How to License Your Company's Data to AI Labs (2026 Guide)

You can license a de-identified copy of your company's internal records, such as email, chat, tickets, CRM and billing history, to AI developers, usually through an intermediary that buys the copy and resells it for model training. Published ranges run from about $100K to $5M, offers are case-by-case and not guaranteed, and the legal homework is yours.

This guide covers who is buying, who qualifies, what the data is worth on the record, how the process runs, and what can go wrong. It is written for owners, CFOs and general counsel at companies with roughly 6 to 1,000 office employees. Every number is attributed to its source, because in this market most numbers come from people selling something.

What does it mean to sell company data to AI?

In this market, "selling your data" almost always means licensing a copy. Your company keeps its original systems and records. A buyer receives an agreed extract covering named systems and date ranges, after names, contact details and other sensitive content have been removed or replaced.

Mode's program FAQ describes it this way: Mode "buys an agreed copy of company-owned operating records for AI research and training," including "email, documents, chat, CRM, and billing data," and removes or transforms sensitive information "before onward delivery." The agreement "defines its scope and any exclusivity."

What buyers want is not your customer list or your financials as such. They want the record of how work gets done. Dan Novaes, Mode's co-founder and CEO, put it this way on The LFG Show: labs are "buying the operational decision-making of companies." A support ticket that moves from complaint to engineering fix to refund approval is a worked example of a multi-step business task. That is the material AI developers need to train agents that do office work.

This is one form of what finance teams call data monetization. For a breakdown of the specific systems buyers ask for, see what data AI companies buy.

Why are AI labs paying for workplace data now?

Because the next product AI developers want to sell is an agent that can do white-collar work, and public web text doesn't show how that work happens inside a company.

The reporting is consistent on direction, even if every vendor's numbers are self-interested:

• Demand. Business Insider reported in October 2026 that AI labs building business agents, including OpenAI, Anthropic and Meta, are driving a market for company emails and Slack messages. The same piece cites a Gartner projection that spending on AI data will grow from under $1 billion to about $6.5 billion in 2027.
• Training methods. Scale AI says "nearly half of all of our new data training projects involve reinforcement learning environments" (Scale). Simulated offices are only as realistic as the real workflows behind them.
• Money committed. micro1 plans to spend $1 billion over the next year licensing company data and pays midsize companies $100,000 to $2 million, according to Business Insider.
• A public price signal. In August 2026, Google won a bankruptcy auction for Spirit Airlines' internal data and custom software with a $10 million bid (Business Insider). micro1 filed a competing $12.5 million bid (Chapter11Cases). The sale still needs court approval; the hearing moved to Nov 6, 2026 after objections from unions and 121 members of Congress (Travel Weekly). Google has not paid for that data yet.

Two cautions. First, much of the hype on social media traces back to sellers and brokers. Novaes said on the podcast that the companies that pioneered this space went from zero to nearly $100 million a month in three months; we found no source naming a company that did so, and we don't repeat it as fact. Second, demand can shift. Novaes himself said "modalities will change" (The LFG Show). For a running list of public deals, see our AI data licensing deals tracker and the logistics lessons in our Spirit Airlines analysis.

Who buys company data, and who actually pays you?

Most small and mid-sized companies won't sign with an AI lab directly. They sign with an intermediary that buys the copy, prepares it and supplies it to developers.

No public source names which AI labs buy Mode's data. Treat any claim that "every big lab" is buying through a given broker as marketing until you see a contract. We compare the intermediaries side by side in AI data brokers compared, and review Mode's program specifically in our BizBuyData review.

Does your company qualify?

Headcount of office employees is the main gate, because value tracks the volume of recorded decision-making. Novaes said on The LFG Show "you got to think like white collar work," and that a restaurant dishwasher's work offers "not much the AI labs can pull from."

Mode has published three different sets of minimums in October 2026:

The program page is the current company-facing figure, so we use it. The referral page limits referrals to the US, Canada, the UK and Australia. The press release counted "full-time U.S. office employees using digital workplace tools," and Mode's application asks for full-time desk employees and years of operating data. Confirm which threshold applies to you during intake.

Beyond headcount, Mode says the strongest fit is a company with "several years of records they own and teams primarily based in the U.S." Nonprofits are eligible under the same "all other" minimum.

Industry pages go deeper on each segment: law firms, accounting firms, SaaS and tech companies, healthcare practices, marketing agencies, nonprofits and logistics operators.

What data do AI companies want?

Mode lists six categories and names the systems it buys from, including Microsoft 365, Gmail, Slack, Jira, Confluence, Salesforce, HubSpot, Zendesk, Notion, Zoom, NetSuite, QuickBooks and Paychex (program page).

