Key takeaways
- Selling company data today means licensing an anonymized copy of internal records to AI labs and data vendors; the company keeps ownership and its daily work doesn't change.
- Companies are paid $100K to $2M or more per dataset, and the largest archive so far, Spirit Airlines', drew bids of $10M and $12.5M.
- micro1 looks for 25+ employees, 2+ years of records in software and data mainly in English; closed companies can qualify if they kept their records.
- The process runs from a two-minute application to payment in about two to three months, and nothing confidential leaves the company before a lawyer-reviewed license is signed.
01
How to sell data as a company: what selling data means today
The short answer
For a company, selling data today usually means licensing an anonymized copy of its internal records, such as team chat, email, support tickets and CRM history, to AI labs and data vendors. The company is paid $100K to $2M or more per dataset, keeps ownership of the originals, and its daily work doesn't change.
Every company already produces data worth selling
Every business leaves written records behind as it works. A sales rep logs a call in the CRM (customer relationship management software, such as HubSpot or Salesforce), a support agent answers a ticket, and two managers settle a problem in a Slack thread. Over a few years, those records add up to a detailed account of how the company does its job.
That account has become valuable to the companies building artificial intelligence. Selling data, in this sense, means letting a buyer use a copy of those records in exchange for payment.
Who this guide is for
Most people who search for "sell my data" want to be paid for their own personal information, such as their browsing history or app activity. That is a different market, served by consumer data apps, and this guide doesn't cover it.
This guide is for business owners and managers who want to sell or license the records their company has built up. Everything below is about company data, sold by the company itself.
How a company data sale works
A company data sale has three parties. The company that created the records, a data vendor (a company that buys records from businesses, cleans and anonymizes them, and packages them into training material), and an AI lab (a company that builds and trains AI models, such as OpenAI, Anthropic or Google DeepMind).
The records travel one way and the money travels back the other, as the diagram below shows. Most owners only ever deal with the data vendor.
Selling to AI buyers vs. other ways to sell data
Licensing internal records to AI labs and data vendors is the newest way to sell data, and for most companies the one that pays the most. Older ways include listing a dataset on a data marketplace, selling customer lists to data brokers, and selling a whole database when a business is sold or closes.
Those older routes buy different records and ask more work of the seller. They are covered briefly near the end of this page and compared in full in where to sell data.
Video series
Selling your company's data to AI, in seven short videos
Watch the series in order, or pick any video from the list. When a video finishes, the next one starts on its own, and the series starts over after the last video.
Video 1 of 7 · 0:47
Introduction: AI companies are buying business data
AI companies are now paying ordinary businesses for their internal records, and some offers have passed $1 million. This video explains what the rest of the series covers.
Get an offer for your company's data →02
Why AI labs and data vendors pay for company data
The short answer
AI labs are training AI agents, software that carries out a job step by step, and a model can only learn a job from records of that job being done. Those records exist only inside companies, so AI labs and data vendors pay companies for them.
AI models learn from examples
An AI model learns by studying a very large number of examples and finding the patterns in them. Those examples are called training data. For years, most training data came from the public internet: websites, books, news and open-source code.
The public internet shows finished results, such as a product page or a published review. It rarely shows the work that produced them, such as the support ticket, the dispatcher's note and the follow-up email behind a five-star review.
What company records add
AI labs now want models that do whole jobs, not just answer questions. An AI agent handling a customer repair has to read the complaint, check the account, book the visit and send the invoice, and it learns that sequence from records that show it done well and done badly.
Data vendors use company records in two ways. Some become worked examples that a model studies like an answer key. Others fill a training environment, a realistic copy of a company's software where an AI agent practices a task thousands of times and is scored on each attempt. It works much like a flight simulator for trainee pilots.
The full mechanism, with examples, is in AI training data: why AI companies want your data.
03
Who buys company data?
The short answer
Company data is bought by AI labs and data vendors. Most companies sign with a data vendor, such as micro1, Mercor, Handshake AI, Polyshares, SourceX or HUMXN, which anonymizes the records and supplies them to the labs. Labs sometimes buy very large archives directly.
The data vendor is the company you sign with
AI labs rarely buy from small and mid-sized companies themselves. They buy finished training material from data vendors, which find the companies, sign the licenses, anonymize the records and build them into training material.
micro1, the data vendor we typically refer companies to, is one of the largest buyers. micro1 announced on October 9, 2026 that it would commit $1 billion over the next 12 months to acquiring and licensing enterprise operational data.
