A guide to licensing your company's data to AI buyers

Some AI data companies pay businesses to license email, chat, documents, spreadsheets and the records kept in business software. What a company is offered depends largely on who has a buyer for that data at the time. AI developers want examples of how real businesses work, such as how people make decisions, use software and handle exceptions. Handshake AI’s employer data program is one public example of a company paying businesses for this material, and several others describe similar programs.

Telegraph Lab, a provider Data Licensing Report is commercially affiliated with, has described to Data Licensing Report one transaction it handled. Three buyers were approached about an online retailer’s data and two made offers. The owner received $300,000 within two weeks of the first conversation and used it to repay high-interest debt. That is one transaction, not an average or an estimate for any other company.

The short version

• Most companies that pay businesses for their data are intermediaries: they prepare the records and license them on to AI labs and other AI companies. Others introduce owners to those firms or handle the archives of companies that are closing.
• The first checks are how much history the company has kept and whether its records show how work gets done. Company age alone does not establish either.
• The same company can get different offers from different providers, because each provider has different buyers at a given time. Archive quality, scope and terms also move the number.
• Payment is upfront, a share of revenue from onward sales, or both. Published ranges are the providers’ own descriptions, not completed deals.
• Applying usually starts with a description of the company, not its data. An inventory, then a sample, come later.
• Before signing, compare scope, permitted uses, exclusivity, duration, payment trigger and the preparation work as well as the price.

What does licensing company data mean?

A data license lets a buyer use a defined set of your records, for defined purposes, for a defined period, in exchange for payment. You keep ownership. Appen’s program page says a partnership grants a defined license for a defined use and does not transfer ownership of the underlying data. Troveo says ownership “never transfers” when it licenses data on an owner’s behalf.

Not every arrangement is a license. SimpleClosure offers both sale and licensing, and its terms of service say that when assets are acquired, all right, title and interest pass to the acquirer. Check which of the two a proposal is before you compare its price.

Am I selling my data or licensing it? covers what you keep, what happens to the copy afterwards and what exclusivity changes.

Why would AI companies want ordinary business records?

Consider what happens when a property manager arranges a repair. A tenant reports a leak. Someone looks at the photos, requests quotes from two contractors, approves one, schedules the visit and checks the invoice against the quote. The emails, images, approvals and invoice together show how a task moved from request to completion: who decided what, with which information, in which tool.

That connected trail is useful to companies building AI systems that carry out business tasks. Handshake AI says it wants data that reflects how real organizations operate: decisions, workflows and tool use. Appen lists decision and escalation records, such as approval chains and resolution outcomes, among the categories it seeks.

So an ordinary archive of completed jobs, support cases or purchase orders can hold more of what buyers want than a polished description of the company.

Why would AI companies want our emails? goes further into what buyers say they use records for and which records they value most.

Could your company’s data qualify?

Three checks come before any conversation about price.

1. Your company has kept its digital history. Look at what still exists in email, chat, shared drives and business software, system by system. A ten-year-old company that keeps one month of chat messages has one month of chat to offer.

2. The records show how work gets done. Projects, customer requests, orders, approvals and problems leave trails across several systems, and buyers care about the connections between them. When you describe the company, start with the work it does; the file types come second.

3. The return justifies the work and the terms. Find out what a buyer needs before committing staff to an export or a cleanup. Then weigh the proposed payment against the effort involved and the rights requested.

Some providers also publish size thresholds, and they differ. As of October 2026, Handshake AI lists a minimum of 20 full-time employees and three years in operation, while Scale AI says its preferred starting point is 40 or more employees, assessed case by case.

Does my company’s data qualify? sets out each provider’s published minimums side by side.

Who pays businesses to license their data?

The end users are AI labs and other AI companies. The companies that pay owners are mostly intermediaries, and they work in different ways. The descriptions below come from each provider’s public pages as of October 2026. The buyers directory lists every provider Data Licensing Report tracks, with the same fields for each.

• Data companies that license and resell. They pay the business, prepare the data and supply it to AI labs. Handshake AI says it de-identifies and prepares data on its side and delivers it to frontier AI labs. Appen, founded in 1996 according to its about page, says it extracts and anonymizes operational data and licenses it to frontier AI labs. Telegraph Lab, which is affiliated with Data Licensing Report, says it licenses work conversations, business-software records and drive files from companies and prepares datasets for AI teams building models and agents.
• Licensing agents and marketplaces. They license your data on your behalf, potentially to several buyers, and share the revenue. Troveo says owners share in the revenue on every sale and that it has more than 40 active buyers, which it does not name.
• Introducers. They match a company with another firm that does the licensing. Avelence says it takes a company profile, with no data, and matches it to a data partner that evaluates, licenses and pays. Its terms say the partner may pay Avelence for a successful introduction.
• Wind-down platforms. They sell or license the records of companies that are closing. SimpleClosure’s AssetHub handles source code and workspace data such as Slack, email and Google Drive. In September 2026 SimpleClosure said AssetHub was open to operating companies as well.

Troveo vs Avelence shows how an agent and an introducer differ in practice.

Why do offers for the same company differ?

Largely because of who has a buyer for that data right now. Most providers in the directory describe passing the prepared data on to AI labs and other AI companies. A provider whose customers are currently asking for a particular kind of record can price yours against that request. A provider without such a request has to judge whether a buyer will turn up later, so it may offer less or pass. For example, a provider supplying a lab that is collecting support-ticket histories might make an offer to a software company that another provider, with no such customer, declines the same week.