The exclusions column is our guidance, not Mode's policy. Mode's own list includes CRM, invoices and payments, so assume those are in scope unless you take them out. You choose which systems and date ranges are included before anything is exported. More detail is in what data AI companies buy, and for software companies, our SaaS page covers Jira, GitHub and support tickets.

How much could you be paid?

Here is everything on the public record that we could source:

• Mode's program page: "Your company's data could be worth $100K–$5M" (as of Oct 29, 2026), with the note that "the calculator is illustrative, not a guaranteed offer" (data.mode.inc).
• Mode's terms: figures "are not offers" and don't guarantee Mode will make one (ToS §3).
• Novaes on The LFG Show: labs are "willing to pay anywhere from a couple hundred [thousand] all the way to $5 million" (00:27), and the outcome could be "six or seven figures" (09:09). He gave a hypothetical of a 200-employee company receiving about $1 million.
• micro1: $100,000 to $2 million per midsize company (Business Insider).
• Mode's Oct 13 press release: no payout range.

Mode says it sets a fixed offer after reviewing "the systems and dates in scope, years of history, record volume and quality, and ownership and access terms." Headcount gets you through the door; those factors set the number.

Our valuation guide turns the published range into a per-employee estimate band by industry, with a calculator. The method, including its limits, is on our methodology page. It is an estimate built from Mode's published figures, not an appraisal and not an offer.

How does the process work, step by step?

Mode has described its process twice, and the two descriptions differ on timing.

Sources: Business Wire, data.mode.inc.

Plan on the longer figure. The work on your side looks like this:

1. Confirm eligibility. Count full-time desk employees, list systems and years of history, and check the minimum for your industry.
2. Map what is off-limits. Find client-confidential, privileged, tax-return, health and other regulated records, and any contracts that restrict use of company records.
3. Get counsel involved early. Your lawyer should review the proposed scope, the de-identification description and how you will notify employees.
4. Get the offer in writing. Amount, payment schedule, exclusivity, rights granted and data-handling terms.
5. Approve scope and support the export. Mode says your team signs in to approved systems and "no passwords are handed over." It describes the effort as "a few hours from your team."
6. Collect payment and keep records. Keep the agreement and scope documents, and ask your tax adviser how to treat the payment.

The detailed version, including agreement terms, is in how AI data licensing works.

Is it safe to license company data to AI?

It can be done carefully, but nobody can promise zero risk, and you should be wary of anyone who does.

What Mode says it does: it prepares the copy in a secure workspace, removes sensitive material, and removes or transforms "personal identifiers such as names and contact details" before delivery (program FAQ). Its demo replaces a customer name with "Company 01" and a contact with "Person 01." The press release says Mode built its own de-identification tools after other vendors wouldn't let it audit their PII removal. Novaes's example: "Dan's going to become... Derek" (The LFG Show).

What that doesn't settle:

• Re-identification. Workplace messages carry context: job titles, project names, unusual events. Swapping names doesn't remove that. The 121 members of Congress who objected to the Spirit sale argued that conventional de-identification isn't enough against AI-driven re-identification (Travel Weekly).
• Leaks. Business Insider reported that Meta suspended an employee keystroke-tracking program after an internal leak (BI).
• Legal definitions. Under California law, "deidentified" data requires reasonable technical measures, a public commitment not to re-identify, and contractual obligations on every recipient (Cal. Civ. Code §1798.140). Ask how the buyer meets each element.
• Insurance. Novaes said Mode is "working on" a third-party policy for potential data leaks (21:34). Until you see a policy in writing, assume there isn't one.

For the full risk checklist, read is selling company data to AI safe?. For the mechanics, see data anonymization for AI training and de-identified vs anonymized data.

What are the limits for law, accounting, medical and nonprofit organizations?

The lowest headcount minimums apply to the industries with the strictest confidentiality rules. That is not a coincidence, and it is not a reason to skip the analysis.

In each case, firm counsel or an ethics adviser should approve the scope. We are not lawyers and this isn't legal advice. Sector detail: law firms, accounting firms, healthcare practices, nonprofits. Agencies face a related question about client-owned work; see marketing agencies.

What about your employees?

Your employees wrote most of what you would license. That has legal and practical consequences.

• California. The employee and B2B exemptions in the CCPA/CPRA expired at the end of 2022 (CPPA FAQ), so California employees' personal information is fully covered. Notice obligations apply.
• Consent is contested. Emory law professor Ifeoma Ajunwa told Business Insider that whether employee consent extends to resale is "an ethically gray area." Employee privacy is also at the center of the objections that delayed the Spirit sale (BI).
• Morale. Staff who learn their messages were sold from a news story or a leak will not take it well. Tell them before export, explain what is excluded, and consider whether private DMs belong in scope at all.