The data vendors that buy from companies
Each data vendor runs its own program, with its own pay structure and its own bar for who qualifies. The table sums up what each one publishes on its own page, checked in November 2026.
| Data vendor | How its program works |
|---|---|
| micro1 | Licenses operating records from established companies, mostly in English, with US companies first. Does the anonymization itself and pays in full within 30 days of accepting the data. |
| Mercor | Connects to a company's tools, extracts and de-identifies the records automatically, and pays by wire, typically 2–4 weeks after signing. Pay rises with the number of tools shared. You can apply to Mercor here. |
| Handshake AI | Advertises $100K to $4M+ for a company's operating records. |
| Polyshares | Reviews the record before pricing it, makes an offer within one to two weeks and pays 30 to 60 days after the data is approved and delivered. You can apply to Polyshares here. |
| SourceX | Licenses records directly from companies for a one-time fee, removes identifying details and sublicenses the data to AI developers. It takes operating, acquired and wound-down companies, works best with 50+ full-time employees at peak, and its licenses are exclusive for AI training. It pays within 60 days of a valid invoice after closing. |
| HUMXN | Buys or licenses operating records such as tickets, CRM history, code, finance records, documents, email and chat. HUMXN says it takes "any business, in any industry, with a few years of records" and lets the company choose an exclusive or non-exclusive license. |
When a lab buys directly
An AI lab sometimes bids for an archive itself when the archive is very large. In August 2026 Google agreed to pay $10 million for the internal emails, Microsoft Teams messages, spreadsheets and calendars of Spirit Airlines, the bankrupt US airline, beating a $7.5 million bid from Mercor, Reuters reported. Deals like that are rare; an ordinary operating company deals with a data vendor.
04
How much can a company make selling its data?
The short answer
A company selling its data to AI labs and data vendors is paid $100K to $2M or more per dataset, usually as a single payment. The largest archives draw eight-figure bids: Spirit Airlines' records drew $10M from Google and a $12.5M bid from micro1 in 2026.
What companies are paid per dataset
A dataset, in a company data deal, is the set of records a company licenses in one contract, usually several years of chat, email, documents, tickets and customer records. The price is set for the whole dataset, not per file or per gigabyte.
What companies are paid per dataset
The highest reported bid for one company's archive, Spirit Airlines'
From acceptance of the data to payment, at micro1
Data vendors publish ranges that match. Polyshares, a data vendor, says on its own site that typical deals run $100K to $2M and above. Handshake AI advertises $100K to $4M+, and HUMXN says on its company page that typical single-company deals "start around $250,000 and run into the millions for large archives."
How micro1 breaks the range down
micro1 publishes how its own payments split across that range on its data partnerships page. It says compensation depends on the quality, uniqueness and value of the data.
- $100K–$500K
- micro1's tier for an established company licensing its everyday operating records.
- $500K–$2M
- Large operational datasets, or ongoing participation across several teams and business functions.
- $2M+
- Highly unique, proprietary operating data with significant value for frontier AI.
The top of the market: Spirit Airlines
The largest archive sale so far shows how high the bidding can go. Spirit Airlines' archive held about 100 million emails and 500 million Microsoft Teams messages, and after Google's $10 million bid, micro1 offered $12.5 million, Fortune reported. Customer profiles were excluded, and passenger details were to be de-identified before any transfer.
Competing bids moved the price by $5 million in a few weeks. Company data has no list price, so a buyer's offer depends on how complete and how rare the records are, and on whether another buyer wants them too.
Ongoing pay for new work
Some companies also earn ongoing pay by doing new work for a data vendor. The company's staff do real tasks from their own jobs using the newest AI tools, or grade how well AI models handle those tasks, and each finished task is paid.
The Information reported that micro1 pays the CEO of a Michigan HVAC company (a business that installs and repairs heating and air conditioning) $1,000 for each task, such as processing vendor invoices or doing payroll. At his planned pace, that comes to more than $150,000 in a quarter.
What moves a company within the range is explained in what your data is worth, and the data value estimate gives a starting range for your own company in under a minute.
05
What company data can you sell?
The short answer
A company can sell the records of how its work gets done: team chat, email, shared documents, help desk tickets, CRM history, project tools, written procedures and code. Records that connect into whole jobs and show how each job ended are worth the most. Patient medical records, privileged legal files and trade secrets are left out.
