It's the phrase you can't avoid right now if you're anywhere near small-business acquisitions: AI rollup.

The pitch goes like this. Take the dusty old buy-and-build playbook from private equity, point it at a fragmented services industry, and instead of grinding out margin through SG&A cuts, use AI to actually transform how the work gets done.

Whether any of that is true is what I'm trying to figure out.

Strip the phrase down and you've got two ideas glued together.

The rollup part is the old PE playbook. Buy a platform business, bolt on smaller competitors, make money through scale, multiple arbitrage, and operational tightening. PE has been doing this for years. None of it is new.

The AI part is where the action is. The bet is that LLMs and AI agents can now do meaningful chunks of the work humans currently do inside service businesses — customer support, scheduling, basic legal review, claims processing, bookkeeping, the lot. If that's true, you can take a 10%-EBITDA-margin services business and turn it into something that looks more like 30%. That's the entire value-creation story in one sentence.

Glue the two together and the pitch is that you get the operational predictability of a PE buy-and-build with the margin profile of a software company. You'll see this pitched as "a PE floor with a VC ceiling." It's a great line. Whether it survives contact with reality is what we're here to figure out.

The clearest framing I've heard comes from Ilya Drozdov, co-founder of Dwelly: an AI rollup is "a tech company whose go-to-market strategy is buying companies." Hold onto that.

Now, two contrasts worth drawing out, because the discourse routinely conflates this with two other things.

An AI rollup isn't a PE rollup with AI sprinkled on top. If your margin uplift comes from SG&A consolidation alone, you're doing buy-and-build. The AI has to be doing real, load-bearing work — typically by automating front-line labour. Otherwise it's just rebranding.

An AI rollup also isn't an AI-native services company. An AI-native services company starts from scratch — green-field, no acquisitions, AI from day one. That's a venture bet on building distribution from zero. An AI rollup buys the distribution pre-built and then transforms the cost structure. Same destination, very different journey.

Right. So which businesses does this actually work for?

Not all of them. The whole game is figuring out which industries the model actually fits, and which ones are wishful thinking.

A few people in this space have written usefully on which industries are ripe and which aren't. The clearest framework I've seen is from Sahil Patwa's piece on industry selection, with a similar framing from OpenOcean. The two overlap heavily, which is reassuring — it suggests there's an emerging consensus rather than one fund's theory.

Stripped down, the businesses you want to be buying have five things in common.

They're fragmented. Lots of small players, no dominant incumbent. This matters for two reasons. One, you can buy targets cheaply — small businesses trade at lower multiples than big ones. Two, there's nobody big enough to crush you while you're consolidating. Industries where ten players control 80% of the market don't work — you'd have to buy big from day one and you've got nowhere to bolt on.

The customers don't really like their suppliers. Low NPS, high churn, lots of complaints on Trustpilot. This sounds like a bug but it's a feature. If incumbents were beloved, there'd be no opening. You want industries where the bar is low and customers are quietly miserable.

The work is repetitive, human, and expensive. This is the AI bit. You need businesses where most of the cost base is people doing fairly structured tasks — claims processors, bookkeepers, schedulers, customer support agents. If the work is genuinely creative or genuinely bespoke, AI can't touch it yet, and the rollup math collapses.

The industry is allergic to software. If incumbents have already digitised, the easy wins are gone. You want sectors still running on Excel, paper, and phone calls — because that's where AI delivers the biggest cost crunch. Software-averse industries are gold.

Customers stick around. Recurring revenue, multi-year contracts, regulatory friction to switching. You don't want to fix a business only to watch its customers churn out. This is why insurance brokerage, property management, and accounting come up so often in this conversation, and why marketing agencies and event production don't.

OpenOcean put together this chart, which I think is the single most useful image in the AI rollup discourse so far. It plots verticals on two axes: how much AI can lift them, and how good the underlying revenue quality is. The dots tell you which industries are typically rolled up versus built from scratch.

