Start here: how an AI holding company works

Begin with the ordinary roll-up

A roll-up is a simple idea: buy several businesses in the same or related markets and operate them together. A buyer might acquire a group of local accounting firms, maintenance companies, or insurance brokerages. The businesses can keep their customer relationships while sharing some management, purchasing, technology, or back-office work.

A holding company is the parent organization that owns those businesses. A vertical is the particular market they serve, such as accounting, property management, or managed IT. An acquisition is the purchase of a business. Enterprise value, often shortened to EV, is one way to describe the value of an operating business: broadly, the value of its equity plus debt, less cash. It is not the same thing as money raised in a funding round.

Conventional roll-ups can gain purchasing power, combine administration, cross-sell services, or bring professional management to small firms. They can also go badly. A buyer may overpay, lose local knowledge, burden the group with debt, or discover that businesses that looked similar do not fit together.

What AI changes in the thesis

The newer thesis is that ownership can make it easier to change how work gets done. A software vendor has to persuade each customer to adopt its product, connect it to existing systems, change employee habits, and keep paying for the tool. An owner can make workflow redesign part of the operating plan across the businesses it controls.

That does not make the technology automatic or the outcome inevitable. It means the owner can coordinate the ingredients: access to workflows and data, engineering time, operating managers, employee training, and a reason to keep improving the system after the first pilot.

A hypothetical accounting group

Imagine a holding company buys several accounting firms. Each firm receives invoices and bank records, prepares work papers, asks clients for missing information, reviews exceptions, and produces reports. The firms may perform those steps differently.

The owner could first document the common process, then introduce shared software to classify documents, draft routine reconciliations, and route unusual items to experienced staff. Engineers could observe where the tools fail and improve them across the group. Accountants would still review judgment-heavy work and remain responsible for the service.

If it works, the firms might serve more clients without lowering quality, complete work faster, make fewer avoidable errors, offer a more consistent customer experience, or reduce the cost of routine tasks. But those are possible improvements, not assumed results. Integration may cost more than expected. Staff and customers may resist the new process. AI output may require so much review that the savings disappear. Demand may not grow enough to use the added capacity.

The important question is therefore not merely “Can AI perform this task?” It is “Can this owner repeatedly improve a real business after paying for the acquisition, engineering, integration, and ongoing operation?”

The main approaches

The companies in this research do not all use the same structure.

Service-business roll-ups buy firms whose work is still performed largely by people. Thrive Holdings has backed platforms in accounting and IT services. Long Lake describes a multi-sector approach to acquiring and operating service businesses. Their central challenge is turning better tools into better operations across many local teams.

Refounding starts with a substantial existing business and tries to rebuild its operation around new technology rather than assembling only small firms. Sequence Holdings uses this language in its writing about BankSouth and The Baldwin Group. The ambition is larger in scope, and so are the integration and governance questions.

Software holding companies acquire software products rather than labor-intensive service firms. Beacon Software describes permanent ownership of niche software businesses and a shared team that helps them build and operate. Serving several unrelated end markets still makes a company multi-sector, even when every acquisition is software.

Technology-led service platforms combine software with service delivery. Eudia, for example, is building in legal work. The boundary between software company, service operator, and acquirer can move over time, so the profiles date what was known rather than forcing every company into one permanent box.

Investors and company builders also appear in the research, but they are not treated as operating holding companies. A sponsor’s fund size does not tell us the acquisition scale or operating performance of a company it backs.

What remains uncertain

The public evidence is strongest on activity: company launches, financing announcements, acquisitions, and the arguments founders and investors make. It is much thinner on comparable, long-term operating results.

Four uncertainties matter most:

1. Execution: Can teams change workflows without harming service quality or losing employees and customers?
2. Economics: Do gains remain after engineering, integration, central overhead, financing, and the price paid for each business?
3. Repeatability: Does the method work across a full group of acquisitions, not only the success stories selected for publication?
4. Time: Are early improvements durable, and who ultimately receives the value?

Company profiles state who reported a result and when. Small source controls open the supporting material. Readers who want the full evidence system can use the methodology, but understanding its vocabulary is not required to understand the business model.

Continue from here

• Read a foundational perspective: visit the Reading Room for selected letters, essays, interviews, and reporting.
• Explore the companies: open the company directory to search by vertical, read a complete profile, or compare the market landscape.