The systems buyers want
Each tool a company uses records a different part of the same work. Together they show a job from start to finish, which is what a model needs.
- Team chat
- Slack or Microsoft Teams threads, where staff discuss problems and make decisions.
- Conversations with customers, suppliers and colleagues, including quotes and follow-ups.
- Tickets
- Help desk and issue trackers such as Zendesk or Jira, which record a problem from the first report to the fix.
- CRM
- The history of each lead and customer, from first contact to won or lost deal and beyond.
- Documents and procedures
- Shared drives, knowledge bases and SOPs (standard operating procedures, the written steps for doing a task).
- Code
- Source code, documentation and bug histories, for companies that build software.
Connected records are worth the most
A connected record is one that can be linked to the others about the same piece of work, usually by a case number, a customer or a date. A support ticket that leads to a chat thread, an invoice and a thank-you email is one connected job, and a model can follow it in order.
Outcomes matter as much as connections. A record that shows whether the deal closed or the ticket was solved teaches a model which actions lead to good results.
What you hand over, and what you keep
Licensing data moves a copy, not the company's records themselves. The diagram below sets what the data vendor receives beside what stays with the company, item by item.
What is left out
Some records are excluded from every deal. Patient medical records, legally privileged material (communications between the company and its lawyers) and trade secrets are carved out, and a company can choose to exclude other categories, such as HR files. Excluding whole departments, such as sales or support, removes the work buyers pay for and lowers the offer.
06
Can my company sell its data? Who qualifies
The short answer
A company usually qualifies to sell its data to AI labs and data vendors if it has 25 or more employees, at least two years of work recorded in software, and records mainly in English. micro1 looks for US companies first, then other Western markets. Closed companies can qualify too, if they kept their records.
The checks a data vendor runs
Before pricing anything, a data vendor checks whether a company has enough connected, readable work on record for an AI lab to learn from. micro1 asks for a company with 25 or more employees. Other vendors set their own bars: Handshake AI looks for 20 or more full-time employees, SourceX for 50 or more at the company's peak, and HUMXN sets no size minimum.
- 25 or more employeesThe count that matters is staff whose work happens in shared software. For a closed company, it is the headcount at its peak.
- Two or more years of recordsMore years pay more. micro1's team calls five or more years ideal.
- Records mainly in EnglishSome records in other languages are fine. Talks with the vendor happen in English.
- Work done in softwareChat, email, tickets, a CRM and shared documents. Work kept on paper doesn't count.
- Any industrymicro1 accepts every industry, from HVAC contractors to software companies. Rare industries tend to pay more.
Closed companies can sell their data too
A company that has shut down can still license its records, as long as it exported them before its software subscriptions lapsed. Shutdowns now commonly include data sales: the head of a service that winds down startups told MIT Technology Review that 47% of its deals now include code or data sales. micro1 and SourceX both accept companies that have closed.
Each check is explained, with a one-minute checker, in does my company qualify?
07
How to sell data to companies, step by step
The short answer
Selling company data takes seven steps: apply, describe your data on an intro call, receive a priced offer, have a lawyer review and sign the license, transfer the records, let the data vendor check them, and get paid. With micro1, the offer arrives 1–2 days after the intro call and payment within 30 days of acceptance.
From application to payment
The steps below follow micro1's process, and other data vendors run close variants. Nothing confidential leaves the company until a license is signed.
Apply 2 minutes
A short form about the company: headcount, years in business and the software the team uses. Nothing confidential is asked for.
Intro call and data manifest Week 1
The company describes its records on a call. A data manifest is the written summary that results: which apps hold records, how many years each covers, what file types and roughly how much. No files change hands.
Priced offer 1–2 days after the call
The data vendor prices the dataset from the manifest. The company can negotiate or turn the offer down at no cost.
Legal review and contract At your lawyer's pace
The company's lawyer reviews the data licensing agreement, which covers permitted uses, exclusivity, anonymization, security, retention and deletion. Then both sides sign.
Transfer 1–7 days
The records move through an encrypted export or an approved connection. The data vendor anonymizes them before anyone else sees them.
Acceptance Up to 30 days
The data vendor checks that the records are complete and match the manifest, then accepts them.
Payment Within 30 days of acceptance
micro1 pays in full within 30 days of accepting the data, usually as a single payment.
How long it takes
The whole process usually takes two to three months from application to payment, by our estimate from micro1's stated timings. Most of that time comes after the transfer, while the data vendor reviews the records and the payment is processed. The diagram below splits the process at the signing.
