Road To Carry is proudly supported by Reef Pass Investors
Reef Pass is a US based investment partnership that specializes in launching serial acquisition and HoldCo platforms (most often launched by ex-PE professionals).
Reef Pass brings decades of experience building acquisition platforms and flexible mandate across incubation and growth stages with a long-term capital base.
If you've ever thought seriously about launching your own acquisition platform, I'd love to hear your thesis!
Note: this partnership does not impact editorial independence.
A few weeks ago, we sat down with Ed Brandman (former CTO of KKR) to talk about how AI is quietly changing the way PE firms themselves operate.
Today, we go one level deeper into the stack — how AI is changing the companies PE (and VC-backed) firms are actually buying.
I recently sat down with Ilia Drozdov, Co-Founder & CEO of Dwelly, a UK-based AI-powered property management platform that:
Recently raised $93 million, led by General Catalyst and Trinity Capital
Made 10+ acquisitions in under two years
Punchline? His first acquisition went from 15% to 40%+ EBITDA margins leveraging AI. But he’s got bigger ambitions—instead of optimizing margins, he’s funneling everything into growth so he can become the dominant player and eventually evolve into a tech-enabled marketplace.
This one is for anyone who has ever rolled their eyes at the phrase "AI-enabled roll-up", or anyone tempted to start one.
State of Play
If there's one thing Ilia wants you to take away from this conversation, it's this:
AI-enabled roll-ups are not "PE roll-ups with a little AI on top." They are tech companies whose go-to-market strategy happens to be acquisitions.
We talked about:
Why AI-enabled roll-ups are structurally different from PE roll-ups
Concrete operational wins in tenant screening, maintenance, and pricing
Why the first Dwelly agency went from 12% to 40%+ EBITDA (and why that's not the point)
How AI-enabled roll-ups approach integration and change management
Which fragmented services industries actually fit the playbook
If you prefer to listen to this interview, you can find it here.
If you’re new to AI-enabled roll-ups, the core thesis is entirely different from traditional private equity. Financial engineering and scaling back-office costs are secondary. The primary goal is fundamentally redesigning how a service is delivered to the end customer from the ground up.
Here is my conversation with Ilia on what it takes to build a modern AI-native acquisition platform.
1. The biggest misconception about AI-enabled roll-ups
Edwin: Let's start with the myth. What do investors and operators get wrong about AI-enabled roll-ups?
Ilia: The biggest misconception is that an AI-enabled roll-up is just a slightly better private equity roll-up. That it’s a PE play with a little technology on top.
That's wrong.
At heart, an AI-enabled roll-up is a tech company whose go-to-market strategy is buying companies. It reimagines how the service is delivered to the end customer. A traditional PE roll-up, by contrast, is about efficiency of scale and consolidating under a platform acquisition.
We don't need a platform acquisition to gain efficiency. Efficiency is coming from fundamentally changing how a service is delivered.
2. Why AI was the thesis from day one
Edwin: You started research in 2023 and made your first acquisition in April 2024. What made AI central, rather than a bolt-on?
Ilia: When OpenAI released the first public versions of ChatGPT, we had an aha moment.
In property management, most of what an agency does is communication. Whether with tenants, landlords, maintenance providers, and local authorities, bulk of the agents’ time is spent on email and phone. If communication could be elevated by AI, the whole service model changes.
To be honest, we didn't foresee how powerful the models would become. We were building on the assumption that AI would enhance what humans do. Today, AI can remove entire administrative workflows, freeing agents to focus on relationships and higher-value advice.
That matters, because in the UK, letting agents (akin to property managers in the U.S.) are among the top three least trusted professionals. After acquiring more than 10 agencies, I can tell you: that isn't because they're bad people. It's because the nitty-gritty admin— things like referencing, compliance certificates, hundreds of tenant applicants—crowds out the actual value work. You end up chasing the gas certificate instead of helping the landlord think about portfolio strategy.
AI changes that ratio of where agents spend their time.
