For established businesses
Sell Data to AI Companies
When a business owner asks how to sell data to AI companies, the real answer is a license: you let an AI company copy and use your old work records, and it pays you for that right. You keep the originals and you keep ownership.
This page covers company records, not personal data. It explains which records count, what to hold back, how pricing works and how a deal runs from first question to payment. When you are ready, a 2-minute estimate shows what your records may be worth.
- You keep ownership
- Offers from several buyers
- No fee to your company
The basics
Company records, not anyone's personal information
The records in question are work records. They show how your staff wrote to customers, solved problems, closed sales and finished jobs. This has nothing to do with consumer lists, browsing histories or anyone's private life, and buyers remove names and personal details from the records before they use them.
A license is permission, not a sale of ownership. The buyer gets a copy under a contract that spells out what is included, what is excluded and how the copy may be used. Your company still owns every record, and in many deals it can license the same archive again later.
AI companies want these records because public writing is running thin. Ilya Sutskever, a co-founder of OpenAI, said in December 2024 that the field had reached "peak data". Real business records are one of the few large sources left, and only businesses hold them.
DillisData works as a data agency. Our job is to find the buyers, gather competing offers and negotiate the terms. The records themselves never pass through us. They move from your systems to the buyer only after you sign.
What counts
Records AI companies license, and where you keep them
Most established businesses already have several of these. Email and shared files usually carry the most weight.
| Record type | Common systems | What it teaches AI |
|---|---|---|
| Email archives | Outlook, Microsoft 365, Gmail, old PST exports | Real exchanges between staff, customers and vendors |
| Chat history | Slack, Microsoft Teams, Google Chat | Quick decisions and handoffs as they happen |
| Help desk tickets | Zendesk, Freshdesk, Jira Service Management | A problem, the steps tried and the final fix |
| Contracts and shared documents | SharePoint, Google Drive, Dropbox, Box | Careful writing by people who know the field |
| CRM history | Salesforce, HubSpot, Zoho | Sales notes from first contact to signed deal |
| Work orders and job records | ERP, field service and dispatch software | How jobs are scheduled, staffed and closed out |
| Call recordings and transcripts | RingCentral, Zoom, Dialpad | Spoken conversations with customers, start to finish |
| Code and internal guides | GitHub, GitLab, Confluence, Notion | How a team builds things and trains new hires |
What to hold back
What most companies leave out, and why
Licensing is opt in, one system, mailbox, folder or year at a time. These are the records companies most often keep out. Every exclusion goes into the contract.
Sensitive personal details
Buyers strip names and contact details before use, but no method catches everything. Exclude anything sensitive outright, such as payroll files, ID scans or medical notes about staff.
Records your industry regulates
Some fields carry legal duties over certain records. Patient charts at a medical practice and client files at a law firm are the clearest examples, and they generally stay out.
Material you do not own
Vendor manuals, purchased research, stock images and licensed software belong to someone else. You can only license what your company created or owns.
Information under a confidentiality agreement
NDAs and customer contracts may bar you from sharing certain information. Check them, or leave the related folders out.
Records under a legal hold
If a lawsuit or an investigation requires you to preserve records, talk to your lawyer before any copy leaves your control.
Secrets you depend on
Pricing models, formulas and code that give you an edge can stay out. Leaving them out still leaves plenty to license.
Pricing
What decides how much a buyer will pay
Four things move an offer the most. Headcount comes first, since more people create more records. Years in business come next, because a long history shows how work changed over time. Then come the systems you use, with email and shared files near the top. Last is location, since buyers pay the most for English-language records from US companies.
Payouts in buyers' own published pricing stretch from near $10K for one company up past $4M. Our estimate runs four of those published pricing formulas, averages them and rounds the result down. It is a careful starting point, not an offer. Real offers arrive after buyers review your company profile.
Most buyers look for about 20 employees and 3 or more years in business. Some accept smaller or younger companies. A business that has closed, or is closing, can qualify too, provided a person with the power to sign can still get to the archives.
Check your number before you call anyone
Four questions on headcount, age, systems and region. No uploads, no passwords, and the estimate appears on screen right away.
The process
From first question to payment
The owner spends about an hour in total. We handle the outreach, the follow-up and the paperwork.
Answer four questions
Size, age, record systems and region. Nothing to upload and no system access.
2 minutesBuyers see your profile
We present a summary of your company to AI companies looking for records like yours, then gather their offers. No records change hands at this stage.
1 to 2 weeksPick a deal
We walk through each offer and its terms with you on one call. You set the scope and sign with the buyer you prefer.
