AI Rollup Strategy: What 'Boring' Businesses Teach Us About Scale

Over the past year, funds like Andreessen Horowitz, General Catalyst, Thrive and Lightspeed stopped fighting over stakes in tech startups and started buying accounting firms, insurance agencies and property management companies. This is not bored investors looking for a distraction. It is a movement with a name, a thesis and a clear business logic, and it has everything to do with what people are now calling AI rollup strategy: the idea that the real value of AI is not in an off-the-shelf tool, but in a layer built on top of each company's specific process.

This movement has a name: the AI rollup. And despite what the term suggests, it does not require an investment fund or a holding company to apply. The reasoning behind it works for any service business owner in Brazil trying to figure out where to start with AI.

What is the 'AI Rollup' and why Silicon Valley funds are buying 'boring' businesses

The playbook works like this: set up a holding company, raise capital, and buy real-economy businesses that technology has historically ignored. Then rebuild each one from the inside out, around artificial intelligence. This is not bolting a chatbot onto an existing operation. It is redesigning the workflow with AI at the center.

These companies get called "boring" because they run on repetitive tasks: answering emails, processing forms, issuing invoices, checking paperwork. That is exactly why they are the most fertile ground for AI for boring businesses. Tedious work is predictable, well defined and measurable, and that is precisely what an automation needs in order to work well.

From SaaS to 'service as software': how AI breaks the link between growth and headcount

For years, software as a service (SaaS) was the most prized business model in tech: build the product once, sell it endlessly, with margins that can reach 90%. Service businesses never had that luxury. They run on human time: more clients means more people, and more people means more payroll. Growth and hiring have always moved together.

What these funds are betting on is exactly the reverse of that equation, a shift from "software as a service" toward service as software AI. A bricklayer can only build more houses by hiring more bricklayers. A factory presses a button and doubles output. The bet is to turn bricklayer-type operations into factory-type operations, using AI to automate the repetitive tasks that today depend entirely on people. It is one of the most direct paths to cutting operating costs without tying headcount growth to revenue growth.

The Long Lake and Nexus case: why a custom AI layer beats using ChatGPT directly

The most cited example of this movement is Long Lake, a three-year-old holding company that has already acquired more than 30 businesses in property management, construction and corporate travel. The detail that changes everything is not the volume of acquisitions, it is how they use AI internally. The Long Lake Nexus AI platform shows that the differentiator was not opening ChatGPT and using it as is: it was building a proprietary platform, called Nexus, that the company claims delivers five times the performance of generic models in the specific context of its business.

They did not replace existing AI models with something built from scratch. They took the best models available on the market, like ChatGPT and Claude, and built on top of them a layer with the workflows, rules and tools specific to each industry. It is the difference between buying a suit off the rack and having one tailored from the same fabric: both cover the body, only one actually fits.

The value is not in the off-the-shelf tool. It is in building a solution tailored to the company's real process.

This distinction is the point most business owners still have not connected: subscribing to a ready-made AI tool solves a small fraction of the problem. The real gain happens when AI is designed around your company's specific process, which usually requires work in custom artificial intelligence combined with well-mapped process automation. That is exactly what a custom AI layer on foundation models means in practice: taking a general-purpose model and shaping it around how your business actually works.

The 4 reasons legacy companies are the best asset for applying AI

Why buy an existing "boring" company instead of building an AI startup from zero? There are four structural reasons:

• The customer already exists and already pays. There is no market to educate and no acquisition cost to recover, revenue is already real and recurring.
• The process is already mapped. The company's real workflow is the input that makes AI perform well, far more than any generic model.
• The competitor also depends on people. Whoever implements AI first can charge less and profit more, building an edge that is not copied overnight.
• Switching providers is costly. Once a client depends on the operation, the tendency to stay is naturally stronger.

The valuable asset was never AI itself. It was the existing customer base and the process already in place, with AI acting as the lever that unlocks margin that used to be trapped in manual work.

A fair counterpoint is worth making: betting everything on technology without discipline has failed before. Between 2020 and 2021, large funds bought software companies at high prices, betting that subscription revenue was the safest asset in the market. As AI advanced, part of that revenue became threatened by the very technology that seemed to guarantee the business. The lesson is not that every AI bet pays off, it is that AI needs to be applied with governance and discipline, which reinforces the importance of structuring AI governance from the start, not as an afterthought.

How to apply this reasoning to your business: 3 questions to find where AI unlocks margin

You do not need a holding company or billions in capital to apply the same logic. You need to honestly answer three questions about your own operation:

• Which tasks eat up hours every single day? These are the first candidates for automation.
• Does your growth today depend directly on hiring more people?
• Is the critical knowledge of the business documented in the process, or does it live only in a few people's heads?

These answers form the map to scale a business without hiring at the same pace as revenue growth. The starting point is not picking the trendiest AI tool. It is timing the hours spent on your company's most repetitive process and using that number as a guide to where AI process automation for legacy companies will pay off first.

The winners of this cycle will not necessarily be the labs building the most advanced AI models, but the owners of traditional businesses who use this technology to change the math of their own operation. If your service business, whether it is accounting, insurance, property management or a professional office, already has clients and a process running, the next step is to understand where AI can fit in without relying on generic tools. Talk to the Agence team to map the most expensive process in your operation and assess where a custom AI layer makes sense.