3. Where AI is creating real operational impact
Edwin: Give me specific examples. What tasks used to be human-only, and what's different now?
Ilia: Start with how properties are marketed.
In a traditional UK agency, the agent lists the property on Rightmove, Zoopla, or OnTheMarket, the three big classifieds. Inquiries come in by phone and email. The agent answers each one manually, runs a live questionnaire, schedules viewings, and runs viewings.
Every call, every email, every viewing costs agent-time. The incentive becomes: find the first good-enough tenant, close, move on.
With AI, the marginal cost of processing an inbound inquiry is effectively zero. Results at Dwelly:
Traditional agency: 1–2 offers per property before signing
Dwelly: 10 validated offers within 3–5 days
Time-to-rent reduced 30%: from 3–4 weeks to under 2 weeks
~5% higher rents on average (via bidding, though UK regulation changes this from May)
The agent's role shifts. Instead of admin, the agent now engages a pre-qualified top-10 tenant shortlist. Humans still verify intent because AI isn't reliable at that last-mile verification yet. Better outcome for the tenant (fair shot), better outcome for the landlord (more and better choices), and the agent still owns the decision.
On maintenance side of property management. the industry average resolution time for a non-urgent request in the UK is 52 days. 52 days for a leaky tap to get fixed.
At Dwelly, we've already cut resolution time by 30%, and we're targeting 10 days. The mechanism is unglamorous: AI maintains full visibility across tenant, provider, and office, auto-follows up, and reassigns providers that don't respond. That's the whole trick.
4. The financial impact, but why it isn't the point
Edwin: What does all of this do to margins?
Ilia: On our first agency, we took EBITDA from roughly 12% to slightly above 40%.
But margin expansion isn't the primary goal.
In every new acquisition, the first question the team asks is: "Are we out of a job because AI is going to replace us?" That's the common misconception. Yes, we could push efficiency hard and expand margins aggressively. But the much bigger opportunity is reinvesting the efficiency gains into elevating the service. As in, giving the agent time to develop deeper landlord relationships and win more market share.
That compounds. We think we can build a $50 billion business in the UK alone, and potentially $200+ billion across Europe.
Compare that to the biggest PE-backed UK property management roll-up today, which has maybe 3–4% market share and is treated as a billion-dollar exit candidate. We're not playing that game.
5. The real customer of the AI is the agent
Edwin: Listening to you, it sounds like the primary user of the AI isn't the landlord or the tenant. It's the agent.
Ilia: Exactly. And our posture on client-facing AI is deliberately conservative at this stage. But there’s been real impact across the board.
Tenants: NPS is above 80. On average, out of 500 tenants moving in, maybe one raises a process complaint.
Landlords: We barely use AI in landlord-facing interactions. Landlord communication is a small fraction of total agent hours, and the relationship is inherently human. These are 10-year-plus relationships around a landlord's largest asset. You don't shock that system. Baseline NPS at acquisition is usually 55–60 (already high), and we see a 5–10 point improvement within six months.
Agents: This is where change management lives. After an M&A event with AI transformation, people are rightly anxious. I've been in their shoes when Uber (where Ilia started his career) merged with a local competitor, my team needed a month just to process what it meant for them. Same pattern here. That's why we have a dedicated integrations team, and why we measure success by team alignment, not timeline.
6. What makes an agency acquirable for this AI thesis
Edwin: How do you evaluate whether an agency is a fit for Dwelly?
Ilia: Some of this is standard. Some isn't.
Business criteria:
Predominantly rental-focused. UK agencies often dabble in sales, block management, and financial advice. We only buy agencies whose revenue is largely rentals.
Diversified customer base. We don't buy agencies where half the portfolio is one landlord. The risk is obvious.
The less obvious criterion:
Strong succession. The typical UK deal is retirement-driven, with a founder who spent 20–30 years building the business is stepping back. We only buy where there's a strong #2 ready to become Managing Director post-close and lead the team through a transformation that fundamentally changes how they work day-to-day.
