Roll-Up Pitches Got Louder. Operator Alpha Got Smaller.

Two stories on the desk this week. First: a Yale case study asks whether operator alpha shows up in search-fund acquisitions. Short answer: margins fell, and the authors say a higher multiple did most of the work. Second: UBS published its Next Generation Report. More than 70% of heirs intend to change their advisers upon inheriting. Different industries. Same shape. In both, the marketing version of the playbook is louder than the data can support.

Operator alpha was rare in the data

If you've sat through an AI-enabled roll-up pitch in the past 18 months, the deck rests on a single assumption. AI does the operating work that historically wasn't happening.

That's the load-bearing assumption.

Yale just tested whether it ever was.

A Yale case study, summarised by Yale Insights on 30 April, looked at where the value came from in 59 search-fund companies that had been sold. EBITDA margins didn't expand. In aggregate they contracted, from 25% at purchase down to 19% at exit. Revenue grew in about 90% of the firms and EBITDA dollars rose in about 70%, but margins fell in about 70%.

The authors say roughly 80% of the rise in enterprise value came from a higher EBITDA multiple, 6.3x at purchase against 15.6x at exit. That is enterprise value, not returns to the funds, and the authors leave debt out. They also say falling margins are not necessarily a sign of poor performance, because a CEO can be investing in growth on purpose.

Stanford's 2024 Search Fund Study reports a 35.1% aggregate pre-tax IRR and 4.5x on invested capital across 681 qualifying search funds in the US and Canada. Yale's sample is a different one and covers enterprise value only, so it can't say how much of that return came from the multiple. In the Yale sample, the value didn't come from better margins.

The pitch implied operator alpha. In the Yale sample margins fell in aggregate and in about 70% of firms. The authors say their data indicate that, in aggregate, search-fund CEOs are not superior operators in the sense of getting more EBITDA out of each dollar of revenue.

Look at how the same strategy is getting underwritten today. Tenet's 2026 investor survey found that 86% of its 102 responses named AI-led margin improvement as a value-creation lever, the most of any lever. The shares add up to more than 100%, so respondents could name more than one: 46% named multiple expansion. About 80% of the responses came from venture capital and growth equity investors.

90% of respondents would consider a 2x EBITDA improvement on acquired companies enough for an AI roll-up to be viable. Those are expectations. The survey measures sentiment. Tenet is an investment firm that backs AI roll-up founders, so read it as one side of the argument.

The same survey ranks integration and change management first among the risks, named in 79% of responses, and "overhyped AI value-creation" second, named in 68%. The trade is in the gap: AI-led margin is the most-named lever, and overhyped AI value-creation is the second-ranked risk. That tension is worth pricing.

Pushback is real. General Catalyst, an investor in both companies, has put numbers on the AI-margin case. It says Crescendo is taking gross margins to 60–65% or more as it rolls out its AI platform to PartnerHero's 200 customers. It also says Dwelly, a UK property-management roll-up that has bought six agencies so far, has doubled EBITDA margins at the agencies where its technology is fully deployed.

Named companies, specific numbers. The Yale dataset is historical, and operators might be doing better now than the record suggests.

The numbers come from the GP side, though, and they cluster in services verticals where AI replaces repetitive call-centre and transactional work. The essay gives no starting margin for Crescendo. For Dwelly it gives no margin levels or period, and it doesn't say how many of the six agencies are fully deployed.

And the sharper version of the objection is that when language models get deployed market-wide, the work gets cheap. The margins don't get fatter. The savings flow to customers, not to whoever runs the business.

That's the part no pro-forma captures. The AI-margin gains and the multiple-expansion premium compress at the same time. Once AI is table stakes, buyers stop paying up for "AI-enhanced" platforms.

Most pro forma tests the AI-margin assumption the hardest. In the Yale sample, the authors say most of the rise in value came from the multiple, which went from 6.3x at purchase to 15.6x at exit. A roll-up that relies on the same multiple expansion, with no AI-margin gain, is making the bet search funds made. That is a bet on the exit market.

The authors make the same point: if multiple expansion becomes unsustainable, they say IRRs will likely tumble and operational excellence will matter more. And with more roll-ups in the market, more sellers will be chasing the same exit window. We covered the structural case in the Founder's Guide to AI-Enabled Roll-Ups. The Yale data just sharpens which question to actually ask.

One way to stress-test the maths sits outside the survey. Commercial Capital's standard teaching example: assume an optimistically high 90% chance of any single bolt-on integration working. Compound that across five deals and the chance all five land falls to 59%. The point isn't the number. 90% per deal feels safe. 59% across the strategy doesn't. Same assumption, different feel. The more deals in the plan, the harder the maths gets, even when each individual deal still looks like a 90.

Operating skill in small-business acquisition is genuinely scarce. Most owners couldn't articulate what made their business work, even when asked. New operators keep finding stuff post-close that an experienced operator would have caught in diligence. AI doesn't fix culture clash, customer concentration, key-person retention, or operating-system mismatch.

Tooling is not the operator who wasn't there.