Your counsel should decide what notice or consent your workforce and jurisdictions require.

How should you prepare before talking to a buyer?

Most of the value of preparation is in knowing what you will refuse before anyone makes you an offer. A one-page inventory does the job:

The rows above are an illustration, not a recommendation for your company. Fill in your own, then check three things against it:

• Ownership. Are these company accounts with admin access, or systems a client or vendor controls? Mode says it reviews "ownership and access terms" before making an offer.
• Retention and holds. Retention policies cap how many years you can offer. Records under a litigation hold or regulator request need your counsel's view before any copy leaves.
• Contracts. Client agreements, NDAs, vendor terms and employee handbooks may restrict how company records are used or disclosed.

With that page in hand, a buyer's first call becomes a scoping conversation rather than a sales pitch.

Who inside your company needs to sign off?

A data license touches more functions than a typical vendor contract. Before you sign, expect to involve:

• The owner, CEO or managing partner, who carries the reputational risk.
• Counsel, internal or outside, for scope, privacy law, professional rules and the agreement itself.
• Finance, to model net proceeds, tax treatment and any exclusivity cost.
• IT or the system administrator, who will sign in to approved systems and confirm what the export covers.
• HR, to plan employee notice and handle questions.
• The board, partners or trustees, where your governing documents require approval for asset sales or unusual contracts. Nonprofits should assume board approval is needed.

If any of these people learns about the deal after it closes, you skipped a step.

How does licensing data compare with other ways to raise cash?

A data license is a payment for an asset, not a loan and not equity. Mode describes it as a way to raise money "without taking on debt or giving up equity" (Business Wire). The trade-offs are different, not absent.

Full comparisons: data licensing as non-dilutive capital and revenue-based financing vs data licensing.

What should be in the agreement?

Before you sign, make sure the written agreement answers each of these:

• Scope. Named systems, date ranges and exclusions, attached as a schedule.
• Rights granted. Training and research use only? Can the buyer resell, sublicense or combine your data with others'?
• Exclusivity. Mode says the agreement defines "any exclusivity." Exclusivity limits your ability to license again later; price it.
• De-identification. A written description of the method and sample output before export. Who checks it, and what happens if identifiers slip through?
• Downstream obligations. Will every recipient be contractually barred from re-identification, as California's definition requires?
• Deletion and retention. What happens to the raw export after processing?
• Payment. Amount, schedule, and conditions. Mode says it charges no broker fee and that the agreement sets the amount and schedule before export.
• Liability. Indemnities, caps, and what happens if a leak or claim traces to the licensed copy.
• Governing law and termination.

Mode's terms say referral and program arrangements depend on signed agreements and can change (ToS). Read the agreement you are actually given, not the marketing page.

What are the red flags?

• Exploding offers. Business Insider described six-figure offers that expire in 48 hours (BI). Real buyers can wait for your counsel.
• Promises that re-identification is impossible, or claims of third-party sign-off on PII removal that nobody will show you in writing.
• A firm dollar amount before anyone has reviewed your systems. Even Mode's own calculator is illustrative.
• Requests for passwords. Mode says it doesn't take them; nobody else should either.
• Unnamed downstream buyers with no contractual limits.
• Pressure to include client files or privileged material on the theory that de-identification makes it acceptable.
• Encouragement for employees to upload employer files on their own. Business Insider reported that Mercor faces a lawsuit alleging it encouraged contractors to upload employer materials without consent; Mercor denies it.

What happens after you are paid?

Keep the signed agreement, the scope schedule and any record of what was exported in one place; you may need them if an employee, client or regulator asks what was shared. Diary any obligations that outlast payment, such as exclusivity periods, deletion confirmations or audit rights. If the license is non-exclusive, the same records, or newer ones, may be licensable again later, though nobody can promise demand will last. Finally, ask your tax adviser how to report the payment before year-end rather than after.

Where do we fit in?

DataLicensingAudit.com publishes guides and an estimate calculator for companies considering this market. We are an independent referral partner of Mode: if you ask us to introduce your company and Mode completes a deal with you, Mode may pay us a referral fee. You don't pay us. Our disclosure page explains the arrangement and Mode's published referral terms.

We are not a CPA firm, a law firm or an appraiser. We don't audit or appraise your data, and nothing here is legal or tax advice. Our job is to put the public facts in one place, flag what to check, and make an introduction if you want one. Talk to your counsel before you sign anything.

If you want to keep reading first, start with how much your data could be worth, then the safety checklist.