The AI-First Roll-Up

Why VC-backed consolidators may struggle but software (and AI) will win

As discussed in a previous essay, the buyout world has intensified its focus on vertical SaaS in recent years. Lately, though, even early-stage VCs have been getting in the mix, floating investment theses inspired by classic M&A-driven roll-up strategies. Not all models are made equal—from tech-enabled roll-ups to acquisition platforms to growth buyouts—but they all envision some combination of software enablement and M&A to achieve scale in new ways. We’ve enjoyed the following recent write-ups—all very well done and worth a read to get a sense of the milieu:

• Slow Ventures on Growth Buyouts (GBOs)

• Equal Ventures on Insurance Brokerage Consolidation

• Tidemark on Tech-Enabled Vertical Roll-Ups

The potential of vertical AI is generating a new level of excitement around these strategies. If you buy the vision that LLM-based tools will drive significant margin recapture for legacy industries, it follows that there’s free money on the table for whoever can drive adoption. What better way to win this race than purchasing legacy businesses and implementing AI-centric automation? Even setting aside scale benefits, multiple arbitrage on the margin differential should yield alpha.

In today’s essay, we will discuss the incarnations of this “AI-First Roll-Up” concept we’re seeing. We’ll summarize the pros and cons, spotlighting the fundamental challenge in our view. Finally, we will introduce a new concept, the “Synthetic Software Roll-Up,” an emerging software aggregation strategy that we see as retaining many of the pros of PE strategies while obviating several of the major cons.

First, let’s begin with some context.

Why are Some VCs Excited About Roll-Ups Now?

The fundamental “why now” for AI-First Roll-Ups is in the name—LLMs have the potential to create significant producer surplus by lowering costs to deliver a service. That said, we started seeing interest in roll-up strategies from our early-stage peers even prior to the current heights of the AI boom.

Venture-backed roll-ups aren’t an inherently new phenomenon. To date, they have an uninspiring track record of delivering venture-backed returns. The most recent boom in aggregating Amazon sellers saw multiple billions invested across Thrasio, Perch, and others. The well-covered outcomes in this category have mostly been flame-outs, at least for those seeking venture-scale exits. Track records in other categories have either been similarly lackluster or are too recent to say. So why revisit a strategy that has been, with few exceptions, a poor steward of VC dollars? And why now?

We see two major reasons. First, while there is a bit of malaise in the current software market, roll-up opportunities seem prime. The post-COVID public-market premiums on profit led to underinvestment in S&M, spurring a lagging growth dip that has investors feeling constrained by SaaS upside. Conversely, an aging baby boomer population contemplating generational turnover of millions of fragmented services business, plus the need to put big AUMs amassed in 2020-21 to work, plus the potential of lower holding periods in a liquidity drought… and we can start to see the appeal of consolidation plays. Almost half of all PE deals in 2023 were of the buy-and-build variety.¹

Second, the current AI era may represent enough of a step-change to support a “this time is different” thesis. It is important to note that we’re amidst an upswell in consolidation interest independent of technology shifts—but modern LLMs undoubtedly bring a new level of credibility to the strategy. We’ll lean into this latter angle as we examine the leading arguments for and against the feasibility & attractiveness of AI-first roll-ups.

The Argument For

1. The opportunity for legacy and service businesses to recapture margin with AI is real. But just to visualize the opportunity, let’s imagine an SMB-focused accounting firm that charges its clients $50 per month and operates at a healthy 30% profit margin (meaning their total costs / client / month are $35, and profits on the same are $15). Further, imagine a novel LLM infrastructure can reduce the person-hours required to deliver a service at a set quality level by 40%, bringing total costs to $21. At the same price point, we’ve now nearly doubled net margins to 58%. Perhaps even more powerfully, we could reduce prices by 40% while still reaping our same top-tier net margin of 30%, aggressively out-competing legacy accounting firms (who don’t have the option of price-matching) while theoretically growing our overall TAM through consumer surplus, Uber-style. The magnitude of AI-driven savings, of course, is the big question, and the maximum yields from AI will inevitably vary wildly between use cases and verticals.

   

2. Some contend that buying is a more effective strategy than building when it comes to digital transformation in tech-laggard sectors. Especially in specific verticals, market structures or incentives may present clear challenges to the adoption of innovation through purchase, making the traditional startup path difficult. These problems may not be solvable by better software products or traditional software distribution strategies. In some cases, vertical software markets have become too crowded, with too little core IP differentiation. The vSaaS opportunities that do exist aren’t greenfield for these reasons—and are likely to remain untapped by software players. Alternatively, the control that comes with buying allows forward-thinking entrepreneurs or investors to enforce technology use and leverage more effective G2M strategies, rather than rely on inefficient or structurally challenged free-market software adoption.