The "two-thirds of roll-ups fail" number you keep seeing isn't from a peer-reviewed study. It's industry folklore. But the compounding maths and the Yale data are saying the same thing. Operating improvement at scale is the hard part. Owner relationships, customer concentration, integration friction: none of that gets cheaper because tooling got cheaper. The pitch is louder. Operator alpha was always rare.

70% of heirs plan to change advisers

Different industry, same shape. UBS just published its Global Next Generation Report. More than 70% of heirs intend to change advisers. The report's framing is wealth-transfer continuity: $83trn will move over the next 20–25 years, and most of the relationships managing it won't survive the handover.

Read it from the founder's side instead. It's not a transition risk. It's a verdict.

Founders in the $5M–$100M bracket are tech-native by default. Every other product we use is AI-first. Wealth management is one of the few categories that's actively gone backwards. Advisers spend roughly 80% of their time on admin and 20% with clients. Only 10–20% of the tools mapped on the Kitces Landscape have mature, publicly accessible APIs. Era registered as an AI-native RIA in March, specifically because traditional firms can't reach the mass affluent economically. Nearly four in ten next-gen-led families now run a single-family office. That's often the only structure where they can buy the tooling and time allocation they actually want.

Then there's the peer layer. Hampton has more than 1,000 founder members and around $8m in ARR. The Moneywise podcast pulls 20–40k downloads per episode. Those numbers are what founders are paying for the conversations the wealth-management industry stopped having.

The answer is hybrid. AI-native tools for the operational work. Peer community for the frameworks and patterns. Specialist humans for the regulated work that actually needs them. The Founder's Guide to Building a Private Investment Office walks through one route into that stack.

That 70% adviser-change number is the next generation voting against the stack they inherited.

On the Radar

Hedge funds bought the platforms. Retail held the gates. Saba Capital, Boaz Weinstein's $6bn fund, went long Apollo, Ares, and Blackstone, shorted the weaker non-traded BDCs, and offered to buy retail stakes at 30–40% below NAV. The trade is on dispersion inside private credit, not against the asset class. The question worth taking to the wealth manager: which side am I on, and how fast can you tell me. Read more →

A 2011 SEC carveout built $100m of Kevin Warsh's wealth. His Senate Banking disclosures highlighted a rule that allows single-family offices to treat senior investment professionals as "family clients." They co-invest alongside the principal family on the same terms, no adviser registration required. Warsh's $100m alongside Druckenmiller's office is the case study. For founders building a sub-$100m SFO, this is the talent mechanism that closes the comp gap with hedge funds. Worth a chat with FO counsel before the next senior hire. Read more →

89 new UHNWIs every day. The US share is rising fast. Knight Frank's 20th Wealth Report puts the global UHNWI population at 713,626, up 162,191 over 5 years. The US created 41% of those, lifting its global share from 33% to 35%, on track to 41% by 2031. For UK and EU founders sitting with the "should we consider US residency" question, the report quantifies the gravitational pull rather than just describing it. Read more →

Direct-deal minimums are collapsing toward $250k–$500k. Saul Wealth Advisors reports minimums coming down from a historical $5m–$10m floor. At the bigger end, Arena Private Wealth co-led a $230m round into AI chip company Positron and took the board seat the VC would normally occupy. Both ends open new access for $5M–$100M founders. The question for the wealth manager: what's the deal selection process, who has board representation, and what's the override on top of the deal economics. Read more →

Family-office wealthtech is consolidating fast. 65% of FOs still run on spreadsheets. Aggregation and reporting tooling has settled around five platforms (Aleta, Eton, Addepar, Asseta, Masttro), each leaning differently on AI document automation, accounting depth, or Principal-facing interfaces. For a sub-$100m FO, the question is simple. Are you paying staff to retype PDFs or to think about portfolios? Sometimes the right answer is software, not a person. Read more →

Sahil Bloom's "New Opportunity Razor." Bloom worked it up after a weekend with James Clear and a few other writers — a filter for which opportunities to say yes to and which to pass on. It's the post-exit version of a problem most founders hit fast. The inbox fills up with angel pitches, board seats, business ideas, advisory roles. Each costs the same currency: calendar and attention. The reader test: does the framework survive application to the last 3 opportunities you actually said yes to? Read more →

The questions that matter when next week's pitch lands

The thread between both stories: the pitch is louder than the data behind it. AI roll-ups get pitched on operator alpha, which was rare in the data. Next-gen wealth is sold on the promise of adviser continuity, even though 70% of heirs have already decided to break it. Founders who do well in this environment pressure-test the assumption underlying the pitch, not the pro forma on top of it.

Two questions land harder than any due diligence checklist when next week's deck arrives. What does this assume about the operator? What does this assume about the relationship? If the answer to either is "the one that historically wasn't there," that's the gap to price.

New on the Site

Last Thursday's article looked at the same problem from the other side. Founders who built companies have habits that work for building, then break the moment they start investing. The gap between operating skill and allocator skill is wider than most expect.

Read it: Why Smart Founders Make Terrible Investors