Services as software: how AI changes the math on an entire business category
AI is changing the unit economics of service businesses so completely that they stop behaving like service businesses. They start behaving like software companies.
By Matei Olaru · · 10 min read
Two years ago I almost raised a fund to buy service businesses to transform with AI. The fund didn't happen because it was too early to go all in on one lane, but the thesis is stronger now than when I wrote the offering memorandum. I've spent the last two years educating mid-market companies on AI modernization and implementing systems I design for them.
AI is changing the unit economics of service businesses so completely that they stop behaving like service businesses. They start behaving like software companies.
The problem with service businesses
Service businesses have a math problem. Revenue scales linearly with headcount. A property management company managing 500 units needs X people. To manage 1,000 units, they need roughly 2X people. Margins stay flat or shrink. Growth means hiring. The business hits a ceiling.
I call this the Linear Ceiling. If doubling your revenue means doubling your headcount, you've hit it. And it's the reason service businesses trade at 3-5x EBITDA while software companies trade at 10-20x revenue. The market is pricing the constraint.
Most companies are applying AI wrong
When I present to groups of CEOs (I recently walked 190 YPO members through this), the first thing I tell them is that most AI projects fail not because the technology doesn't work, but because it doesn't have access to the same information your people do.
AI project failure today is a context problem, not a tech problem.
The second thing I tell them: almost every company starts with the wrong kind of AI. They buy ChatGPT licenses. The KPIs don't really move.
This is what I call Grease. Grease makes a human 10% faster at their existing job. Your support rep still reads the ticket, still looks up the order, still types the response. She has an AI assistant that drafts a reply she can edit. She's slightly faster. She still touches every ticket.
A Cog replaces the workflow entirely. The ticket arrives. An AI agent reads it, checks the order status in your ERP, identifies it's a routine request, pulls the tracking number, and sends the response. Your support rep never saw it. She's handling the 20% of tickets that actually need human judgment.
That's not a 10% improvement. That's a 10x structural change.
Grease is AI-enablement. It's a feature. Cogs are AI-native. They are assets. Features depreciate. Assets compound.
How to find the cogs
You can't find cogs at the department level. "Automate procurement" is a wish, not a buildable instruction. You have to break work into its smallest pieces, what I call Atomic Units. A trigger in, an action out. "When an invoice PDF arrives in this inbox, extract line items, match against the open PO in NetSuite, flag any variance over 2%." That's a buildable instruction.
What happens when you build enough cogs
Take a property management company. They spend 50-65% of their costs on tenant communications, maintenance coordination, and owner reporting. Most of that work is handling unstructured information and making routine decisions based on patterns.
Build cogs for the routine tenant inquiries (24/7, instant response, no staffing). Build cogs for maintenance ticket routing and vendor matching. Build cogs for automated financial reconciliation and owner reports.
The result isn't that you fire half the staff (that's a reductive way to think of AI). The result is the same team manages three times the properties. Revenue goes up 3x. Costs go up maybe 20% for the AI tooling. EBITDA margins expand from 10-15% to something approaching 25-35%.
In truth, this capacity unlock is also a limited view. True AI modernization means building systems and processes for machines.
When the Linear Ceiling breaks, businesses start exhibiting properties that look like software:
Marginal cost drops. Each additional customer requires a fraction of the effort it used to. The cogs handle the routine work at near-zero marginal cost.
Data compounds. Every interaction feeds back into the system. The AI gets better at predicting problems, routing requests, generating accurate reports. A competitor starting from scratch doesn't just face a labor disadvantage. They face a data disadvantage that grows every month. Your business stops just having data and starts having memory.
New markets open up. When your cost structure required X humans per Y units managed, lots of small markets didn't justify the overhead. When cogs handle the bulk, suddenly a 50-unit building in a secondary market is profitable. You can compete on price in price-sensitive segments and still make money. You can offer premium service levels at standard prices.
Why this matters now
Adoption is very low among service businesses and the businesses that move first get a compounding advantage that grows every month.
A service business that deploys cogs today will, in two years, have a data advantage, a cost advantage, and an operational advantage that a competitor starting from scratch literally cannot catch. The gap compounds because the data compounds.