Articles · Guide
What company data can you sell to AI labs?
What data can I sell? Your company's own operational records, such as SOPs, knowledge bases, support tickets, CRM history, project histories and internal chat. AI companies license them because they show how real businesses run. Data you hold for clients is generally excluded.
Check your company's records
What data you can sell, by record type
AI labs can't scrape how a real company runs, so data companies pay operating businesses to license their records. The table lists the record types in our research, what each shows and what to watch for. Source code and call recordings raise their own licensing and consent questions, so ask a buyer about them before you include them.
Record type | What it shows | Watch-outs |
|---|---|---|
SOPs and process documentation | How a process runs, step by step, in your own terms | Usually the easiest to license, since it is your own work with little personal data. Check for client names and third-party content |
Knowledge bases, wikis and internal documents | The rules, templates and know-how your staff rely on | Often mix your content with vendor manuals, licensed reports and client material |
Support tickets and chats | Real customer problems and how staff resolve them | Full of names, contact details and account numbers, often in free text. Customer contracts may limit use |
CRM and sales history | How an account moves from first contact to close or renewal | Customer and contact lists are personal data, and some contracts restrict use of customer information |
Project histories and files | How work moves from request to delivery | Deliverables may belong to clients, so check each project's contract |
Internal chat and email | How decisions get made | The heaviest personal-data load. Decide first whether to include messages at all |
Accounting exports | How money moves through the business | Account numbers, tax IDs and customer names, and sector rules may apply |
Source code and engineering documentation | How software gets built and maintained: code documentation, engineering workflows and bug tracking | Open-source components and customer code raise licensing questions, so ask a buyer before you include raw source code |
Call and meeting recordings | Speech and live problem-solving | Voices and personal details, and recording-consent rules vary |
Records that are your own work and carry little personal data, such as SOPs and project histories, are the easiest place to start. Records full of people, such as tickets, chats and email, carry the most personal-data removal work, whether you or the buyer does it.
Buyers differ in which records they want, so ask any buyer for its own list before you spend time on a record type.
Check your company's records
Check whether your company meets our referral criteria. The check asks four questions about your company and records.
A quick way to sort your records
Before you contact anyone, sort each record type into one of three groups.
Group | What goes in it | What to do |
|---|---|---|
Strong candidates | Your own work with little personal data: SOPs, process documentation, internal templates and project histories your team wrote | Inventory them first and check the ownership of anything embedded in them |
Needs work or a decision | Records with people in them or with contract limits: tickets, chats, CRM history, email and recordings | Plan the removal of personal data, read the customer contracts and decide whether to include them |
Leave out | Client data, third-party content, privileged messages, credentials and employee files | Exclude them from the inventory, and tell any buyer you have done so |
The second group is where most of the work sits, because every record in it needs a decision or cleanup. Revisit the sort after you read the customer contracts, because one clause can move a whole record type from the second group to the third.
What makes a set of records worth more
No buyer we know of publishes a pricing formula. Standard valuation reasoning, and what a buyer can do with the records, point to a few factors:
- History. Years of consistent records cannot be recreated, so a long archive is hard for a buyer to find elsewhere.
- Breadth. Records that span several systems and teams show more of how the business runs than one system's export.
- Depth and uniqueness. Specialist work that few companies document in detail is harder to replace than generic records.
- Consistency. Records kept in the same format, with the same fields, year after year, need less work to combine.
- Preparation. Clean files with little personal data need less work before a buyer can use them, whoever does that work.
Our guide to data valuation explains how each factor bears on price, and how the terms of a license move it.
How this differs from selling customer lists
Many people hear "selling data" and think of data brokers, which sell lists of people: names, emails, phone numbers and buying habits. That is personal data, and it carries the most legal risk. Operational licensing is a different transaction. The buyer wants to see how your company works: the procedures, the problems your staff solve and the way work gets done.
Customer contact details appear in those records incidentally, and much of the cleanup effort goes into removing them. If a record is mostly a list of people, treat it as personal data first and as an asset second.
What you can't sell
Some records stay out of any package, however useful they look:
- Data you hold for clients. Buyers generally exclude it. If it sits under a client contract or NDA, it usually isn't yours to license.