One callCollect payment
The buyer pulls the records directly from you, strips personal details, reviews the data and sends payment. Published timelines run from 7 to 60 days.
7 to 60 daysProof
Data licensing is already a big business
These deals were reported publicly. DillisData was not part of any of them.
Reddit's AI licensing deals added up to this much, according to the company's own IPO paperwork.
SEC Form S-1, Feb 2024Shutterstock's 2023 income from AI developers who licensed its content library.
PetaPixel via Bloomberg, Jun 2024Meta paid this for 49% of an AI training data supplier.
TechCrunch, Jun 2025Of Slack, email and Jira history that earned Cielo24 hundreds of thousands of dollars.
Forbes, Apr 2026Timing
Why the timing matters
Nobody can promise how long buyers will keep paying at this level. These are the forces at work right now.
Public text is running low
Researchers at Epoch AI projected in 2024 that the supply of public human writing could be used up somewhere between 2026 and 2032. Private business records are one of the deep sources left.
Scarce records earn more
Buyers pay a premium for kinds of records they lack. Once several companies in your field have licensed theirs, the next offer in that field tends to be smaller.
Demand will not stay this high
Labs keep collecting, and they keep finding other ways to train. As that happens they will need less of this. Nobody can name the month.
Archives earn nothing in storage
Old mailboxes and drives cost money to keep and pay nothing back. A license changes that without deleting a single file.
Our intake is capped
Each month we cap how many companies we sign up, so every deal gets real attention. Late sign-ups roll to the following month.
By industry
Guides for your industry
Each guide covers the software that industry runs, the rules that apply and what usually stays out.
Law firms
Firm operations records such as admin email, blank templates and intake workflows, with client files kept out under the duty of confidentiality. Read the law firm data licensing guide.
Healthcare practices
Scheduling, billing workflows, internal email and SOPs, while patient records protected by HIPAA generally stay out. Read the healthcare data licensing guide.
Insurance agencies
Agency management history, service email and renewal workflows, with policyholder details handled under strict exclusions. Read the insurance agency guide.
Consulting firms
Proposals, methods, internal training and project workflows, with each client's confidential work reviewed against your agreements. Read the consulting firm guide.
Manufacturers
Work orders, quality reports, maintenance logs and supplier email that show how products actually get made. Read the manufacturing data guide.
Retail and e-commerce
Support tickets, merchandising notes, vendor email and store procedures, with shopper personal details removed or excluded. Read the retail business guide.
Real estate brokerages
Transaction workflows, listing copy, agent training and office email from years of closed deals. Read the real estate brokerage guide.
Logistics companies
Dispatch notes, load records, carrier email and exception handling that show freight moving in the real world. Read the logistics company guide.
Construction companies
Bids, RFIs, change orders, daily logs and job email that record how projects get built. Read the construction company guide.
Questions
Questions owners ask
Is it legal to license company records to AI companies?
In general, a company can license records it owns. Limits come from privacy laws, industry rules and contracts with customers or staff. That is why you choose what goes in and leave out anything restricted. Have your own lawyer read the agreement before you sign.
Is this the same as selling customer data?
No. You license a copy of work records, like email, tickets and documents, and you keep ownership. The buyer scrubs names and personal details from its copy, and you can exclude any system or folder.
What does an AI company do with our records?
It uses a cleaned copy to train and test AI models so they learn how real work gets done. The contract lists the uses it allows.
Do we have to gather or clean the records ourselves?
No. The buyer copies the records and removes personal details. Someone on your side may need to grant access or run one export.
Can we license the same records to more than one buyer?
Often, yes, as long as no contract makes one buyer exclusive. Some buyers ask for exclusive or permanent rights. We point those terms out so you can decide whether the price is worth giving up later deals.
How small is too small?
The usual floor is roughly 20 people on staff and at least 3 years of operating history. Some buyers go lower, and closed businesses can qualify too.
Why use an agency instead of contacting buyers directly?
You can go direct. We already know which buyers want your industry, we collect several offers at once, and we read each contract for lock-in terms. Our fee is paid by the buyer, added to what it pays you, so it costs you nothing.
Does DillisData ever hold our records?
Never. We work on the deal, not the data. Your records go from your systems to the buyer you picked, under the contract you signed.
Get started
See what your company's records could bring in
Four questions, two minutes, one cautious number on your screen. If it looks too low to pursue, close the tab and you are out nothing.
No fee to you. No files shared. Limited openings each month.