If you've been on X this month, you've probably seen people talking about the "data broker" business model.
The premise is simple but I will warn you there's a LOT more that goes into this business than people are making it seem.
The business model in a nutshell:
Frontier labs (Anthropic, OpenAI, SpaceX, etc.) have used all the data on the internet to train their models and need private company data to keep improving.
They need to buy this data from somewhere, so there's been a recent explosion of companies who find businesses with the right data then de-identify it, package it, and sell it to the labs.
As someone who is always looking at different ways to make money with AI, this made me super curious so I started reaching out to people in this world.
This led me to a guy named Ryan Lock, founder of Polyshares, who does this for a living. His firm tracked a million dollars in revenue in its first month brokering private company data to the labs, and the average deal he does is about $300,000.
When I interviewed Ryan, I asked him a ton of questions about how this industry works, how deals get done, and how everyday people can get a piece of the action even if they don't have relationships with the big labs.
Here's a summary of everything he told me:
The labs are interested in a very particular set of data. They want to watch work happen. A client email comes in, the team argues about it in Slack, someone proposes three fixes, one gets picked, the answer goes back to the client. That whole chain, across systems, is the asset they're paying for.
This means that depth of history beats raw volume.
Ryan learned this the hard way from his first company. He ran an 8-figure service business through cell phones. His team would call the homeowner, call the contractor, call the rep. From a data perspective, his operation looked like "opportunity created, opportunity closed." Because of this, his company's data wasn't worth much of anything because they couldn't see the chain of data that led to a sale.
If you want to get into this game, here's Ryan's playbook:
Step 1: Qualify before you pitch.
He looks for this criteria:
- Companies with 50 or more employees
- At least 3 years on the same core software
- White collar
- Ideally remote-first (the work lives in email, Slack, and docs)
- US-based
- English
Under 50 employees gets hard because legal and due diligence cost the same as on a big deal. If they switched every system three years ago, only those three years count toward the valuation.
Step 2: Understand the corpus.
Ask what systems they run. Email, Slack or Teams, CRM, ticketing, project management, shared drives. Full context across systems is what gets scored. A partial export of one tool isn't worth nearly as much.
Step 3: Tell them not to touch the data.
A vetted third party handles de-identification, on-prem for high finance and legal. Every "Corey" becomes "employee 632" so the relationships in the data survive. An owner who strips names himself can destroy the value of the whole set.
Step 4: Protect the term sheet.
He never signs a perpetual license. A 24-month exclusivity is the normal ask. If the owner plans to sell the company, he times the exclusivity to expire before the exit so the next buyer inherits a sellable asset. The check doesn't get factored into EBITDA. It's one-time money.
Step 5: Decide your role.
- Option one: refer the deal to a broker like Ryan and collect a finder's fee (Ryan pays 6%). On a $300,000 average deal, that's $18,000 for an introduction.
- Option two: act like a business broker for data. Help the owner inventory and present the corpus the way a good M&A advisor presents financials, and charge an advisory fee on top.
In my opinion the easiest way to get started is by finding one company that fits Step 1 and connecting it to a buyer.
- Corey
PS: If this business model interests you at all I highly recommend watching the full interview I did with Ryan. This space is HOT, but I don't know how long it'll last.
A few notes/tweets/cool things sourced from the AI community.
1)
|
5:18 PM ā¢ Oct 8, 2026
3300 |
Retweets |
37635 |
Likes |
|
|
ā
Reverse engineer any software.
2)
Give your Grok Bot access to Claude/GPT models.
3)
Very cool idea from my friend the Boring Marketer.
Thanks for reading the Corey's Notes newsletter. I'd appreciate it if you sent it to a friend that might find it interesting.
Be sure to check out t he Build With AI podcast on Apple, Spotify, and YouTube. If you're a non-technical entrepreneur who wants to learn how to integrate AI into your business, you'll love it.
Be back next week.
-Corey
ā