Will AI Improve Returns on Search Fund Acquisitions?

What the Search-Fund Research Says

By way of background, the 2026 Stanford Primer on Search Funds is a must-read practical guide for anyone interested in search funds or entrepreneurship through acquisition. Because searchers are first-time business buyers with limited operating experience, the primer identifies desirable characteristics for a target company, including a competitive advantage or moat, high recurring revenue, multiple avenues for growth, and a reasonable valuation.

The primer also describes the post-closing period as one of learning and evaluation, followed by attention to systems, controls, reporting, and operational improvements. With its primary focus on overcoming common pitfalls in the early transition period, the primer does not go into much depth on what happens as the CEO begins evaluating growth plans.

The 2026 Stanford Search Fund Study provides an updated view of search-fund returns since the 2021 report and company characteristics over time. It reports that recently acquired companies often show lower returns, in part because their CEOs are early in their tenures and have not yet had time to make substantial operational changes.

With the most recent cohort, target purchase prices increased to a median of $16.0 million—the second-highest level on record—while the EBITDA multiple paid fell to 6.2x, from approximately 7x in the two prior cohorts.

Returns vary by cohort. Very recent acquisitions showed lower returns, which is typical when CEOs are early in their tenure and have not yet had time to make substantial operational changes. As time progresses, returns typically grow. For example, the 2021–22 cohort improved from the prior study, increasing ROI from 1.5x to 2.7x and IRR from 23% to 30%.

Outcomes from older, completed cohorts ranged widely. The earliest cohort, consisting of companies acquired in 1996 or earlier, achieved 6.7x ROI, while the next cohort, from 1997–2000, returned less than half of that—2.7x ROI. While the study does not go into detail on the 1996–2000 period, it is reasonable to wonder whether those companies were affected by the high valuations during the dot-com bubble.

To my knowledge, A. J. Wasserstein, a professor at the Yale School of Management, is the only person to have published substantive research focused specifically on search-fund value creation and returns.

In On the Nature of Entry Multiples, Wasserstein cautions against focusing too heavily on the purchase multiple. Price matters, but a low price cannot rescue a business that does not grow. Over a longer holding period, the quality of the business and its ability to produce compound growth over time matter more than whether the buyer saved an additional turn of EBITDA at entry.

In Exploring Search Fund Entrepreneur Economics, Wasserstein and his coauthors use an illustrative five-year model to show how different variables affect the searcher’s potential payout.

In the base case, a 5.0x entry multiple produces a modeled payout of $2.61 million. Reducing the entry multiple to 3.0x increases the modeled payout to $4.39 million. But growth also matters. Increasing annual EBITDA growth from 7% to 11% raises the modeled payout from $2.61 million to $4.02 million.

This study used a model, not an empirical data. Its value is in showing that both purchase price and compound growth can materially affect the economics. It also illustrates why a searcher cannot evaluate a deal by looking only at the entry multiple.

Wasserstein’s recent research, A Mathematical Analysis of Value-Creation Attribution in Search Fund Projects, analyzes 59 fully exited, traditionally funded ETA companies. The sample represented approximately 35% of the known traditional search-fund exits included in the 2024 Stanford study, although the samples are not perfectly comparable.

Across the sample, enterprise value grew from approximately $996 million at entry to nearly $4 billion at exit. Using a mathematical decomposition, the authors attributed approximately 80% of that increase to expansion in the EBITDA multiple and approximately 20% to growth in EBITDA dollars.

In plain terms, enterprise value is generally calculated by multiplying EBITDA by a valuation multiple. A company can therefore become more valuable because its EBITDA increases, because the market applies a higher multiple to that EBITDA, or because both occur at the same time.

The EBITDA growth itself came from two opposing forces: revenue growth contributed approximately 39% of total enterprise-value creation, while EBITDA margin contraction reduced value by approximately 18%.

This finding is surprising. It suggests that, in this particular sample, the increase in enterprise value was driven more by the valuation multiple applied at exit than by growth in EBITDA. It does not mean that search-fund CEOs created 80% of the value through operational improvements, nor does it mean that buying well or improving the business is unimportant.

Multiple expansion can reflect many factors, including company growth and scale, market conditions, interest rates, buyer demand, strategic value, and the quality of the sale process. The analysis also cautions that it cannot cleanly separate CEO skill and effort from broader market conditions. It identifies a relationship between the components of value growth, not necessarily a single cause.

It is nonetheless important that multiple expansion accounted for approximately 80% of the increase in enterprise value. From an investor’s perspective, the finding is a reminder that search-fund returns may depend not only on operating performance, but also on the conditions under which the company is ultimately sold.

What We Know About AI’s Impact

The next question is whether AI can improve one or more of the operating variables that influence a company’s performance and, ultimately, its value.