3. Speed wins. While AI-powered software selling into the space may eventually level the playing field, those that get to market first will benefit from a first-mover advantage in two ways. First, they may be able to rapidly capture market share by offering a price competitors cannot match (at first)—in high-retention verticals such as accounting or insurance, this impact could be significant. Second, acquiring legacy businesses valued based on their pre-AI margins offers multiple arbitrage opportunities now that may evaporate when / if software democratizes the efficiency benefits and multiple roll-up platforms replicate the strategy (driving up prices). This is connected to the preceding point—acquisition enables accelerating speed to surplus.

4. Understanding the range of operations that can be automated or streamlined with AI should make it easier to underwrite potential efficiency improvements. As Slow points out in its GBO deck, “you can figure out exactly what to build by owning and operating an asset… everyone agrees, it just makes sense.” As we’ve pointed out in our last essay on early product-market fit, unintuitive and unforeseen challenges, workflows, and needs are common in industry-specific plays—a true insider view into how legacy businesses operate could help streamline development of impactful AI-powered tooling.

5. First-party ownership of data should provide a valuable corpus to train models. As mentioned in our essay, “The Future of AI is Vertical,” access to proprietary training data is a potential cornerstone of vertical AI defensibility. Even for software businesses, not all platforms will have turnkey access to customer data, especially in an era of heightening scrutiny on data privacy. If your acquired businesses throw off enough relevant data to be valuable, owners will have unfettered access off the bat.

6. Regardless of the incorporation of AI, vertical software models don’t capture enough value to achieve venture-scale in many industries. To illustrate this point, let’s return to our prior example of an accounting firm that charges $50 / client / month. Let’s imagine that firm has 1000 clients, yielding revenue of $600k (exact figures don’t matter). As a very small business, even a core software vendor would probably be capped at a fairly low ACV—let’s say $20k (3-4% of top line). Even assuming low gross margins for the accounting firm, it’s capturing an order of magnitude more “industry dollars,” making your realistic TAM much larger. Of course, at scale vSaaS penetration rates tend to be much higher than traditional services providers—but in a rosy bull case (your AI-powered accounting firm unstoppably consolidates a highly fragmented industry, racking up double-digit market share) you can envision a much larger value-capture opportunity.

A few incarnations of roll-up strategy stand out to us as most credible. (A) Test-bed acquisition(s) at the growth stage that serve to optimize the software product and opportunities for vertical integration (rather than as an ongoing growth strategy); (B) high-initial-ownership plays (e.g. incubations) that offset heavy downstream dilution; and (C) companies leveraging M&A to accelerate G2M acceleration but able to compartmentalize those efforts, keeping focus on the core software platform and maintaining the ability to divest the roll-up down the road. We will talk more about (B) in our Metropolis case study below, before introducing the most attractive alternative of all, in our eyes: the Synthetic Software Roll-Up.

The Argument Against

1. It’s critical to remember the reasons why software trades at a multiples several times higher than services businesses (even high-margin ones). The higher share of industry dollars means taking on the associated industry risks. COVID-19 hit MindBody hard but it delivered a fatal blow to Gold’s Gym. Despite construction seeing some of the highest bankruptcy rates across sectors, diversification across sub-verticals, enterprise sizes, and products allowed Procore to achieve a 117% retention rate mid-pandemic.² The software model allows the free market a hand in selecting your customers, which can be tough when structural issues mean slow uptake, but also offers insulation from cyclicality and a less fragile business overall.

2. At any appreciable scale, roll-up strategies can be difficult to execute well. While operational improvements have much more promise given AI, expertise in the selection and pricing of acquisition assets is not trivial. In 2008, Harvard Business Review found that two thirds of rollups were value-neutral or worse.³ Beating large, now ubiquitous PE firms at their own full-time game seems like a bad bet, leaving AI development and integration as a more likely moat. Not to mention that debt is more expensive in a normalized (non-ZIRP) interest rate environment and the current FTC & DOJ are bringing a new level of scrutiny to roll-ups perceived as anti-competitive practices—would-be next-gen Krafts and Waste Managements beware.⁴

3. Typical reasons why PE-backed companies struggle with innovation are likely to remain relevant: debt service crowding out R&D spend, pressure to drive net profits to drive EBITDA multiple arbitrage, and the complexities of the core operating business at scale. Keeping up with competition will not be easy in a market hungry to fund AI-powered software. After a time, the pure economic incentive to maintain not just good software but industry-leading software is weak—after all, your prize isn’t 200%+ YoY growth as it would be with a startup; it’s a few additional points of margin at best. Assuming your cost-saving AI is eventually democratized to competition, why keep up the costly R&D effort? And if you will inevitably swap in best-in-breed AI software, why build it internally in the first place when you could focus that capital on pure PE-driven implementation & integration of top 3rd party AI tooling?

