How to sell your company's data to AI labs
Selling your company's data for AI training usually means licensing a de-identified copy to AI labs and AI companies. AI labs pay established companies for de-identified copies of their internal records: email, CRM, accounting, contracts, support tickets, team chat and code. In most deals the company licenses a copy and keeps the original, so it still owns its data and can license it again.
Why AI labs buy company data
AI labs have already trained their models on most of the public internet. What they lack is the record of how real businesses work: how a firm prices a job, handles a dispute, closes its books or answers a customer. That record lives in private systems. Labs use it to train models and to build practice environments where AI agents work through realistic office tasks.
In October 2026 the AI data company micro1 said it would spend $1 billion over the following 12 months buying and licensing operational data from companies, with Citi and Hercules Capital providing the capital. In April 2026, Forbes reported that companies that had shut down were selling old Slack messages, email and Jira tickets to AI developers.
What kinds of records sell
Buyers want years of real work spread across several systems. Our data partner's public price model gives each type of record a base value. The base value is then scaled by company size, years of history and sector (see the price guide).
| Record type | Examples | Base value in the model |
|---|---|---|
| Accounting or ERP | Ledgers, invoices, purchase orders, inventory | $18K |
| CRM | Accounts, deals, activity history | $18K |
| Internal and client email | $14K | |
| Support tickets | Zendesk, Jira Service Management, help desk logs | $14K |
| Team chat | Slack, Microsoft Teams | $14K |
| Call recordings | Sales and support calls | $14K |
| Contracts | Signed agreements, redlines, amendments | $10K |
| Databases | Operational and product databases | $10K |
| Client files | Case files, engagement files, project folders | $9K |
| Projects or matters | Project plans, matter histories, timesheets | $9K |
| Code repositories | GitHub, GitLab | $18K |
| CAD, engineering or design files | Drawings, models, specs | $18K |
| Jobs and field operations | ServiceTitan, Procore | $12K |
| Orders and point of sale | Shopify, Square | $10K |
| HR and recruiting | Workday, Greenhouse | $7K |
micro1 lists a similar set: standard operating procedures, knowledge bases, internal documents, CRM data, project histories, quality assurance processes and records of how teams make decisions.
What company data is worth
- On our data partner's public price model, estimates run from about $20K to $1.5M. A 25-person accounting firm with 11 to 17 years of email, accounting, CRM and client files comes out at about $90K to $180K. A 150-person manufacturer with more than 18 years of email, ERP, CRM, contracts and databases comes out at about $230K to $460K.
- In October 2026 our data partner said one company had recently been offered $480K, and that most companies get offers between $100K and $1M.
- micro1 publishes its own tiers: more than $100,000 for a qualified partnership, more than $500,000 for large datasets, and more than $1 million for highly unique data.
These figures are a rough guide. Only a written offer after review is a real price.
How the process works
- Check what you own and what you may share. Look at customer contracts, NDAs, platform terms and any duties to regulators.
- Describe your records. Through finData this is our data partner's 10-minute form: which systems you use, how many years they cover and roughly how much data there is. Estimates are fine.
- Get a written offer. Our data partner reviews your answers and sends an offer that states the terms.
- Sign, then share only what you approved. Access is read-only and covers only the sources named in the agreement. Identifying details are removed before any AI lab sees the data.
- Get paid when an AI lab licenses the data.
If you do not sign, nothing is shared and nothing happens.
What you keep
You keep your company, your systems and your original records. AI labs license a de-identified copy. You choose which systems to include and can leave any of them out. Our data partner's security page says access is read-only, data is encrypted, the originals are deleted after processing, and the data is never published or sold to your competitors.
What to check before you share anything
- Email and chat often hold other companies' confidential information, covered by NDAs and commercial agreements. A guide from the law firm Frankfurt Kurnit Klein & Selz makes this point, and notes that in the Spirit Airlines bankruptcy a software vendor claimed that records generated by its platform belonged to the vendor.
- The same guide says Slack's API terms prohibit third-party application providers from using API data to train a large language model, so the way you export chat history matters.
- Health, legal, financial and insurance records come with duties to clients and regulators. Check those duties before you share anything.
- Our data partner's own security page says no de-identification method is perfect. Leave out anything you would not want exposed.
None of this is legal advice. Take advice from your own lawyer before you sign.
Questions
Do I sell my company's data or license it?
In most deals you license a de-identified copy. You keep your company and the original records, and you can license them again later.
Can a small company sell its data to AI labs?
Yes. Size is not a hard limit, although value rises with staff and years of history. On our data partner's public price model, a 10-person firm with 3 to 5 years of email and accounting records comes out at about $20K to $40K.
Is it legal to sell company data for AI training?
It depends on your contracts, the kind of data and the countries involved. Check your NDAs, customer contracts, platform terms and any duties to regulators, and take legal advice before you sign.
How do I license business data to AI companies?
Check what your contracts let you share, describe your records to a data marketplace, and review its written offer. If you sign, the marketplace gets read-only access to the sources you approve, removes identifying details and licenses the copy to AI labs.
Does my data leave my control?
With our data partner, access is read-only and limited to the sources you approve, and it starts only after you sign. Data is encrypted, identifying details are removed, and the originals are deleted after processing.
See what your company's records are worth
Answer six questions and get a range from our data partner's public price model. It takes a minute and costs nothing.
Get an estimatefinData introduces established companies to our data partner, a marketplace that licenses business data to AI labs. If you license your data through them, they pay us a referral fee. You pay nothing. Details are in the questions on our home page.
Sources
- The Next Web, "micro1 commits $1bn to buy company data for training AI agents", 9 October 2026: thenextweb.com
- Forbes, Anna Tong, "AI's new training data: your old work Slacks and emails", 16 April 2026: forbes.com
- Daniel Goldberg, Frankfurt Kurnit Klein & Selz, 1 October 2026, "Selling company data for AI training: what to know before you say yes": technologylaw.fkks.com
- Our data partner's public price model and security page (linked from the home page).