A few things jump out. Insurance brokerage, property management, and customs brokerage sit top-right — high revenue quality, high AI uplift, rollup-friendly. That's the sweet spot. Investment banking is top-left — beautiful revenue quality, but AI can't really touch the work, so it's not interesting for this strategy. Customer support BPO is bottom-middle — AI can demolish the cost base, but the underlying revenue quality is so weak you'd be transforming a business nobody wants to own.

The chart isn't gospel — these are judgement calls, and reasonable people will argue about where to put the dots. But as a starting map, it's the best thing I've found.

If you're trying to assess whether an industry is interesting for an AI rollup, the question isn't "could AI do this?" — it's whether all five conditions hold at once. Plenty of industries score high on two or three. Very few score high on all five. The ones that do are the ones you keep seeing in the headlines.

If you want to play with this, there's a public industry screener that scores sectors against criteria along these lines. Worth ten minutes of your time, even just to see how the framework gets applied in practice.

Right. We know what an AI rollup is and which businesses it works for in theory. Time to look at who's actually doing it.

Enough theory. Let's look at the people on the ground.

The first thing to know is that this is a young category. Most AI rollup companies you'll come across are less than two years old, raised their first round in 2024 or 2025, and are still in the early innings of executing. There aren't yet outcomes to point to. What there are, though, are some interesting case studies of how the strategy plays out when someone actually tries to do it.

Three examples in three different industries.

Property — Dwelly. A UK rollup of letting agencies. In two years, Dwelly has acquired 10+ agencies, become one of the UK's 15 largest letting firms by scale, and raised $93m from General Catalyst and Trinity Capital — with reports it's now in talks for another $200m. The publicly disclosed numbers are striking: the first agency they acquired went from roughly 12% to over 40% EBITDA, time-to-rent dropped 30%, and maintenance resolution time fell by the same. But co-founder Ilya Drozdov has been clear that margin expansion isn't the point: "We don't need a platform acquisition to gain efficiency. Efficiency is coming from fundamentally changing how a service is delivered." The bigger ambition is to turn Dwelly into a transactional property marketplace.

Accounting — Zinco. A Spanish AI rollup of accounting firms. What makes Zinco interesting is how scrappily it started. Co-founder David Alonso Martinez has said publicly that they bought the first acquisition without any institutional equity — using bank debt, a vendor loan, equity rollover from the seller, and their own money. They talked to "100 accounting firms within 3 weeks… around 10 meetings a day" before closing. A useful data point that this strategy doesn't always start with a venture cheque.

Staffing — Pioneers. A US AI-powered staffing business. Founder Leo Shangin's path is notable because he didn't set out to build a rollup at all. Pioneers started as an AI-first service business; M&A came later, when buying agencies turned out to be faster and cheaper than organic growth. By the first acquisition, the company was at $3m of run-rate revenue. As Leo put it: "We didn't even think about becoming a rollup. Then I talked to Dan and he told me about this whole AI rollup thing. So we ultimately realized that we are an AI rollup. But it just happened."

The full landscape is broader than these three. There's a public database of AI rollup companies currently tracking over 150 of them — operators in everything from accounting and real estate to logistics and clinic admin. Worth a browse.

Three things.

Most are tiny. A handful have made meaningful numbers of acquisitions. Most have done one or two and are still figuring out how to integrate them. The category is closer to the starting line than the headlines suggest.

The strategies vary wildly. Some are buying first and bolting on AI. Some are building software first and acquiring later. Some are venture-backed and need to deliver a 10x in seven years. Some are bootstrapping with bank debt. Lumping them all together as "AI rollups" obscures more than it reveals.

Exits are a known unknown. Tenet's investor survey calls the exit market "a big unknown right now," with PE consolidators expected to be the most likely buyer (79% of respondents) and IPO a secondary route (32%).

One of the survey's anonymous respondents put it more bluntly: "We're 5 years too early for it to be generational. Plenty will get burnt in the short term." Euclid Ventures, writing on the same theme, point out that previous venture-backed rollup waves — most recently the Amazon FBA aggregators — have an uninspiring track record of delivering venture-scale returns. That doesn't mean every AI rollup will go the same way. Some will work. Some won't. We're just too early to know which is which.

