This article is based on my conversation with Sam Hields, Partner at OpenOcean VC, on the RolyPoly podcast.
Most of the people building AI-native roll-ups come from finance. They've done M&A, they understand spreadsheets, and they see a structural arbitrage in fragmented service markets. That's a real edge. But Sam Hields, partner at OpenOcean and the first VC I've had on RolyPoly, comes at this from the opposite direction. He worked as an engineer at JP Morgan, ran operations at Uber, and was inside Community Fibre through a £400M LBO. Then he became a VC.
That arc shapes how he looks at every roll-up pitch that lands on his desk. Sam isn't asking "can this team execute an M&A playbook?" — he's asking do they actually understand what AI can automate, and is the front office in scope, or only the back office? That distinction sounds subtle. It isn't. It's the difference between a PE deal and a venture case, and most founders pitching AI roll-ups don't realise they're being filtered on it.
What I came away with after 70 minutes was less a tour of OpenOcean's portfolio and more a framework: a way of separating the AI roll-ups that have a venture-scale outcome from the ones that are really just back-office consolidation plays with AI sprinkled on top. Here's what I learned.
The Real Prize Isn't Software — It's the 95% That Isn't
Sam opened with a framing I think is going to define the next decade of venture investing. He referenced a recent Sequoia piece arguing that the next trillion-dollar company will be "a software company masquerading as a services company."
The logic is simple. Software is maybe 5% of the services economy. The rest is people at desks doing what an engineer would call CRUD work — creating, reading, updating, and deleting records, sending emails, running calls, updating spreadsheets. As Sam put it, "essentially a skin on a database." That's the work AI can now do, for the first time, at meaningful scale.
So when VCs argue about whether the next great company is going to come out of pure software, Sam thinks they're playing on the wrong board. The board is 10x larger and sits under the application layer. It's the services economy itself.
What's interesting is that this reframes what roll-ups are for. They're not a financial engineering trick. They're a go-to-market strategy. As Sam said: "Roll-up is just a method." You could try to sell an AI platform as SaaS into a traditional SME and watch them refuse to adopt it because the owner is "on the way out." Or you can acquire the business outright, get its clients, and re-platform the operation on AI from the inside. The roll-up is how you get to the customer when the market is structurally set up to keep you away from them — which is why the buy-and-build integration strategy only really works when the operating model on top is doing the heavy lifting.
The Thatcher Window — Why You Only Get to Consolidate Once
This was the framing I keep going back to. Sam compared the current AI services moment to Margaret Thatcher's privatisation wave in 1980s Britain — when the British economy was being called "the sick man of Europe" and the government privatised the banks, the water, the rail. There were costs, but it unlocked enormous productivity gains, and the country got wealthier.
Here's the bit that matters: you can only privatise something once. Once it's done, the latent productivity is unlocked, and the next wave of value creation has to come from somewhere else. Sam thinks the same is true of AI in services.
"Once AI has eventually transformed and re-platformed services onto AI, it's done until the next thing. So I feel like there's a bit of a time window here to get it done. And I think that's probably why a lot of people are rapidly looking at it."
If that's right, the people who are building AI-native roll-ups today aren't entering a stable market. They're competing for a one-shot consolidation wave. A lot of money will be made. A lot will be lost. But the window doesn't reopen.
That framing changes the calculus on speed. If you're sitting on capital and waiting to find the perfect vertical, the analysis itself is a form of risk. The companies that move first will set the integration standards, win the brand trust, and build the operating moats.
The Filter Sam Applies: Front Office or Back Office?
Here's the question that separates a venture case from a PE case. Sam was direct about it: if you can only automate the back office, you're in PE territory. If you can automate the front office too, that's a venture case.
He used scaffolding as a counterexample. Scaffolding is fragmented, ubiquitous, hard to staff, and trades at low multiples — it ticks a lot of roll-up boxes. But the labour in scaffolding is people climbing up structures and putting them together. AI without robotics can't touch that. So even though the macro setup is attractive, the unit economics don't transform.
