How Much Does AI Cold Outreach Actually Cost? (Real Numbers)

I want to walk you through the economics of an agent I built for a client in commercial real estate. Because the numbers are striking enough that they’re worth spelling out.

My client’s firm buys medical office buildings. Their whole business model depends on finding building owners who might be open to selling — doctors who own their practice space, property groups that hold aging medical campuses, dermatologists who’ve been in the same building for twenty years and might be ready to cash out.

The challenge: there’s no public list of these people. You have to find each one, figure out who actually owns the building, locate them on LinkedIn, and reach out personally. Done by hand, that’s a few people a week. Not enough to build a real pipeline.

So we automated it. And the math on what we built is what I want to share.

The Outreach Bottleneck

The core constraint was this: LinkedIn is the right channel for these owners, but LinkedIn doesn’t have an API. They’ve intentionally blocked programmatic access to prevent exactly the kind of outreach I’m describing.

Traditional automation tools can’t touch LinkedIn. Any script that tries to interact with the platform gets flagged and the account gets restricted.

This is a problem that had no good solution a year ago.

What Virtual Machines Changed

Over the past year, a new category of AI capability has become practical: virtual machines controlled by AI.

The concept is simple in principle. Instead of a program talking directly to a platform through an API, you spin up a small computer — a virtual machine — and you prompt an AI to control that computer the way a human would. It moves the cursor. It types. It clicks buttons. It reads what’s on the screen and responds to it.

From LinkedIn’s perspective, it looks like a person sitting at a keyboard. Because functionally, it is — just an AI-controlled one.

For my real estate client, this unlocked everything. We built an agent that:

1. Takes a spreadsheet of property addresses
2. For each address, researches who owns the building (pulling from public records, company websites, Google)
3. Finds the owner on LinkedIn
4. Drafts a personalized outreach message that references their specific property, their background, and the general proposition
5. Logs into the client’s LinkedIn account via virtual machine and sends the message

The whole process runs in the background. Hundreds of contacts per day if needed. No human involvement until someone replies.

The Numbers

Here’s where it gets interesting.

Each message costs approximately 70 cents to send. That includes the AI compute, the virtual machine runtime, and the research layer. Not 70 dollars — 70 cents.

Now consider what happens on the other end.

If a building owner says yes and the deal closes, the commission on a $10 million medical office building at 2% is $200,000.

So the input cost per contact is $0.70. The potential output per deal is $200,000.

You could run 285,000 personalized outreach messages for the cost of a single commission. Realistically, you’d hit your target volume in days, not months.

Before this agent, my client’s team was manually reaching out to a handful of people per week. That’s the human limit. With the agent, the constraint becomes how many qualified targets exist in the market — not how many hours the team has.

Why This Works for Outreach Specifically

There’s a principle I’ve found useful when deciding where to apply AI: start with tasks that happen at high frequency and have a clear, measurable outcome attached.

Outreach checks both boxes. It’s something you’d do every day if you could. And every response you get is either a qualified lead or a no — the outcome is measurable.

When the potential deal value is large (like $200K) and the cost per contact is tiny (like $0.70), the ROI calculation becomes almost silly. You’re not debating whether it’s worth it. You’re just asking how fast you can run it.

This is different from AI projects that promise vague “efficiency gains.” The math here is direct: contact more qualified people, close more deals, at a cost that’s essentially rounding error relative to the revenue.

The Same Pattern in Other Industries

I’ve seen the same virtual machine approach applied in other contexts.

A property investor I work with needed to pull listings from MLS — the real estate database that also blocks API access. We built a similar agent: logs in, navigates to the search results, extracts the properties that match their criteria, dumps them into a spreadsheet, and triggers the next stage of the workflow. It runs continuously. Their lead flow went from manual and intermittent to consistent and automated.

The pattern is the same: there’s a platform you need to access, that platform blocks programmatic access, virtual machines act like humans and bypass the restriction.

As more platforms have locked down their APIs over the past few years, this approach has quietly become one of the most practical tools in the automation stack.

What This Means Right Now

A year ago, this wasn’t available. Six months ago, it was technically possible but not reliably practical. Now it works.

Most businesses don’t know this option exists yet. That’s a window.

The firms that figure this out in the next twelve months will build outreach and sourcing capabilities that their competitors will spend years trying to catch up to. Not because the technology is hard to access — it isn’t — but because most people aren’t looking for it.

If you have a sales or sourcing process that’s bottlenecked by “we can’t reach enough people fast enough,” this is probably the right place to look.

The math tends to make the decision obvious.