Because AI adoption is still relatively new, we do not yet have reliable data on AI-driven returns for search-fund acquisitions. However, several studies provide useful clues in related settings.

In Generative AI at Work, a field study of 5,172 customer-support agents examined what happened when employees received access to an AI assistant. The tool increased the number of customer issues resolved per hour by approximately 15%, on average.

The results were not uniform. Less-experienced and lower-skilled employees saw the largest gains in speed and quality, while the most experienced and highest-skilled employees saw smaller gains in speed and small declines in quality. The study also found evidence that AI assistance helped disseminate the practices of more effective employees and facilitated worker learning.

In The Uneven Impact of Generative AI on Entrepreneurial Performance, researchers randomly gave 640 Kenyan small-business entrepreneurs access to a GPT-4 business assistant through WhatsApp. The study found no statistically detectable average effect on business performance. But the results varied significantly by entrepreneur: higher-performing entrepreneurs improved by roughly 15%, while lower-performing entrepreneurs performed approximately 8% worse.

The lesson is not simply that AI is good for some businesses and bad for others. Results depend on how people select, interpret, and implement AI advice. The same tool can produce different results depending on the task, the quality of the employee, and the quality of the underlying business.

A foundational study, Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence, extends the evidence to more professional workers. Noy and Zhang randomly assigned 453 college-educated professionals to complete incentivized writing tasks with and without access to ChatGPT. Access to ChatGPT reduced the time required to complete the tasks by approximately 40% and increased output quality by about 18%. But the study involved short, simulated writing tasks—not firm-level profitability or long-term business performance.

A related study, Navigating the Jagged Technological Frontier, examined 758 Boston Consulting Group consultants performing realistic consulting tasks with and without GPT-4. On tasks within the system’s capabilities, consultants using AI completed more work, worked faster, and produced higher-quality results. On a complex task outside the system’s capabilities, however, consultants using AI were 19% less likely to produce a correct answer.

The ability to identify where AI is reliable and helpful may be as important as the ability to use the tool itself.

More recent research provides evidence that AI can affect firm-level outcomes, not just individual worker productivity. In Generative AI and Firm Productivity: Field Experiments in Online Retail, Fang, Yuan, Zhang, Donati, and Sarvary analyze large-scale randomized field experiments involving millions of users and products at a leading cross-border online retail platform.

Over 2023 and 2024, the platform integrated generative AI into seven consumer-facing workflows, including customer service, product matching, advertising, and seller services. The effects varied by workflow, from no detectable impact to increases in sales of up to 16.3%.

The gains operated primarily through higher conversion rates rather than larger cart values, suggesting that AI improved the customer experience by reducing search, information, communication, and personalization frictions. The study also found no deterioration in product returns or customer ratings.

These findings are encouraging, but the setting matters. The retailer had the scale, data, infrastructure, and ability to integrate AI into multiple high-volume workflows. A typical search-fund acquisition may have none of those things.

A practical question emerges from my conversations with search CEOs. They tend to describe the benefits of process improvement in terms of efficiency and consistency, allowing the business to pursue new opportunities.

If AI gives a team time back, that capacity may become available for pursuing new product lines, geographic regions, or customer bases. It may help a company respond more quickly, document its processes, or make institutional knowledge easier to share.

When a searcher is considering buying a stable, profitable business that already operates in its own distinctive way, the relevant question may not simply be whether AI can reduce costs.

The better question may be: What new opportunities might greater efficiency and consistency make possible?

What Does This Mean for Searchers?

At this point, I would not say that AI will improve search-fund returns as a general rule.

The research supports a more cautious conclusion. AI may help some companies improve after acquisition, but the benefit will depend on the business, the use case, the people implementing it, and the quality of the process being improved.

Search-fund returns remain affected by purchase price, leverage, growth, holding period, exit valuation, and market conditions. AI may influence some of those variables, but it cannot substitute for buying a durable business at a reasonable price or developing a credible plan to operate it well.

CEOs need time to learn the business and make operational changes. The Yale research suggests that value creation in ETA is often driven more by multiple expansion than by EBITDA growth. The AI research suggests that productivity gains are possible, but uneven and highly dependent on the task.

Taken together, these findings suggest that AI should be treated as a potential operating tool—not as a guaranteed source of value creation.

The searcher still has to identify a good business, understand how it actually operates, preserve the institutional knowledge that may not appear in the financial statements, and determine whether the proposed improvements are realistic.

 

CONTINUING THE CONVERSATION

These are the kinds of questions we explore within the Challenge Point CEO Roundtable—a conversation among search-fund CEOs and guest mentors focused on the practical realities of building value after acquisition.

If you are interested in joining the conversation, learn more about the CEO Roundtable or book a call with me.