08
Is selling company data legal and private?
The short answer
Selling company data is legal when the company has the right to license the records, personal details are anonymized, and the contract sets how the data may be used. The data vendor does the anonymization. A lawyer should check the company's rights, customer contracts and app terms before any data leaves.
How the records are anonymized
Anonymization means removing or replacing every detail that identifies a person, such as names, emails, phone numbers and addresses. Data vendors replace each detail with a consistent stand-in instead of blacking it out, so "Hey Sarah, can you call Mike about the Acme invoice" becomes "Hey Employee 14, can you call Contact 203 about the Client 7 invoice," and the conversation still makes sense.
The company doesn't anonymize anything itself, and records too sensitive to clean safely are deleted instead. The method, with a before-and-after example, is in how company data is anonymized.
The legal checks before you sign
Owning records doesn't always mean a company may license them. Customer contracts, privacy laws and other companies' confidential material inside the records can all limit what is licensable.
- Only the company can license its dataAn employee or contractor can't sell an employer's files, however they got them.
- App terms decide how data leavesSlack, HubSpot and Zendesk restrict AI training on data pulled through their APIs, so records usually leave through each tool's own admin export.
- Privacy law applies until the data is anonymizedGDPR in Europe and CCPA in California cover personal data, so the handover itself needs a contract.
Every check is set out in is it legal to sell company data? This page is not legal or tax advice.
09
Selling data vs. licensing it: which one are you doing?
The short answer
Almost every company that "sells" its data to AI labs and data vendors is licensing it. A license gives the buyer a copy and the right to use it for agreed purposes, while the company keeps ownership. A sale transfers ownership, and mostly happens when a company is acquired or closes.
What a license gives the buyer
A license is a contract that lets a buyer use data for agreed purposes, for an agreed time, while the owner keeps it. The license says whether the buyer may use the data for training, for testing or inside a product, and what happens to the raw copy when the term ends.
An exclusive license means the company licenses the records to one buyer only. Exclusivity pays more because no competitor can get the same data, but it rules out every other deal for the term.
What deletion can and can't undo
A deletion clause removes the buyer's copy of the records at the end of the term. It can't remove what a model has already learned from them, which is why anonymization and exclusions are the company's real protection. Deal terms, from flat fees to revenue shares, are covered in data licensing: how the deal works.
10
Other ways to sell data and earn money
The short answer
Besides AI labs and data vendors, a company can sell data on data marketplaces such as AWS Data Exchange, to data brokers, or as an outright database sale. Marketplaces suit a dataset the company can keep updated; brokers pay least and carry the most legal risk; database sales happen mostly in acquisitions and shutdowns.
The three older channels
Channel
Data marketplaces
Online storefronts such as AWS Data Exchange and Snowflake Marketplace, where a company lists a dataset, such as prices or shipping rates, and other businesses subscribe. Pay is recurring, but the company must clean, document and update the data.
Channel
Data brokers
Companies that buy personal information, such as contact lists, and resell it. Brokers pay little per record, and privacy laws give customers the right to opt out of the sale.
The riskiest channel for a company's reputation.
Channel
Outright database sale
The whole database changes hands, usually when the business is sold, wound down or in bankruptcy. Personal data stays bound by the privacy promises made when it was collected.
The channels are compared side by side, with a tool that picks one for your situation, in where to sell data.
Selling data is one form of data monetization
Data monetization is any way of earning a financial return from data, either by using it to run the business better or by selling or licensing it. Licensing records to AI buyers is its newest form. The full set of models is in data monetization: models and strategy, and dated, sourced deals are collected in data monetization examples.
11
Mistakes to avoid when selling company data
The short answer
The costliest mistakes are deleting old tools before exporting them, sending samples before a lawyer has reviewed the deal, scrubbing records by hand, excluding whole departments and accepting a rushed offer. Each one destroys records buyers would pay for or weakens the company's position.
Mistakes before the deal
- Deleting old tools and workspacesCancelling a chat tool or help desk can wipe out years of records. Export the history with the tool's own admin export before closing any account.
- Scrubbing the data yourselfDeleting names or threads by hand breaks the links between records that make them valuable. The data vendor anonymizes with consistent stand-ins.
- Assuming more files mean more moneyVolume matters least in pricing. Showing that records connect into whole jobs adds more than gathering every file.
