AI Rollups: The Acquisition Playbook for AI Holding Companies

Building a new AI product from zero costs six months and a team. Buying a small, revenue-generating AI tool costs a multiple of its monthly profit and takes a weekend. That arithmetic is why the "AI rollup" — an acquisition-led holding company that bolts existing AI businesses onto a shared operational core — has become the fastest-growing structure in the space. This is the playbook: what to buy, what to avoid, and what actually breaks after the deal closes.

Why Rollups Beat Builds (Most of the Time)

The honest version of the math: a competent agent infrastructure — model routing, evaluation, observability, deployment plumbing — is 60-80% identical across products. Once you have it working for one vertical, the marginal cost of running a second product on the same rails is a fraction of building it again. We covered this shared-infrastructure dynamic before, but the acquisition angle changes the calculus further. A solo founder with a working SaaS and $8k MRR has already absorbed the build risk, found the niche, and proven willingness to pay. You are not buying code. You are buying validated demand at a discount to what it would cost you to reproduce it.

The counter-case matters too: acquisition only works if you genuinely can operate what you buy. Buying a product whose moat was its founder's personal distribution and then watching revenue decay for three months is the classic rollup failure. Deal volume is a vanity metric. Integration discipline is the real game.

What to Buy: The Four-Point Screen

Not every profitable AI tool is a good acquisition. We screen on four axes:

• Demand durability over hype cycle. Does the product solve a problem that exists regardless of which model is on top this quarter? Tools whose value is entirely "wrapper around a frontier model capability" die when the capability commoditizes.
• Infrastructure compatibility. Can the product migrate onto your agent stack within 30 days without a rewrite? Every week of parallel infrastructure is margin leakage.
• Revenue concentration. One customer at 40% of revenue is a sale of a customer relationship, not a business. We want no single customer above 15% and churn under 5% monthly.
• Founder dependence. Interview the top five customers. If they say "we bought because of [founder]," you are buying a person, not a product. Price accordingly or walk.

The Economics: How These Deals Are Actually Priced

Micro-SaaS and AI tool acquisitions in the $5k–50k MRR range currently trade around 2.5–4x annualized profit, sometimes lower for sellers who want a fast, clean exit. Structure matters more than headline price. Earn-outs tied to 90-day revenue retention align incentives far better than a lump sum — the seller's continued involvement during migration is usually worth more than a 0.5x price difference.

The hidden cost most first-time buyers miss: model spend. A product built on premium API pricing at low volume becomes structurally unprofitable when you scale it unless you route intelligently. We rebuild the inference layer of every acquisition as step one, typically cutting unit model cost 40-70% via tiered routing before touching anything user-facing. Cost-per-success, not per token, is the metric that decides whether the acquisition pencil out.

The 30-Day Integration Sprint

Every acquisition runs the same sprint structure:

• Days 1–7: Freeze and observe. No changes. Instrument everything. You need a baseline before you can prove the migration didn't break anything.
• Days 8–20: Migrate the invisible 80%. Model routing, evaluation harness, alerting, deployment. Users should notice nothing.
• Days 21–30: Ship one visible improvement. Faster response, lower price, or a feature the old roadmap never reached. Signals to customers that the acquisition is an upgrade, not an extraction.

Skip the freeze week and you will spend the next quarter guessing which of your "improvements" caused the churn spike. We learned this the expensive way.

What to Walk Away From

Three walk-away signals that have saved us more money than any due diligence spreadsheet:

• Growth that only ever came from Product Hunt launches. Spiky acquisition without retention is a lottery ticket, not a business.
• Untraceable revenue. If the seller can't show Stripe records matching claimed MRR in the first hour of diligence, the deal is over.
• A seller in a hurry for a reason they won't name. Legal exposure, a dying API dependency, or a key employee already interviewing elsewhere — there is always a reason, and you will inherit it.

The Bigger Shift

The rollup model only exists because AI collapsed the cost of operating software. Five people can now run a portfolio that needed fifty in 2020. That means the binding constraint on portfolio size is no longer headcount — it's judgment. Which products to buy, which to kill, which to merge, and in what order. The holding companies that win the next cycle will not be the ones that closed the most deals. They will be the ones whose decision quality compounded fastest. That framing, more than any multiple, is what "AI rollup" should actually mean.