There’s a few pieces on AI Rollups floating around and I think it’s worth getting familiar with the model as it looks like a trend. This presentation has a full tear down of the model.

The tl;dr of that deck is that if you build a vertical SaaS product you can grab more return not by making pure software sales, but instead by buying businesses and then leading the transformation of applying the software to that business; this is known as the growth buyout. The oft-cited example of the model is Metropolis, who worked out number plate recognition for paid entry/exit to car parks, sold the software, then once they’d proven the opex gains started buying car parks and doing the change management themselves.

This has sparked the launch of a wave of imitators. See this analysis piece as a typical example.

There’s been a few examples in recent YC batches, and their Summer ‘25 call for startups also includes a section on Fullstack AI. Rocketable is a YCW25 batch startup following the AI Rollup model. The plan here is to purchase profitable SaaS companies throwing off cash and use that cash to bootstrap more purchases, Omaha style. The investment thesis is the application of AI/agents allows full automation of any work done by humans within these small SaaS co’s (as it’s likely to be generic one assumes). This feels tricky; the exact businesses willing to sell in this niche are likely to be those that are on paper killer small businesses today but very likely to be disrupted on a 5 year horizon, either by general purpose agents (as they are thin automation layers) or by the rapid decrease in the cost of lines of code. So the trick is going to be having a very concentrated portfolio of bets of absolute golden companies that have both AI moat and strong unit economics. Finding those companies and persuading them to sell is an interesting challenge (and likely a test of patience). There have been many doomed attempts to imitate Berkshire.

There’s also an assumption buried deep here that feels untested to me: “software costs approach zero”. From using a lot of AI development tools I think that’s unlikely to be true in vertical SaaS, mostly as the domain is so complex and you have to have a very precise understanding of the requirements that’s difficult and can’t be scraped from textbooks. You’re also typically innovating on what’s there already, and a lot of the complexity of the software is hidden from the customer in the backend (it’s not easy to clone). Not all software costs are equal and we should think carefully about where AI will win.

OffDeal is another YC backed company company that has published their blueprint for a rollup that takes on investment bank M&A. Somewhat unusually, there’s tonnes of detail in this strategy doc so I’ve pulled our a few interesting bits below to provide a tear down of how a roll up model works in practice.

First up, note how they’ve moved down market to deliberately unprofitable business that traditional M&A would not consider:

Our first battleground is lower-middle-market M&A—deals in the $5-30 m range that big banks ignore because $100-300k fees can’t support a bloated legacy deal team. A two-person pod plus AI agents does make money at that ticket size, and the high deal velocity gives us dozens of live reps each quarter.

The full strategy is then laid out:

Short-term. Remain in SMB M&A until our processes and software reach maturity

Mid-term. Move up-market, competing for $100m+ deals.

Long-term. Move even more up-market ($1bn+ deals) and add adjacent advisory services—capital raises, debt advisory - on the same platform, aiming for a full-service franchise.

They also make a strong case on the level of scrappiness they can ship with:

Because every application is internal, we can release features at “good-enough” and harden them through live use rather than long test cycles. This creates a rapid product feedback loop.

This resonated - I think this particularly suits software with probabilistic outcomes where getting consistent quality is hard. Your own team will be far more tolerant of occasional failure/the nuance of the tool, so heavy use of LLMs in this setting as opposed to say selling software into this niche makes sense.

It’s noticeable as well that the tooling needed in this niche neatly fits the jagged edge of what models are good at today, e.g. heavy use of deep research modes.

Towards the end there’s a nice call out of the problems with the model:

Launching any full-stack startup, let alone an investment bank is hard:

• Many competencies, zero excuses. We must stand up data infrastructure, AI tooling, deal execution playbooks, and a brand that convinces owners to entrust their life’s work - all at once. There is no single “core feature” we can hide behind.

• Long lead, lumpy revenue. A SaaS demo converts in weeks; a sell-side mandate must be sourced, launched, and closed before a dollar lands. We carry months of payroll while a single busted deal can wipe a quarter’s pipeline—so the capital curve is steeper than pure software.

• Skepticism by default. Industry insiders often view a VC-backed investment bank run by “tech outsiders” as Silicon-Valley hubris. Until we post results, we’re assumed wrong.

The cashflow volatility called out here feels like the big one - M&A activity is pro-cyclical and lumpy in the way it pays out. A venture model where you’re trying to build a high growth juggernaut at pace while running a loss with limited runway might not be a great fit, particularly as you scale out and opex grows (will that headcount advantage from use of agents and AI hold as the deal complexity ramps up?). Perhaps this is the other, unwritten, driver of starting with smaller deal sizes - the volume will be higher which means your pipeline is less concentrated on a few big, slow moving, bets. Also interesting how this strategy is clearly written to persuade with the two weaker, more fixable, issues either side of the filling in the s*** sandwich.

As you try to move upmarket through the niches, you suspect it might be hard to keep winning business rapidly in a market where trust is everything and buyers will be heavily influenced by brand, staying power and a lengthy track record.