Mistakes during the deal
- Sending a sample before legal reviewA sample is a disclosure, even a handful of tickets. Nothing should leave the company before a lawyer has read the agreement.
- Excluding whole departmentsLeaving out HR or medical records is normal. Leaving out sales or support removes the work buyers pay for.
- Accepting an exploding offerOffers that expire in 48 hours are common in this market. A serious buyer will still be there after your lawyer reads the contract.
- Signing exclusivity without pricing itAn exclusive license should pay noticeably more, because it rules out every other buyer for the term.
12
How to start selling your company's data
The short answer
To start selling your company's data, list every tool your team has used and the years each covers, check that you qualify, get a starting range from the estimate, then apply to a data vendor for a priced offer. Applying takes two minutes and nothing confidential is shared before a contract is signed.
Four things to do this week
- List every tool the team works in, such as chat, email, shared drives, the CRM, the help desk and project tools, with the years each one covers. That list becomes your data manifest.
- Stop any deletion: keep old workspaces running, or export them with each tool's admin export before cancelling.
- Check the basics in does my company qualify? and get a starting range from the data value estimate.
- Find a lawyer who can review a data licensing agreement, so the offer doesn't wait on legal review.
Why the timing matters
Prices are high because AI labs and data vendors want far more company records than companies are offering. MIT Technology Review expects that data prices will likely stabilize in the coming year. A company with years of connected records loses nothing by finding out its number now.
When you're ready, apply to micro1 for a priced offer. micro1 pays The Corner a referral fee out of its own money, and its published terms don't take that fee out of the company's payment.
Questions
Frequently asked questions
How do I sell my company's data?
A company sells its data by licensing an anonymized copy of its internal records, such as chat, email, tickets and CRM history, to AI labs and data vendors. The steps are: apply, describe the records on an intro call, receive a priced offer, have a lawyer review and sign the license, transfer the records, and get paid after the data vendor accepts them. With micro1 the offer arrives 1–2 days after the call.
Can I sell data to companies?
Yes. A business can sell or license its own records to AI labs and data vendors, which pay $100K to $2M or more per dataset for records that show how work gets done. It can also list a dataset on a data marketplace, or sell its database when the business is sold or closes. Individuals selling their own personal data use a different kind of service.
How much can a company make selling its data?
A company can make $100K to $2M or more per dataset by licensing its internal records to AI labs and data vendors, usually as one payment. The largest archives draw eight-figure bids: Spirit Airlines' records drew a $10M bid from Google and a $12.5M bid from micro1 in 2026. The price depends on how connected, complete and rare the records are.
Is selling company data legal?
Selling company data is legal when the company has the right to license the records, personal details are anonymized, and a contract sets how the buyer may use them. Customer contracts, privacy laws such as GDPR and CCPA, and app terms (Slack, HubSpot and Zendesk restrict AI training on API data) can limit what is licensable, so a lawyer should review the deal before any data leaves.
Who buys company data?
Company data is bought by AI labs, such as OpenAI, Anthropic and Google, and by data vendors that supply them, such as micro1, Mercor, Handshake AI, Polyshares, SourceX and HUMXN. Most companies sign with a data vendor, which anonymizes the records and turns them into training material. Labs sometimes buy very large archives directly, as Google did with Spirit Airlines.
Does my company need to be big to sell its data?
No, but it needs enough work on record. micro1 asks for 25 or more employees, at least two years of records and data mainly in English, and other data vendors set bars from 20 to 40 employees. Every industry can qualify, and closed companies can too if they had 25 or more staff at their peak and kept their records.
Do I lose ownership of my data when I sell it to AI companies?
No. An operating company that sells its data to AI labs and data vendors almost always licenses it, which gives the buyer an anonymized copy and limited rights while the company keeps ownership of the originals. An exclusive license stops the company licensing the same records to anyone else for the term. Ownership transfers only in an outright sale.
How long does it take to get paid for company data?
Selling company data usually takes two to three months from application to payment, by our estimate from micro1's stated timings. The priced offer arrives 1–2 days after the intro call, the transfer takes 1–7 days after signing, the data vendor reviews the records for up to 30 days, and micro1 pays in full within 30 days of accepting them.
Bottom line
Find out what your records are worth.
Apply in two minutes. Nothing confidential is shared, and you'll have a priced offer 1–2 days after the intro call.
Not legal, tax or financial advice. Figures come from micro1's published pages and public reporting as of November 2026.