Short answer
Company email archives can be licensed for AI training, and buyers value them because email records years of client and vendor conversations. Email also holds the highest concentration of personal, privileged and third-party confidential content, so deals depend on choosing the right mailboxes, excluding high-risk material, de-identifying the rest and reviewing rights before anything moves.
Email is the oldest and deepest record most companies have. A fifteen-year-old business may have fifteen years of client threads, vendor negotiations and internal decisions sitting in its mail system. That depth is why AI data buyers ask for email. It is also why email needs more care than any other source.
Key takeaways
- Email is valuable for its history and for the client and vendor conversations it holds.
- It is also the riskiest source: personal messages, legal advice and other companies' confidential information all live in inboxes.
- Not every mailbox belongs in a deal. Shared, role-based inboxes are the easiest fit.
- Exclusion comes before de-identification. Some categories should never be included.
- No email should move until a written license is signed.
Why AI data buyers want email
Email shows how a company deals with the outside world. A single thread can hold a customer's request, a clarifying question, a quote, an objection and an agreement. Across thousands of threads, an AI system can learn how commercial conversations actually unfold.
Email also links other systems together. A support ticket, a CRM update and a project task often start with one message. Reporting on this market describes the most valued package as a connected chain running from a client's email through the internal discussion to the reply. Email is the first and last link in that chain.
The scale can be large. Frankfurt Kurnit describes a $10 million winning bid for Spirit Airlines' internal business data, while noting that the proposed sale drew union and vendor objections.
What makes email harder than other sources
It is personal. People use work email for life. Review emails and calendars for doctor's appointments, school pickups and personal plans before deciding what can be included.
It is privileged. Conversations with lawyers sit in the same inbox as everything else. Disclosing them can put attorney-client privilege at risk.
It belongs partly to others. Messages from customers, vendors and partners often contain their confidential information, protected by NDAs and commercial agreements.
It is unstructured. Names, addresses and account numbers appear anywhere in free text and in attachments, which makes them harder to find and remove than fields in a database.
Which mailboxes fit a deal
| Mailbox type | Typical content | Fit for licensing |
|---|---|---|
| Shared role inboxes (support@, orders@, info@) | Customer requests and replies | Usually the best fit, after de-identification |
| Sales and account mailboxes | Client and prospect conversations | Possible, subject to customer contracts |
| Operations and project mailboxes | Vendor coordination, scheduling | Possible |
| Executive mailboxes | Strategy, board matters, legal advice | Often excluded |
| HR and payroll mailboxes | Personnel, pay, benefits, health | Exclude |
| Legal mailboxes | Privileged advice | Exclude |
Starting with shared, role-based inboxes keeps the first deal simple. They contain business conversations by design and far less personal material.
What to exclude outright
A law firm that advises on these deals recommends excluding high-risk material entirely rather than relying on de-identification to fix it. For email, that list includes:
- privileged legal communications,
- HR and personnel matters,
- payroll and tax records,
- health and benefits information,
- union and other protected workplace communications,
- customer or partner information that a contract says must stay confidential,
- content licensed from third parties,
- data from regulated systems.
See what data can't be licensed for AI training.
How de-identification works for email
After exclusions, the remaining messages are de-identified. That usually means removing or replacing names, email addresses, phone numbers, street addresses, account numbers and signature blocks, in the body and in attachments.
Three limits are worth knowing:
- Free text is hard. Automated tools miss things in long, informal messages.
- Context can identify people. A message about "our only customer in Tulsa" may point to one company even with the name removed.
- Removing a name does not remove sensitivity. A confidential pricing discussion is still confidential.
Agree in writing on who performs the work, which method is used and how the result is checked. See de-identification for AI training data.
Rights to check before licensing email
- Customer and vendor agreements. Look for confidentiality clauses and limits on sharing communications.
- Privacy policy and employee notices. Do they cover this use of personal information? Regulators have warned that companies cannot quietly change the rules for data they already collected.
- Privacy laws. State laws such as California's can treat a transfer of personal information for payment as a "sale" unless the data meets a strict de-identification standard.
- Recording and regulated content. Voicemail attachments and regulated records bring their own rules.
Our guide to whether it is legal to sell company data to AI companies covers each of these.
A practical sequence
- List your mail systems, the years they cover and the number of mailboxes.
- Sort mailboxes into the table above.
- Remove the excluded categories.
- Check contracts and notices for what remains.
- Get written offers that state scope, de-identification method and permitted use.
- Sign a license before any message moves.
DataRelai never takes or handles your raw data, including email. You describe your systems. There are no uploads and no access. We qualify what you describe, find buyers and work with the buyer through closing. See if your company qualifies.
Frequently asked questions
Can a company legally license its employees' work email?
Often it can license business email it controls, but that depends on its contracts, its notices to employees and the privacy laws that apply. Have your attorney review before agreeing to anything.
Do we have to include every mailbox?
No. Scope is negotiable. Many deals cover only selected shared or team mailboxes.
Are attachments included?
They can be. Attachments often hold the most useful documents and the most sensitive ones, so they need the same exclusion and de-identification steps.
Is email worth more than chat?
Not necessarily. Buyers weigh variety and substance. One report quotes a buyer saying a rich chat history can outweigh another company's chat, email and calendars together.
Sources
- Frankfurt Kurnit Klein & Selz, "Selling Company Data for AI Training: What to Know Before You Say Yes" (October 1, 2026)
- Crowell & Moring, "Again, AI Does Not Change The Law: FTC Guidance on Unfair and Deceptive Practices Involving Privacy Policies" (February 2024)
This guide is general information, not legal, tax or financial advice. Whether a particular dataset can be licensed depends on what it contains and on obligations your company already has.