is out in London, rolling up property management and scaling really well, while also doubling the EBITDA margin of the property managers they’re buying. We’ve also done a deal in Germany in the accounting space.

So increasingly now—well, my answer was at first the US—we also have bets in the UK and Germany, and we’re looking at this strategy really closely in India, too. Europe has much lower entry multiples. People sometimes rag on Europe, saying that it’s hard to build a big company because there are so many markets. In some ways, the AI-enabled roll-up is really perfect for that because we can buy our way into new markets and iterate, and it’s the same back-end technology to automate accounting, for example.

Increasingly, this will be a more global strategy. But to date, it’s been New York, San Francisco, London, and Berlin.

When you go global, do you keep the teams here in the US and the operations overseas? How does that work—

We normally—

—talent-wise?

We normally invest in a separate company to do an HOA property management roll-up, for example, in the UK. That’s Dwelli, and they’ll expand to broader Europe, for example. Right now, none of our roll-ups are really working internationally, with the exception of one that’s rolling up software companies. They’re based in Canada but operate between Canada and the US.

I mean, more on the talent side. Is the talent in those geographies as AI-native as it could be in the US? How does that compare quality-wise?

I think it’s less the US versus international and more the San Francisco Bay Area versus everyone else.

Okay.

I think San Francisco just has a much higher density of talent and quality of talent than really anywhere else in the world. But I think it is exciting that there are a lot more homegrown, great companies in New York, London, and Berlin, and the quality of talent is rising in all of these other places—so much so that we definitely are investing more in Europe and investing more in India.

But in terms of where the absolute top-tier talent in the world is, I think it’s still here in the Bay Area. There’s nothing special about the Bay Area except that it attracts all the really great people to leave wherever they are and come work here. So that’s been the draw.

You’ve been at General Catalyst for a couple of years now. What’s the biggest lesson that you’ve learned from Hemant?

Definitely advocating for the team and paying up when you have to. I got to learn from HT some epic stories about how GC got involved in Stripe, and I won’t repeat all of the details there, but—

Ooh.

There are some companies you just really want to be in, and it’s less about the terms and more about catching them early.

Mm.

You can buy 10% or 20% of an iconic company at seed, Series A, or Series B. A lot of people have the mentality of, “Oh, this is a $100 million check. Let’s spend all our time. This is more important than a $1 million seed check.”

But the reality is, normally, your $100 million check is buying you 10% of a company. Your seed check for $1 million is also buying you 10% of the company. So it’s really important to ask: What are going to be the next 50 iconic companies? Ideally, try to catch them early and pay just as much attention to buying 10% of some seed business as 10% of an established business.

So at GC, we’ve really shifted a lot of our focus to seed, with the acquisition of La Famiglia and bringing in Jeanette, who’s running our Europe practice and is a very prolific seed investor from La Famiglia. Venture Highway, the India seed firm that we also acquired, we brought in recently. And then Yuri came to lead our US seed practice. He had a successful seed firm.

I think that mentality is really being pushed through: can we buy 10% to 20% of companies early, rather than thinking that certain types of investments are more important than others? It’s really about how we get 10% or 20% of those iconic companies.

So intense.

Yeah. There’s a lot going on there.

There’s a lot going on. So what are you most looking forward to in the next 12 months?

In the next 12 months, I think becoming much more public about our thesis. Some of our best roll-ups have decided not to do any press whatsoever, which I can understand. They’re cash-flow positive. Some of the companies in the roll-up will hit $100 million of EBITDA, and they’re less than 2 years old.

Oh, my God.

We have companies like that who say, “If we’re generating hundreds of millions of free cash flow, we don’t need to be out fundraising every 3 to 6 months, doing the press tour, and promising people AGI is around the corner.”

If you have a good business, you’re making a lot of money, and you’re seeing AI implementation really working, a lot of our companies have remained really quiet or more stealthy, despite doing amazingly well. But I am excited to hear these stories come out much more publicly, and obviously we’ll send them your way as well.

In the next 6 months or year, as we have these amazing proof points that I get to see from being on the board of many of these companies, we’re going to start telling our story much more to the world, with an angle of working now with more crossover firms and starting to prepare these companies to go public. It’s really an amazing journey.

For me, it feels a lot like 2016 in crypto. There were a few people who thought it was legit, like USV and a16z, but most of the market was just trying to be polite because they didn’t want to offend USV, a16z, and Founders Fund, which was obviously really involved directly in Bitcoin. They were just trying to be polite, but they didn’t really believe in it. There were a few people who did.

I think we’re kind of there in the AI roll-up right now, where ourselves, Elad, Thrive, and a few others, including 8VC, have embraced it, but I think most of the market is still pretty skeptical. Similar to the crypto evolution and where Bitcoin and stablecoins are today, you see very few people who say, “Hey, there’s no value there. It’s not interesting.”

I think that’ll be us in a few years, too. So it’s been really fun being on what I think is a journey that will most definitely be proven true.

Oh, so fascinating, and really cool to capture it at this point in time.

Absolutely.

I’m really excited to do a rapid-fire interview with every single one of the companies.

No, I think they should definitely do more press. We have a few that are starting to do more, including Titan MSP, Udia, and Crescendo, but we’re excited for the rest, too.

Okay. Well, Marc, it was a pleasure to have you on. Thank you so much for sharing more—

Yeah, of course.

—about the Creation Fund.

Yeah, thanks so much for having me. It was a real pleasure.