So the strategy exists, real people are trying to make it work, and the early evidence is genuinely mixed. Which brings us to the question everyone's been arguing about: does any of this actually work?

This is the question the whole space is arguing about. The honest answer is that we don't know yet — but some pieces are clearer than others. Let's break it down.

Probably yes — but uneven, and the bar investors are setting is high.

The bull case is straightforward. Take a 10%-EBITDA-margin services business, use AI to automate a chunk of the front-line labour, and you can plausibly turn it into something nearer 25–30%. Euclid Ventures walks through the maths in their piece using a stylised accounting firm — drop person-hour costs by 40%, and at the same price point you nearly double net margin. The arithmetic isn't controversial.

What is controversial is whether anyone's doing it at scale yet. Tenet's investor survey found that 90% of investors want to see 2x EBITDA improvement in acquired assets to consider an AI rollup viable. That's a high bar, and almost nobody has demonstrated it across multiple acquisitions for long enough to know it's repeatable rather than one-off cost-cutting. The same survey identified "overhyped AI value-creation" as the second-biggest risk in the space, behind only integration and change management.

The lever exists. Whether anyone can pull it consistently is a different question.

This is genuinely contested.

The bull case is that buying the business is sometimes the only way to actually deploy AI inside it — because customers won't switch software, so selling it to them directly is a doomed motion. Buena, a German aggregator of residential property managers, is the clearest example. Buena started as a vertical SaaS vendor selling into the Hausverwaltung (property management) industry. The founder's own line on LinkedIn: "We didn't find product market fit. We burned through $14M over a few years." In 2023 they pivoted to acquiring property managers directly, did 20+ acquisitions, and raised $58M from Google Ventures and others. The customers wouldn't buy the software, so Buena bought the customers.

The bear case, made most clearly by Andrew Ziperski, is that rollups have a long history of disappointing returns — and that running M&A, integration, and AI deployment simultaneously is much harder than it looks. Harvard Business Review puts the M&A failure rate at 70-90%, a sobering base rate to be underwriting against even before you add the AI transformation piece. And the most recent rollup wave — Amazon FBA aggregators — has been a graveyard, with Thrasio burning through $3.4B and filing for bankruptcy after 200 acquisitions and a chaotic integration playbook. AI rollup founders are betting they can do better.

Both arguments are reasonable. Which one wins probably depends on the operator and the industry.

No — and this is worth saying clearly. Most AI rollup companies are less than two years old, no one has exited at scale, and the strategy is being underwritten on assumptions about what PE consolidators and public markets will pay in five years' time. The "PE floor with a VC ceiling" pitch is, for now, a claim — not a demonstrated outcome.

Three takeaways for someone trying to make sense of it.

The industry-selection framework is portable. The five conditions — fragmented, low NPS, automatable workflows, software-allergic, sticky customers — are useful for any acquirer thinking about AI-driven margin expansion, not just people doing rollups. Whether you ever pursue an AI rollup or not, the underlying sector logic is something to internalise.

The 2x EBITDA bar is probably not your target. Investors in this space want 2x because that's what venture math demands. If you're acquiring one business and don't need to deliver a 10x return to a fund, a 25-50% margin uplift over several years is genuinely valuable and probably more achievable. Don't let venture-scale expectations distort what's reasonable to underwrite.

The most useful signal is what the operators are learning, not what the decks claim. The founders — and others like them — are figuring out the operating reality in real time, and the gap between pitch deck and execution is going to widen as more deals close. Following a few of them publicly is probably the highest-bandwidth way to track whether this thing is actually working.

For now, the most useful thing isn't to pick a side in the bull-bear debate — it's to know enough to recognise the difference between operators who are taking the strategy seriously and operators who are repackaging buy-and-build. The next twelve months will tell us a lot, and the people watching closely will be in the best position when the dust settles.

Disclaimer: Stripped back to just the facts and a few opinions. Views are my own, sources public, none of it financial advice. Best consult a professional before doing anything with real money.