- Personal data you have no right to share. Health, financial and children's data have their own rules, and personal data of people in the EU or UK needs a lawful basis under GDPR before it leaves your company. Our guide to whether selling company data is legal covers the rules.
- Third-party content. Licensed reports, stock material, vendor manuals and anything else you hold under someone else's license.
- Work you can't prove you own. Under US copyright law, material written by contractors belongs to you only if a signed agreement gives you the rights, either as a work made for hire or by a written transfer.
- Privileged communications. Messages with your lawyers may lose their protection if shared, so ask counsel before including them.
- Credentials and secrets. Passwords, API keys and access tokens turn up in chats, tickets and code. Remove them before any records leave.
- Employee files. HR records, payroll and similar files are personal data about your staff.
A worked example
Example: a 150-person engineering services firm lists its record systems: a wiki, a ticketing system, a CRM, a project tracker and a shared drive. The wiki and project tracker mostly hold the firm's own work, so they are strong candidates. The tickets name customers and need heavy cleanup. The CRM holds customer lists that some contracts restrict. The shared drive mixes the firm's documents with client deliverables and licensed reports. The firm starts with the wiki and project history, leaves out client deliverables and asks counsel about the CRM.
Exporting each record type
You don't need to export anything before a buyer asks, but knowing how each system exports saves time later:
- SOPs and wikis. Export pages with their folder structure and the date each was last edited.
- Tickets and chats. Export with ticket IDs, timestamps, status and the reply that resolved each ticket.
- CRM history. Export accounts and opportunities with stages and close dates. Leave contact fields aside until you have decided how to handle personal data.
- Project files. Export by project, and note the contract that covers each one.
- Chat and email. Hold off until you have decided whether to include them at all.
Whatever the system, keep the dates, the IDs and the links between records. A ticket joined to its resolution is easier for a buyer to use than two loose files.
Where specialist buyers fit
Some buyers take mixed operational records from many kinds of company. Others concentrate on one kind of data. If most of what you hold is a single kind, such as call recordings, medical records or video, ask each buyer whether it takes that kind before you prepare anything. Our comparison of AI training data buyers lists each buyer with its published size rule and terms.
Closed or closing companies
The same record types may still be licensable after a company closes. The difference is who can sign: someone with authority to license the records has to act for the company. Our companion site covers that case in selling company data after closing.
Questions to answer before you approach a buyer
These are the questions a buyer is likely to ask first, and you can answer them from your own systems:
- Which systems hold the records, and how many years does each go back?
- Roughly how much is in each: documents, tickets, accounts?
- Which records are your own work, and which belong to clients, vendors or contractors?
- Which records contain personal data, and how much?
- Which customer contracts, NDAs or tool terms restrict reuse?
- Who in the company can sign a license, and who must approve it?
Write the answers down. They are also the start of the inventory you will need later. Whatever you decide to include, be able to describe each set in one sentence: what it is, which years it covers and which system it came from.
Next steps
Once you know which records you hold and own, the next step is preparing them for licensing. For the full process, from ownership check to payment, start at the handbook on selling company data.
Check your company's records
Check whether your company meets our referral criteria. The check asks four questions about your company and records.
Frequently asked questions
What kind of company data can you sell to AI companies?
Operational records that show how your company works: SOPs, knowledge bases, support tickets, CRM history, project histories and internal chat and email. You license them for a fee and keep your own copy.
Can you sell customer data to AI companies?
Usually not as it stands. Customer personal data falls under privacy law, customer contracts often limit its use, and buyers generally exclude data you hold for clients. Records with personal data properly removed are a separate case.
How much data do you need to sell to an AI company?
There is no standard minimum across buyers, so ask each one. Before you do, note how many years each system covers, roughly how many records it holds and how consistent they are. Those are the first things a buyer will ask about.
Can a small company sell its data?
It depends on the buyer. We refer companies with 30 or more full-time employees. Other buyers, sector specialists and marketplaces set their own rules, so a smaller company should ask each one directly.
Can a company that has closed still sell its data?
Possibly. The same record types may still be licensable after a company closes, and someone with authority to sign for the company has to license the records. The eligibility check on this site covers companies that are closing or closed.
Sources