7. Integration: 3–4 months, not 30 days
Edwin: Walk me through the first 90 days. How does it compare to a PE integration?
Ilia: You might be surprised at how similar the chapters are. The steps are: announcement prep, back-office migration (finance, HR, bank accounts), process mapping, system rollout, data migration. Where we diverge from a standard PE roll-up is emphasis.
Day 1 announcement: We want the team to feel welcomed, as opposed to being signaled to expect the classic PE playbook of losing the brand, the office, and senior management within a few months.
First month: We sit with the team, answer questions, and do all the back-office work in parallel. Process mapping is collaborative. We explain our operating model, we learn theirs, and sometimes fold their approach into Dwelly's framework going forward.
Timeline: Smaller integrations run 3–4 months. Our most recent acquisition, an 11-branch, 80-person business for example, runs longer. We measure success by alignment and portfolio health, not by days.
A PE roll-up in our space typically integrates faster. Usually closer to a month of intense work with some cleanup after. I recently saw an agency sold in my neighborhood where the office was closed inside three weeks.
We don't do that. With AI, maintaining tens or hundreds of brands is feasible. Local brand equity matters; we don't need to consolidate it.
8. The end state: a transactional marketplace
Edwin: Where does this go long-term? Is AI roll up just a race to the bottom?
Ilia: Eventually, we want to build an end-to-end transactional marketplace covering every transaction between landlord and tenant, whether we own the agency or simply facilitate the transaction for another.
The inspiration is Beike in China, which was originally a large traditional agency, now a publicly traded marketplace trading at roughly a 30x EBITDA multiple on the NYSE. Similar models exist in Brazil with QuintoAndar.
For us, the roll-up is a cold-start problem. Meaning, it’s a way to solve the chicken-and-egg problem of starting a two-sided marketplace. The roll-up is the means. The marketplace is the end.
That framing is also why the VC side has taken an interest.
9. Which industries fit the AI roll-up model?
Edwin: Not every fragmented services industry is a candidate. How should people stress-test their own AI roll-up thesis?
Ilia: Applying AI isn't rocket science, but building foundation models is. You'd be naive to think your competitors won't deploy AI in your space, your country, or both.
So the question becomes: what's the moat that prevents fee and margin erosion once everyone has AI?
Go back to fundamentals:
Brand: (less relevant to us, but powerful in some verticals)
Network effects: including offline ones. Our internal "TaskRabbit" of local contractors gets better with scale, and that alone defends us from smaller competitors.
Distribution and customer stickiness: Landlords don't switch agencies casually.
If your AI-enabled thesis doesn't have at least one of those, you'll enjoy short-term profits and then get competed down to commodity margins.
Behind the Curtains Sneak Peek
Road To Carry is cooking up an AI for PE content series.
So many of you have asked how AI is actually being used inside PE firms and portfolio companies. Honest answer: I'm not the expert. And after fielding a steady stream of cold outreach from random AI vendors, I decided to roll up my sleeve and find someone I trust.
In the coming weeks, we'll launch the series with an ex-PE investor turned product manager at one of the leading AI-for-PE companies.
Topics in the works:
AI in PE, where do I start?
AI as your first-pass analyst: getting 80% of a market map done before lunch
How to write a good prompt (introduction to prompt engineering)
From company name to qualified target in 5 minutes: the AI enrichment workflow
Finding companies before they're on anyone's radar: AI-powered deal origination
Anything missing? Hit reply and let us know. Excited for this? Also let us know.
Any topics I should cover next? Share thoughts with [email protected]
Were you forwarded this newsletter? Subscribe here.
Make sure you receive Road To Carry every Thursday and Saturday!
Everyone: reply to this email with a “hi”
Gmail mobile: Click the 3 dots (...) at the top right corner, then "Move," then "Primary"
Gmail desktop: Go back to your inbox and move this email to the "Primary" tab
Other users: Follow these instructions