We Didn't Buy the Company. We Bought the Right to Rebuild It.
TL;DR: The next decade's biggest AI opportunity isn't founding another startup — it's taking traditional companies that already have customers, trust, and 20 years of accumulated process knowledge, and rewriting their cost structure with agents. At Mercury, we just did exactly that with a labour agency. No acquisition, no leveraged buyout. A revenue-sharing partnership that gives us skin in the game and them a rebuilt funnel, GEO/SEO infrastructure, and a CRM-run back office. Same math as an AI roll-up. None of the debt.
I am James, CEO of Mercury Technology Solutions. From my office in Hong Kong, watching the AI roll-up thesis go mainstream in private equity circles, I keep noticing everyone assumes you need to buy the company to fix it.
You don't. You need the right to re-engineer it — and a reason to do it well.
The Roll-Up Math Everyone's Excited About
The logic is simple. McKinsey estimates around $5 trillion of US business assets could change hands by 2035, driven largely by Baby Boomer owners retiring with no successors. Private equity's classic play: buy boring companies, consolidate back offices, cut costs, add leverage, sell higher.
AI roll-ups add a new lever: redesign the work itself — with AI doing the machine work and humans doing the judgment work.
A services company running a 10% EBITDA margin doesn't need more revenue to triple EBITDA. It needs AI agents eating the repetitive work — data entry, document processing, routine client communication, ticket triage — until 10% becomes 30%. Same revenue, same customers, three times the profit.
Larson Gross, an accounting firm founded in 1949, ran 7,000 tax returns through AI this filing season. Accountants saved 31% of their time on average. One workflow that took 180 hours a year now takes 15 — the math is obvious once you see it. What's less obvious is where the actual moat sits.
The Moat Isn't the Model
Claude, GPT, Codex — everyone rents the same models. The model is a commodity.
The moat is what's inside the company you take over: the client relationships, the historical case files, the edge cases nobody documented, the SOPs, the senior staff's judgment, and every correction logged after every mistake over 20 or 30 years.
Compiled properly, that becomes:
Rules → Agents → Human Review → Correction Log → New Rules
You're not "adopting AI." You're compiling a traditional company's tacit knowledge into an operating system. That loop — not the model — is the asset, and it's what makes the company scalable without linear headcount.
What We're Doing With a Labour Agency
Here's where I stop theorizing.
Mercury is now working closely with a labour agency under a revenue-sharing model. The structure matters more than the industry:
- We don't collect a consulting fee and walk away. We get paid as a function of the business we actually help create. Incentives aligned, no agency problem.
- We're rebuilding the digital acquisition layer — GEO (generative engine optimization) and SEO engineered for how search actually works now, not how it worked in 2019. If the AI answer layer — ChatGPT, Perplexity, Gemini — can't see you or cite you, you don't exist to the next generation of buyers.
- We're re-engineering the customer acquisition funnel end to end — because a funnel built for walk-ins and referrals has a hard ceiling, and this business is ready to grow past it.
- We're rebuilding the back office around a real CRM — leads, placements, compliance trails, client communication, all in one system instead of someone's spreadsheet and long-term memory.
Think of it as a roll-up without the buyout. PE buys the company to get permission to re-engineer it. We got permission through a contract, and our upside is tied to their upside. For a labour agency, SEO and GEO aren't marketing line items — they're the difference between a funnel that grows and a funnel that ages. 並非是我們擁有資產，而是我們擁有槓桿 — it's not owning the asset that matters, it's owning the leverage.
Why Boring Industries Win
Accounting. Property management. Insurance. MSPs. Medical billing. Payroll. Staffing.
These industries look boring precisely because they're process-heavy, compliance-bound, and relationship-driven — which means they're full of exactly the tacit knowledge that compiles into agent systems, and full of exactly the repetitive work agents eat first.
The labour agency we're working with fits the pattern perfectly: decades of accumulated placement knowledge, a customer acquisition funnel that grew organically and was never engineered for SEO or GEO, and a back office held together by experienced humans doing machine work. That's not a weakness. That's raw material.
And the scaling question — can this company handle three times the business? — stops being about hiring more people and becomes about whether the system we're building can absorb volume. It can. That's what we built it for.
The Takeaway
Stop asking "how do we add AI to our company?" Start asking "what would this company look like if the work itself were redesigned?"
If you own a traditional business: your twenty years of SOPs, edge cases, and client trust are an asset AI-first startups would kill for. Don't let them expire with your retirement.
If you build AI systems: the best PMF in the market isn't waiting to be found. It's already sitting inside boring companies with loyal customers and broken funnels. You don't need to buy them. You need a structure — revenue share, partnership, whatever aligns the incentives — that gets you inside the machine.
The companies that figure this out won't be the ones with the best AI models. They'll be the ones who compiled their own judgment into systems first — and rebuilt their SEO, GEO, and operations around it.
Mercury Technology Solutions: Accelerate Digitality.