It works like applying for a business loan. Lenders looking at the same accounts reach different decisions, because each has its own appetite for an industry, a company size and a type of loan at that moment. One approves, one declines and a third offers different terms. None of those answers settles what the business is worth.

Providers say as much in their own materials. Miro Advisory says dataset value “depends on buyer demand, quality, volume, rights, permitted uses and other factors,” and that it does not represent every AI developer. Avelence’s terms say it does not claim to represent every buyer or guarantee the highest available return. micro1 says its current demand is strongest for companies based in the United States, the United Kingdom and Canada. In the retailer transaction described by Telegraph Lab, which is affiliated with Data Licensing Report, three buyers were approached and one made no offer.

Beyond buyer demand, the archive, the scope and the terms move the price. Archive quality covers how much history survives, how well it connects across systems, and how much has to be removed to protect people and clients. Scope is which systems, years and record types are included; Mercor says connecting more tools generally results in a higher payout. Exclusivity, permitted uses and the payment model change what the same records are worth to a buyer.

How much is my company’s data worth? lists the ranges providers publish, and how AI companies price company data explains the inputs behind them.

How do offers work?

Published processes generally follow the same order: a short description of the company, a call or review to scope the data, a written agreement, then export, de-identification, delivery and payment. Handshake AI describes an intake form followed, if there is a fit, by a scoping call on data parameters, de-identification, terms and timeline.

Early figures are estimates, not offers. Avelence labels its calculator range a model estimate rather than a buyer offer, and Handshake AI’s estimator says any payout depends on the data, due diligence and acceptance of its licensing terms. A number becomes firm once the provider has seen enough of the data and the terms are written down. How do offers work? covers each stage and what can change between an indicative figure and a signed one.

Should you take upfront payment or a revenue share?

Upfront payment is a fixed amount paid when a milestone in the agreement is met. A revenue share pays you a portion of each onward sale, so income depends on how often, and for how much, the data sells. Some offers combine the two. As of October 2026:

• Appen says payment is triggered by acceptance of the data, not by submission.
• Troveo says owners share in the revenue on every sale. It does not publish the percentage.
• Replay says its offers are typically an upfront cash payment plus a perpetual revenue share.

The choice trades certainty and timing against possible later income, and the trigger and calculation that count are the ones in your proposal. Upfront payment or revenue share? works through both. The directory also lists providers that describe upfront payment and those that describe a revenue share.

What are a data inventory and a data sample?

An inventory describes what you have: the systems, the years each one covers, rough volumes and anything you would exclude. A sample is a small set of representative records that lets a provider check quality before it firms up an offer. What is a data inventory? and What is a data sample? explain what goes into each.

Neither is needed at the start. Avelence says its first step uses only a description of the company and its systems, with no system connections or uploads. Replay says its valuation form gives an initial view without touching raw customer or employee data. Working in that order (description, inventory, sample, then full export) keeps your team’s effort in step with a provider’s interest. The sample can serve the seller too: Appen says nothing is used without the company’s sign-off on a representative sample pack.

What should you compare before signing?

Price is one line of the agreement. These terms decide what you are agreeing to, so check each one in the agreement itself.

Term	What to check
Scope	Which systems, years and record types are included, and what is excluded, such as particular clients or privileged material
Permitted uses	What the buyer may do with the data, such as training or evaluation, and whether it may sublicense or resell it
Exclusivity	Whether you can license the same records to anyone else, and for how long you can’t
Duration	How long the license runs, and what happens to delivered copies when it ends
Payment trigger	The event that releases each payment (signing, delivery, acceptance or each onward sale) and how long after it the money arrives
Preparation work	Who exports, who de-identifies, who pays for that work, and how many of your staff hours it takes

Published terms already differ on these points. Polyshares lists payment at net 30 to 60 days after the data is approved, shared and anonymized. On preparation, Troveo says the company does the exports with its guidance, Appen says it bears the extraction, anonymization and transfer costs, and Mercor describes linking the company’s accounts and running automated extraction.

Before signing, also read what’s the catch, whether it is legal to sell company emails and records and the other questions owners ask.

What does the market look like now?

Three dated public facts give a sense of scale. They describe different situations and are not prices for any one company.

• Spirit Airlines. Bloomberg Law reported that Google was named the winning bidder, at $10 million, for de-identified business data from defunct Spirit Airlines in an August 2026 bankruptcy auction, with Mercor’s $7.5 million offer named as the backup bid. As of late September 2026 the sale still needed approval from the bankruptcy court, Moneywise reported.
• Closing startups. Forbes reported in April 2026 that SimpleClosure had processed nearly 100 data deals for defunct companies in the prior year, typically paying $10,000 to $100,000 per company. Those figures are the company’s statements, relayed by Forbes.
• Operating companies. Handshake AI says its qualified employer data partnerships typically range from $100,000 to $4 million, depending on the data (as of October 2026). That is Handshake’s own description, not a record of completed deals.

The market facts page keeps a dated, sourced list of deals, programs and reporting.

How do you get offers, or check one you already have?

Start with a company profile. The application is one short form about your company: what it does, its staff and years in operation, the software it uses and how far back each system goes. It asks for no data, exports or files. It comes to our team for review. We contact you about possible licensing routes and any further inventory details needed.

Compare an offer you already have. If a provider has made you an offer, the benchmark walks through the terms listed above one at a time (payment and its trigger, any revenue share, exclusivity, duration, permitted uses and the work expected of you) so you can see what the offer covers and what it leaves open. “Unknown” is a valid answer to every question.

Data Licensing Report may earn a referral fee when a business that applies through the site completes a deal with a participating provider. How we make money explains the arrangements.