$5T opportunity: AI Roll Ups

The AI roll-up thesis targets a $5 trillion opportunity over the next decade as small-business owners retire without family members or employees ready to take over. Greg Eisenberg cites McKinsey’s estimate that about 1 million businesses, including accounting practices, insurance agencies, and property-management firms, will sell by 2035. The strategy is to acquire established firms at their current earnings and use AI agents to take on routine service work, with people reviewing the output. If agents can reliably shift work and costs, businesses with current profit margins of roughly 5–10% could potentially reach 30–40% margins, or roughly three to four times the profit. That is a possible outcome, not a proven result.

Eisenberg says Thrive Holdings has acquired close to 50 accounting firms in the past 24 months and committed another $1 billion to acquisitions. At its Larson Gross firm, with five offices and 200 employees, AI built on OpenAI Codex helped process 7,000 tax returns. Accountants saved an average of 31% of their time, and one accountant reduced a task from 180 hours a year to 15. Interns who had prepared returns began training to review them instead. General Catalyst, meanwhile, set aside $1.5 billion for the strategy and invested more than $750 million across at least 10 companies.

Some early portfolio results are striking but should be treated as signals rather than proof. General Catalyst-backed Long Lake says it acquired 18 property-management businesses, reached $100 million in EBITDA in under two years, and doubled margins; Crescendo says AI handles about 90% of its front-line customer-support tickets. General Catalyst companies typically pay about 60–70% cash for acquisitions, with the seller rolling roughly 30% equity into the new company. The broader playbook is to build an AI platform, acquire trusted firms, run agents alongside employees, and shift back-office work to agents as reliability improves. Acquiring more firms on the same platform can spread the operating system while aiming to preserve client revenue and lower costs. But reported figures are often self-reported by young companies seeking investment, and the businesses have not been tested through a recession.

Smaller operators can pursue a related model without building a large fund: a holdco that owns several businesses, each run by a general manager and supported by shared agents, rules, back-office systems, and dashboards. The idea, which Eisenberg calls “multi-pronorship,” is to build the operating layer once so future integrations become easier. Direct founder involvement can help preserve client knowledge and ensure people review agent work. The system should capture global and business-specific rules, tested examples, repeatable runbooks, and a log of corrections. Agents can handle intake, prepare work, and check drafts against those rules, with plain-English instructions specifying their duties, limits, and when to seek human help.

The recommended safeguards are to let a reviewer block work but not send it, and to require a person to approve client work, especially in regulated or high-stakes settings. Operators should track five measures weekly: profit margin, human time per job, how often agent drafts need fixing, client retention, and whether key employees are happy and staying. Repeated human corrections can be turned into agent rules and tests; after a few hundred jobs, the accumulated operating knowledge may be harder for competitors to copy than the AI models. Higher profit alongside departing clients is a warning sign that efficiency gains may be coming at the expense of service.

To find a first acquisition, Eisenberg suggests serving one industry with agents on real work, building relationships with owners, and pursuing a takeover when an owner is ready to step back, potentially after six, nine, or twelve months. He favors direct outreach over business marketplaces unless a marketplace is highly curated, and advises buying slowly, leading integration directly, and valuing firms on current earnings rather than anticipated AI gains. Roll-ups can still fail through overpaying, acquiring faster than integration can keep up, or cultural breakdown. His proposed transition is to keep agents in the background for the first 30 days without changing what clients see, give senior leaders meaningful roles and real upside, and retain human approval and accountability for now. Jobs may change and layoffs may occur; AI does not remove the risks of acquisition and integration.