4. Setting aside operational challenges, it remains unclear what the margin lift from AI will be—and opportunity across industries is unlikely to be universal. Software built internally at AI-First Roll-Ups, in some ways, are taking a page from the recent “Service as a Software” venture meme, which sees companies selling an outcome and internalizing the margin risk. I.e., they are making a bet that they can automate manual an increasing share of work over time. These have been famous last words for many startups whose margin uplift materialized too slowly because it was (i) too hard technically, or (ii) too hard to escape reliance on services once addicted to the high growth rates their venture backers demanded.

5. For most VCs, the juice simply may not be worth the squeeze. Aggregation will get you a few turns on revenue, and AI margins could perhaps push multiples a bit higher. With acquisition being the only way to reach an attractive scale, however, significant downstream dilution will make venture-scale outcomes a tall order. Larger acquisitions can make more sense than tiny ones, as Andrew Ziperski outlines in his post on VC-backed roll-ups, but require more M&A capital. The numbers could work more favorably with a non-traditional approach:

   1. Backers of platforms who make acquisitions but also continue selling software to the whole of the industry. For example, the founder never intends to continue rolling-up into scale, but rather acquires one or a small number of operating businesses to optimize the software product and workflow. Perhaps that is simply a software company that tucked in a strategic asset vs. a roll-up, but in this case the business is still software at its core, with no need for massive dilution. Or in rare cases, a founder intends to do both on an ongoing basis. Metropolis—as we will touch on below—seems to fall into this camp today. It’s unclear to us how it will work to sell software into your direct competitors but perhaps there are precedents we aren’t thinking of at the moment.

   2. Incubators and company-builders that have majority ownership going into the PE scale-up raise. A platform could build the core technology, perhaps fund the first 1-2 acquisitions as product sandboxes, and own 50-100% of the business. Then, even if a PE firm layers later on with a massive check to fuel continuing rollups, the founding firm would retain substantial equity. A strong outcome would be feasible even if the asset’s exit multiples end up somewhere between services and software. This can be difficult for most VC funds to execute, however, given the delicate dance between talent and initial equity, not to mention formation-stage and / or incubation experience. Lower initial ownership may work if valuation growth between initial funding and M&A capital is substantial enough—perhaps what the Slow team envisioned in their GBO thesis.

In our view, the strongest argument against the AI-First Roll-Up is the simplest, though perhaps most subjective. Namely, that success requires building two separate industry-leading enterprises under one roof: an AI-driven software company and an operating company (e.g., insurance brokering, dentistry, accounting). Software is about rapidly building enterprise value: build a product once and sell it to many customers, with minimal marginal cost to distribute and (hopefully) happy recurring users. Consolidation plays, on the other hand, are about financial engineering, integration, cost management, and of course the nitty-gritty of running the successful underlying operating business. As the PropTech world has seen, bifurcated models (e.g. AIco + OpCo) can bring significant organizational and capitalization complexities. But nailing one of these strategies is hard enough—executing both well simultaneously may prove a geometrically harder task.

Metropolis: A Case Study

Metropolis, a platform for parking facility management and payments, has become a central case study for the GBO and is instructive in thinking about AI-First Roll-Ups. Founded in 2017, the business set out to transform the parking industry by using computer vision to automate parking transactions and reduce reliance on onsite attendants. CEO Alex Israel, having previously co-founded and sold ParkMe in 2015, saw the opportunity for a more seamless consumer drive-in / drive-out experience, a lighter-cost solution for real estate owners, and a nice payments stream to tap into.

Despite the innovative technology, Metropolis faced challenges getting legacy players to adopt its platform. To overcome this, the company shifted to acquiring parking lots in addition to selling its software, with its most notable deal being a take-private of SP Plus (the largest player in the space) in 2023 for $1.5B. To date, Metropolis operates in 40+ major markets, has 6m+ drivers using their platform, and has raised nearly $2B to power its dual software / roll-up strategy. While not LLM-driven, the company’s strategy is very much in line with the AI-First Roll-Up theory: use tech to automate manual work and recapture margin in a legacy industry, outcompeting low-NPS incumbents, and acquiring them to drive adoption faster than traditional G2M would allow.