Compare that to building management, which OpenOcean has invested in. Sam's view: in building management, AI today can attack both ends — the back office (finance, admin, scheduling) and the front office (customer comms, tenant relationships, contract management). The entire pie is addressable. That's what makes it a venture case, not just a leverage-and-strip-cost PE deal.
This is the filter most founders miss when they pitch. They get the back-office story right — "we'll consolidate the finance function, reduce headcount by 40%, expand EBITDA from 15% to 35%." Sam is asking the next question: what happens to the people doing the customer-facing work? If the answer is "nothing, AI can't really touch it," then the multiple expansion thesis is real but the venture-scale upside isn't there.
The Founder Triangle: Tech, Domain, and M&A
When Sam talks about red and green flags in pitches, he keeps coming back to a triangle of skills you have to have on the founding team:
- Tech / product expertise — someone who actually understands what AI can do today and what's coming in six months. Not someone who has read about it.
- Domain expertise — operators who've lived in the vertical and know where the work actually is.
- M&A discipline — someone who's done acquisitions and can run the transactional machine at pace.
What surprised me was his ranking. He thinks the tech is the least concerning piece. Not because it's unimportant, but because in the age of foundation models and accessible APIs, the engineering required to extract real productivity gains has never been more tractable. ChatGPT made AI accessible to everyone, and increasingly AI itself can help you ship.
Sam said that if he had to rank the three skills by how replaceable they are, he'd actually rank the tech as the most accessible — and the domain expertise as the scarcest. That's a contrarian take, especially coming from a VC who came up as a software engineer. But it explains why he keeps pushing founders to prove they understand their vertical before they impress him with a model demo. A great LLM can be rented. A great operator in compliance, in building management, in legal — that person is rare and they're the one who tells you which workflows are actually automatable and which ones the regulator won't let you touch.
The M&A piece is the third leg. Not glamorous, not rocket science by Sam's read, but if you're trying to do 10-20 acquisitions a year — and that's roughly the cadence he's seeing from the companies that get into stride — you need someone who runs that machine without breaking it. The integration question is the one that follows immediately after, and it's where most roll-ups stall — which is exactly why a disciplined post-merger integration checklist matters more for AI-native roll-ups than the prettier numbers in the IC memo. We see this repeatedly at PMI Stack: founders close the deal and then discover that the AI transformation they underwrote takes 18-36 months of operational lift to actually realise. Without an operator who can stand inside the acquired business and drive that change, the EBITDA expansion stays on the model.
Capital Stack: Why Most VCs Are Wrong About Debt
This is where Sam gets the most interesting, and most contrarian. He's clear that using venture equity to acquire profitable, cash-flowing service businesses is one of the worst trades a founder can make.
The logic: equity is your most expensive capital. If you raise it at a startup valuation and use it to acquire a target at 3-5x book, you're effectively paying multiple times the underlying value with the most diluting capital you'll ever take. That math is brutal for founders, and it doesn't make sense from the investor side either if there are cheaper alternatives.
The alternative is debt — typically one-times leverage in the early days, meaning if the platform is doing £20M in turnover, you can probably get £20M in debt. The acquired companies throw off cash. Acquisitions land that immediately improve EBITDA. The leverage is appropriate to the business being built.
Sam acknowledged the obvious counterpoint: General Catalyst's approach is the exception. They've been public about not wanting their portfolio companies to take debt early, because debt forces a more risk-averse posture that flies in the face of swashbuckling, high-growth venture behaviour. Sam respects that argument but sees it as a function of GC having "billions to deploy" — they have to put equity to work. Most VCs aren't General Catalyst.
