Guide › Dark data

Dark data: the unused business records frontier AI labs pay for

Every company keeps years of email, chat, documents and tickets that nobody looks at again. The industry calls it dark data, and until recently it was only a storage cost and a privacy risk. Now frontier AI labs, the companies building the most advanced AI models, pay to license de-identified copies of exactly those records.

By Sean Key · Last reviewed November 8, 2026

At a glance

Gartner describes dark data as information a company collects, processes and stores in the course of business but generally doesn't use for anything else. TechTarget (2021), paraphrasing Gartner, checked Nov 8, 2026 (opens in a new tab)
In a Splunk survey of business and IT leaders, the average estimate was that 55% of an organization's data is dark. IT Brief, reporting Splunk's State of Dark Data survey, 2019, checked Nov 8, 2026 (opens in a new tab)
Much of it is the everyday record of how work gets done, which is what AI training data companies such as micro1 license. micro1's partnerships range from $100K to $3M+. micro1 creator guidelines (refer.micro1.ai/creators/guidelines), checked Nov 8, 2026
Licensing doesn't change your duties under privacy law. Data still has to be de-identified, and personal data shouldn't be kept longer than needed.

What counts as dark data, and what AI labs want

Dark data	Where it usually sits	Fit for AI training
Email archives	Google Workspace, Microsoft 365, old mail servers	Strong. Shows how decisions get made and work gets passed on
Team chat	Slack, Microsoft Teams	Strong. Real problem-solving, in sequence
Process documents	Shared drives, wikis, SOPs, templates	Strong. How the work is meant to be done
Support tickets	Zendesk, Freshdesk, Jira Service Management	Strong. A problem, the steps taken and the outcome
CRM notes and sales records	Salesforce, HubSpot	Good. The customer lifecycle, step by step
Project records	Jira, Asana, Monday	Good. Planning and delivery over time
Source code	GitHub, GitLab	Good, though usually through code-specific buyers such as SimpleClosure's marketplace
Server logs and sensor data	Monitoring and IoT systems	Usually not what this market is buying

What makes the records valuable is that they show real people doing real work, connected over time. Why AI companies pay for business data explains why that matters to frontier AI labs.

Three ways to monetize dark data

Use it yourself

Analytics, search tools or internal AI assistants built on your own records. Valuable, but it takes investment and earns nothing directly.

Sell data products

Packaged datasets or insights sold through data marketplaces. Suits companies whose data describes a market, such as prices or transactions.

License it for AI training

A one-time or ongoing license of a de-identified copy. The buyer handles the export and de-identification, and your company keeps the original. What your data is worth sets out the published figures.

Why dark data rarely shows up as an asset

Accounting rules generally don't let a company put records it generated itself on its balance sheet, so their value never appears in the accounts. That's one reason owners overlook them. It matters most when you're selling the business or closing it, because the data can be worth more than you expect.

Before you license dark data

Find it. List your systems and check retention settings. Some tools delete history automatically, and canceling a subscription can delete years of records.
Read your privacy notices. They decide what the data can be used for, and who you'd need to tell.
Decide what's out of scope. Most companies leave out health records, legally privileged material, trade secrets and client confidential files.
Insist on de-identification. Real names and contact details should be replaced before a frontier AI lab sees anything. My de-identification guide explains how and what to ask.
Don't keep data just in case. Privacy law says personal data shouldn't be kept longer than needed. If you're not going to license it, your retention policy still applies.

Common questions about dark data

What is dark data?

Information a company collects and stores in the course of business but doesn't use for anything else: old email, chat history, documents, tickets and logs. Gartner popularised the term.

What are examples of dark data?

Email archives, Slack or Teams history, shared-drive documents, support tickets, CRM notes, project records, server logs and sensor data.

How do companies monetize dark data?

Three main ways: using it internally for analytics or AI tools, selling data products through marketplaces, or licensing a de-identified copy to AI training data companies, which pay to help frontier AI labs train their models.

Can you license dark data that contains personal information?

Only with care. The data has to be de-identified, your privacy notices and the law where you operate have to allow the new use, and you may need to tell staff and customers. Take advice if you're unsure.

How much is dark data worth?

It depends on the volume, quality and uniqueness of the records. micro1's partnerships range from $100K to $3M+, and it looks for companies with 30+ employees and several years of records.

Getting a figure

Registering costs nothing and doesn't commit you. micro1 reviews what your company holds and makes a proposal.

I'm not employed by micro1. If your company signs up through my introduction, micro1 may pay me a referral fee, which costs you nothing and doesn't change your terms.

More guides

About the author

Sean Key has been a freelance consultant for 22 years, mainly delivering data and digital projects into the NHS. He runs Applied Clinical Judgement, which connects professionals with AI training work, and helps established companies weigh up micro1's Company Data Partnerships program.

He isn't a lawyer and isn't employed by micro1. If your company signs up through his introduction, micro1 may pay him a referral fee at no cost to you.

Sean Key on LinkedIn (opens in a new tab) · Book a call (opens in a new tab)