The rationale for their roll-up play runs that a traditional software G2M couldn’t work in this market. SaaS faces very real challenges in this vertical. Parking lots are fundamentally real estate and owners are landlords. They generate steady, profitable cash flow and don’t want to be bothered with the nitty gritty of management—hence these can be 5+ year contracts. So by acquiring the facilities themselves, they would expand TAM by capturing the whole parking management pie and accelerate access to bigger deals on the software side. This is made more difficult by the fact that parking is a non-fungible good—even a substantially better consumer experience is unlikely overcome proximity from a consumer choice POV.

A strong SaaS wedge needs to have a good leverage point with either buyers or their customers, and neither were easy to realize here. So was the initial friction more attributable to a dead-on-arrival software G2M motion or to product-market fit issues? We think it’s a worthwhile question for vertical SaaS skeptics but we can only prognosticate without being closer to the business. Regardless of the answer, given where Metropolis found itself at the time, a GBO strategy was a clever tack with promise to remove friction and reignite growth.

Perhaps the more operative question is whether Metropolis’ roll-up strategy is one to replicate. As you may have gleaned from preceding sections, we are generally skeptical of startups’ ability to dual-track software + M&A early and of the dilutive nature of raising large sums for buyouts later on. That said, what is interesting about Metropolis’ strategy is its stated intention to continue selling its technology to 3rd party facilities, rather than keep its advantage proprietary. As mentioned above, keeping up product velocity while managing a top-tier operating business will be hairy—but let’s assume for a moment they can do both well. That does open up an interesting lane: namely, to divest the operating businesses down the road. Having secured big, multi-year software contracts and taking out leaders that could’ve hampered growth, the value to tech business is clear. Meanwhile, the roll-up can trade on its improved scale and efficiency thanks to Metropolis’ tech. While it sounds like an operational nightmare and we think there are lighter-touch ways to leverage consolidation to juice adoption, it might be just crazy enough to work. And if so, Metropolis won’t be the last to adopt this strategy.

If you’re interested to learn more about Metropolis, we would suggest checking out profiles from Eli at Verticalized, Luke at Linear, and Eric Roseman (who leads strategy for the company).

The Synthetic Roll-Up

While venture capital and buyouts coexist on the same private equity spectrum—and, in the case of some firms, are intermingled—the two strategies were conceived to take very different bets. Venture capital generally seeks minority stakes in innovative businesses with high growth potential, high gross margins, and low capital intensiveness, chasing winners with big cash-on-cash returns. While buyout firms like all those things too, their model is less predicated on on growth rate and margin profile—they leverage control checks, non-equity financing, and shorter holding periods to drive IRR. If the lifeblood of VC is growth, the lifeblood of buyouts is scale.

Every so often a market dislocation or technology shift convinces venture to play in buyouts or vice versa. The LLM boom is ushering in one of those moments—venture-backed roll-ups are back with an AI twist. It’s more a credible hypothesis than most we’ve seen over the last 20 years: if AI can drive multiple expansion by recapturing margin, legacy service businesses trading at 0.5-2x revenue should look cheap. The trick, of course, is how you get from A (status quo) to B (fully integrated AI) in fragmented, late-adopter markets. To be clear, we do believe there’s significant Enterprise Value creation to be had in AI-first aggregation strategies—the vast majority of which will be vertical in nature. The question is primarily whether these strategies are good uses of venture dollars.

There is more than one way, however, for venture-backed platforms to aggregate businesses, drive efficiencies across them, and capture economy of scale. Startups that aim to maintain software product velocity and excellence while executing a roll-up face substantial complexity and operational risk. Conversely, models we classify as “Synthetic Software Roll-Ups” (SSR) avoid the bulk of these challenges, while still benefitting from industry consolidation dynamics to accelerate product and G2M.

An SSR—still fundamentally an application-layer software platform—holistically powers, helps operate, or directly contributes to the growth of legacy businesses. It brings to the table SaaS, yes, but also a highly vertical specific wedge and likely some network element that contributes to a healthier business model for the operator and / or consumer surplus for their customers. This “roll-up” doesn’t own the underlying business but it is enough of a “partner” to earn share in its upside—either a share of revenue / profit or via some other value-based model. Like a PE consolidation play, there are usually network effects to scale involved and properly choosing targets is key to the G2M. SSRs may even benefit from a GBO-lite (e.g. small pilot tuck-in or two) to fuel product development rather than scale. Their success, however, remains predicated on the uptake of the core technology platform. When the right problems in the right markets are identified, SSRs they can grow very quickly, more akin to a marketplace businesses than SaaS.