He's also watching the holdco vs. fund structure debate play out. A traditional vintage-based VC fund (10+10 years) wasn't designed to hold and operate roll-ups indefinitely. Holdco structures let you raise differently at different levels — equity at the holdco, vertical-specific equity at portfolio entities, debt sleeves layered in for liability protection. Sam suspects the holdco model wins out, and the next generation of roll-up financing will look more like the financial engineering you see in PE than the clean cap tables venture is used to.
The other piece he flagged is rollover equity in acquired-company sellers — increasingly common, increasingly smart. The first acquisition into a new geography isn't just a deal, it's a beachhead. You're not buying a business so much as buying the general manager who'll run that territory for the next decade. Rolling them into holdco equity aligns them on the long-term outcome instead of the short-term earn-out, and it preserves the founder's cap table for the harder fights ahead.
Verticals to Watch (And Verticals That Are Already Crowded)
I asked Sam where the white space still is. He was careful — he doesn't want to turn buyer's markets into seller's markets by saying the quiet part on a podcast. But he gave me three frames I think are useful.
First, accounting and bookkeeping in Western markets are over-discovered. Sam jokes that OpenOcean gets pitched an accounting roll-up roughly every week. It's an obvious vertical because finance people feel they understand the work. But that obviousness creates a problem: in Western and Northern Europe, an AI-enabled accounting roll-up is up against lower-mid-market PE plus every other AI-enabled roll-up that's already entered the market. Sam called that "quite a worrying kind of dynamic." That doesn't mean accounting is dead as a thesis, but the geography matters. The same play in Southeast Asia or other under-discovered geographies looks very different.
Second, look for resilient revenue. A holiday lets roll-up in Dubai or the Gulf right now would be vulnerable to anything that disrupts travel — geopolitics, economy, weather. Accounting, conversely, is regulatorily anchored: come rain or shine, businesses have to file their accounts. That regulatory stickiness is what makes the revenue VC-shaped. Sticky, sub-cyclical, ideally with an annual renewal cadence that feels like ARR. Building management is another example he keeps coming back to for the same reason.
Third, watch what AI does to the bottom of the TAM. Sam gave a great example from law: lawyers can't economically get out of bed for a £500 piece of work because their cost structure won't price below a certain floor. So they turn away clients who'd happily pay £500. AI-enabled legal services can serve that bottom segment profitably, expanding the TAM downward. Some of those clients then graduate to bigger work over time. The trade-off, he flagged, is that pricing may also collapse in segments as AI commoditises certain work — so the same dynamic that expands TAM at the bottom may compress prices at the top. Net effect: probably positive for share, ambiguous for revenue. It echoes what Vadim Rogovskiy described on an earlier RolyPoly episode — agentic AI doesn't just compress cost, it rewrites where the revenue actually sits.
What This Means If You're Building or Backing One
If you boil down Sam's thesis, the actionable test is this:
- Is the front office in scope? If AI today can only attack the back office of your target vertical, you might still have a fine PE deal — but you don't have a venture case, and you probably shouldn't be raising venture money.
- Do you have the triangle? Tech, domain, M&A — all three, and don't underweight the domain operator just because the tech is more visible.
- Are you using the right capital? Equity to fund operating losses and platform infrastructure. Debt — when you're ready — to fund the acquisitions of cash-flowing assets.
- Are you optimising for the window? This is a one-shot consolidation moment. Speed isn't optional; analysis paralysis is a tax.
- Is the revenue resilient? Regulated, sticky, sub-cyclical revenue is what makes the underwriting hold.
The companies I've spoken to who are getting this right are not the ones with the slickest decks. They're the ones whose founders can walk into a room and explain, in concrete terms, which 100 workflows are now automatable, what they'll do with the headcount that comes off, and how they'll keep the customer-facing work feeling human even as the back end becomes autonomous. That's a different conversation than "we're going to roll up X vertical."
Sam closed by saying we're going to be counting this stuff in months, not years. The window is open now. It won't be open forever. If you're building or backing in this space, that framing is worth keeping front of mind.
