How to sell your company data for AI training

Why AI labs want real operational data

Models are increasingly asked to act as assistants and agents that plan, hand off tasks and follow processes. Public text rarely shows that. Internal records do: a support ticket shows a problem being diagnosed step by step, a Slack thread shows a decision being argued out, and a project document shows how a plan changes over time. That kind of material reflects how work is really done, which is hard to recreate synthetically.

Who buys company data

AI labs directly

The largest labs sometimes negotiate directly, but generally with very large data holders. Most mid-sized companies will not be approached this way.

Data partnership firms

Firms such as micro1 work with labs on one side and companies on the other. They assess the data, scope a package, handle de-identification and manage the transfer. For most businesses this is the realistic route.

Data marketplaces

Marketplaces list datasets for sale or licence. They suit packaged, already clean datasets more than raw internal records, and you usually do more of the preparation yourself.

What typically qualifies

• 30+ employees (micro1's published requirement), including closed companies
• Based in the US or other Western markets
• Records mainly in English
• Several years of history across Slack or Teams, email, documents, tickets and CRM notes

Volume and history matter most. Our guide to what company data is worth covers the factors in detail. Companies that have shut down or are winding down can also qualify through a separate defunct-company track; our partners page explains how advisors and insolvency practitioners can refer them.

What is usually excluded

• Personal data you cannot reliably de-identify
• Client data covered by an NDA or confidentiality clause
• Regulated health data without a proper legal basis
• Anything you do not clearly own or have the right to license

How the process works

1. Assessment. A short review of your company size, systems and history.
2. Scoping. Agreeing which sources, date ranges and teams are in scope.
3. Written scope you approve. Nothing moves until you sign off on exactly what is included.
4. De-identification. Names, emails and personal identifiers are removed or masked, and the data is reviewed.
5. Transfer. The agreed package is transferred securely.
6. Payment. Paid under the agreed terms once the data is approved.

The commercial side of step three is covered in our data licensing primer.

Risks and how to manage them

Privacy law

GDPR, the UK GDPR and US state laws such as CCPA apply to personal data in your records. De-identification reduces the risk but does not remove your obligations, so ask how it is done and get advice.

Contracts

Check client agreements, supplier contracts and investor terms for restrictions on how data can be used or shared.

Platform terms

Some SaaS providers restrict how exported data, or data pulled through their API, can be used. Check the terms of each tool before including it.

Employee notice

Your staff wrote much of this material. Depending on where you operate, you may need to tell them, and it is good practice either way.

Reputation

Consider how customers and staff would react if they learned about the arrangement, and be ready to explain the safeguards.

What it pays

There is no public price list. For reference, micro1's published range ($100k to $2M+ per data package), subject to micro1's review and approval; not guaranteed. The amount depends on volume, history, domain and data quality.

Next step

Our free valuation asks about your company size, history and systems and gives an indicative range in about a minute. It is an estimate only; any real offer comes from micro1 after review. If you are also weighing other ways to earn from your records, see how to monetise internal documents.

Questions to ask before signing

• What exactly is in scope? Ask for a written list of systems, date ranges and record types, and confirm anything not listed stays out.
• How is de-identification done and checked? Ask who removes names, emails and other identifiers, and whether you can review samples before transfer.
• Is it a licence or a sale? Confirm whether you keep ownership and whether the rights granted are exclusive or non-exclusive.
• When and how is payment made? Understand whether payment follows approval of the data, delivery, or milestones.
• What happens at the end? Ask how long the buyer may keep and use the data, and what deletion commitments apply.

Common mistakes to avoid

The most common mistake is cancelling tools before exporting history, which can permanently remove the most valuable records. Another is assuming every record can be included: client work covered by confidentiality terms usually has to be carved out, and finding that late slows everything down. Some companies also skip internal sign-off and only involve their board, co-founders or investors at the end, which can stall a deal that was otherwise ready. Finally, treat any early estimate, including ours, as a guide rather than a promise. The figure that matters is the one set after the buyer reviews the actual data.

A short internal checklist helps: list systems and how far back each goes, confirm admin access and export options, flag NDA-covered material, read each provider's terms on exported data, and agree who has authority to sign. With that in hand, the conversation with a buyer is faster and more predictable.