We see several archetypes as falling under the Synthetic Roll-Up concept: franchise models, certain vertical sales-enablement platforms (e.g. management company copilots), some GBOs (e.g. ones that aim to drive core value from software, despite small-scale M&A), and others within the “Vertical X” spectrum (a paradigm we love outlined by Matt Brown).⁵ All of these seek to take advantage of the same undercurrents behind AI-First Roll-Ups: the promise of AI and the power of consolidation across fragmented stakeholders in one industry. Admittedly, they lack the certainty of scale that control transactions offer. Which is why wedge products must be chosen carefully for industry-specific PMF (more on how we consider that in our last essay). But done well, they can draft off PE consolidation that’s already happening in just about every well-suited sector. By selling into and partnering with consolidators, they achieve the same win-win we described in Metropolis’ case: startup gets accelerated G2M into big accounts, roll-up gets their EBITDA lift.

Despite our excitement for SSRs, we aren’t counting out GBO or roll-up strategies. What we don’t buy is that they are necessary because vertical software (or SaaS generally) is dead. These narratives—“under-digitized markets are that way for a reason,” “vertical markets are too small,” and “X just can’t work in this market”—have gained consensus then been proven wrong, over and over again, for the last 20+ years. The eCommerce software space was already highly saturated when Shopify first started but their hyper-focus on the issues of SMB merchants won the day. Benchling, selling into R&D leaders, faced high skepticism around the difficulty of G2M into low-incentive buyers. A wave of startups failed to find the right wedge into the personal injury law space before EvenUp, which has catapulted to ~$50m ARR after reimagining their product with LLMs.⁶

As we look forward to the technology and business models that will continue unlocking the massive opportunity in vertical tech, we believe variations on the Synthetic Roll-Up will feature heavily—as in these example, opening up big, untapped avenues of value capture across our economy while staying true to the core principles that make software businesses so powerful. And perhaps AI-First Roll-Ups will too. Despite our skepticism, if smart people like the founders of / investors in the above companies are exploring something, so are we.

In future essay, we’ll elaborate on opportunities for the SSR model (and perhaps come up with a better name). We’ll also share more on our latest investment—Indie Health—which inspired our deep dive on this topic.

Thanks as always for reading Euclid Insights. For more suggested reading, check out the rest of our sources here.⁷ Know of a VC-backed roll-up we missed in our list? Please let us know and share any other feedback in the comments below!

1. Bain Capital (2024). “Buy-and-Build: Deal Breakers and Deal Makers.” Dry Powder Podcast.

2. Meritech Capital. (2020). Procore IPO S-1 breakdown. Meritech Capital.

3. Harvard Business Review. (2008). Seven ways to fail big. Harvard Business Review.

4. Federal Trade Commission (2024). FTC and DOJ Seek Info on Serial Acquisitions, Roll-Up Strategies Across U.S. Economy.

5. Brown (2024). Invisible Asymptotes in Vertical Software. Matt Brown’s Notes.

6. Palazzolo (2023). Legal AI startup EvenUp in talks to raise at $1 billion valuation. The Information.

7. Dukes (2023). Metropolis: Unlocking urban infrastructure value. Verticalized.

Equal Ventures (2023). Tech-Enabled Consolidation of Insurance Brokerages. Equal Ventures.

Federal Trade Commission. (2024). FTC and DOJ seek info on serial acquisitions and roll-up strategies across U.S. economy. FTC.

McElhaney (2019). Buyouts lead to less innovation, new research shows. Institutional Investor.

Lyseng (2024). Turning customers into investments. Cautious Optimism.

O'Malley, Bornstein, Brussell (2023). From digitally native brand to digitally native franchise: a new model. Forerunner Ventures.

Roseman (2024). You hate parking? Let’s fix it. Break the Routine.

Slow Ventures (2024). Growth Buyouts: the Dawn of the GBO. Slow Ventures.

Sophinos (2024). 069: 4 types of VSaaS companies: the new frontier of SaaS. Linear.

Sophinos (2024). 071: The hybrid VSaaS playbook: Veeva edition. Linear.

Tidemark Capital. (2024). Are tech-enabled vertical roll-ups the future or the past? Tidemark Capital.

Ziperski (2024). A few thoughts on VC-backed rollups. Andrew’s Substack.