Who buys business data for AI? A buyer-by-buyer guide
At least a dozen companies now pay businesses to license their past email, documents, chat and tickets for AI training. Here is what each one buys, who qualifies, and what the terms look like, as of November 2026.
The short answerAs of November 2026, the main companies licensing business records for AI training are Mode (BizBuyData), Handshake AI, micro1, Scale AI, Mercor, Polyshares, Grepped, Endless Data and Labelbox Prism. Brokers such as Ventra shop records to several of them. For closing startups, SimpleClosure and Sunset handle sales. Most want established companies with at least 20 to 50 staff and several years of records, mainly in the US, Canada, the UK and Australia. Exclusivity and payment timing vary widely, so terms matter as much as price.
Key findings
- At least a dozen companies now license business records for AI training. Most launched or expanded programs between June and October 2026.
- Most want 20 to 50 or more staff and several years of records. Professional firms (law, accounting, finance, medical) can qualify smaller at some buyers.
- Exclusivity varies the most. Terms range from non-exclusive to perpetual exclusive licenses that bar any future sale of the same records.
- Payment timing ranges from about two weeks after signing to about three months from first contact.
- Published payout ranges are marketing claims. We found almost no independently confirmed completed deals for operating companies.
How to read this guide: Verified means confirmed by the company's formal terms or an independent news outlet. Company claim means a marketing statement or a single source we could not confirm. Payout ranges on company websites are marketing, so we label them as claims. We have no completed-deal data for most buyers.
Why AI companies are buying business records
AI developers have used up much of the easy public text. What they lack is a record of how real work gets done: how a support team resolves a ticket, how a contract is negotiated, how a project moves from first email to delivery. Companies hold years of exactly that.
The market went mainstream in 2026. Shut-down startups were the first sellers: SimpleClosure had arranged close to 100 such deals at $10,000 to $100,000 each by April 2026 Verified 1. In August 2026, Google paid $10 million for bankrupt Spirit Airlines' data, which held about 100 million emails and 500 million Teams messages, outbidding Mercor Verified 2. By October, several companies had launched programs aimed at operating businesses, not just closed ones.
There is no federal law in the US that bans these sales Verified 2. That does not make every sale lawful: client contracts, professional confidentiality rules and privacy laws can still limit what a business may license.
The buyers at a glance
| Company | Type | Size floor | Countries | Exclusivity | Timing |
|---|---|---|---|---|---|
| Mode (BizBuyData) | Direct buyer | 6 to 10+ law, 10 to 15+ accounting, finance, medical, 30 to 50+ others (its pages differ) Claim | US, CA, UK, AU | Set in agreement | About 3 months to payment Claim |
| Handshake AI | Direct buyer | 20+ staff, 3+ years Verified | Not stated | Not published | Contact in 1 to 2 days Verified |
| micro1 | Direct buyer | 30+ staff Claim | US first, then UK, CA | Not stated | 30 days to approve data, pays within 30 days of invoice Claim |
| Scale AI | Direct buyer, recurring | 40+ preferred Verified | Not stated | Not stated | Not stated |
| Mercor | Direct buyer | None published; leadership has targeted 100+ Claim | Not published | Not published | 2 to 4 weeks after signing Claim |
| Polyshares | Direct buyer, resells to labs | None stated Claim | Not stated | Exclusive by default; published deals were perpetual Claim | Offer in 1 to 2 weeks, net 30 to 60 Claim |
| Grepped | Marketplace | Not published | Not stated | Some 2 to 3 year licenses Claim | Not published |
| Endless Data | Aggregator | About 30+ Claim | US, UK, CA first | Non-exclusive Claim | Fixed sum agreed before sharing Claim |
| Labelbox Prism | Direct buyer | Not published | Not stated | Exclusive by default Claim | About 21 days to first payment Claim |
| Troveo | Licensing agent | None | Not stated | Time-limited Claim | Revenue share on each sale Claim |
| Ventra | Broker | 30+ staff, 3+ years; excludes legal, healthcare, financial, accounting Claim | US, UK, CA | Not stated | Offers in 3 to 7 days, net 30 to 60 Claim |
| SimpleClosure | Wind-down marketplace | Closing startups, now some running companies | Mostly US | Not stated | Deals of $10K to $100K Verified |
| Sunset | Wind-down service | Closing companies | Not stated | Not stated | Not published |
Scroll the table sideways on a phone. Sources for each entry are in the profiles and source list below.
Buyer profiles
Mode (BizBuyData)
Mode Mobile, Inc., best known for consumer apps, launched BizBuyData on October 13, 2026 Verified 3. It licenses operating records such as email, documents and tickets. Its pages describe an estimate, then a 14-day nonbinding letter of intent while it reviews the data, a final offer that can differ from the estimate, 15 to 30 days to export and de-identify records, and payment about 60 days later, roughly three months in all Claim 4. It says it asks for no passwords, charges no broker fee, and lets the company choose systems and date ranges Claim 4. Its size floors differ between pages, and its advertised range of $100,000 to $5 million is marketing Claim. Professional ethics guidance urges caution for firms holding client information: the American Bar Association says lawyers need informed client consent before client information goes into AI tools that learn from it, and AICPA ethics staff said in November 2026 that the default answer to selling client data for AI training is no 5.
Handshake AI
Handshake, known for its college hiring platform, runs a data partnership program that licenses company data to AI labs. It lists sources including Slack, Gmail, Google Drive, NetSuite and QuickBooks, and asks for 20 or more full-time employees and three or more years in business Verified 6. It advertises $100,000 to $4 million or more per company Claim 6. Exclusivity and payment timing are not published.
micro1
micro1 buys SOPs, tickets, CRM and finance records, legal documents and internal communications. It asks for 30 or more full-time staff with English records, starting with the US, then the UK and Canada, and runs a separate program for companies that have closed Claim 7. It says it has 30 days to approve data after transfer, then pays within 30 days of invoice Claim 7.
Scale AI
Scale runs ongoing data partnerships covering email, chat, documents and business systems such as Jira, Salesforce and Zendesk, with records de-identified inside the company's own environment. It prefers 40 or more employees and lists software, finance, legal, healthcare, industrial, public sector and retail Verified 8. It describes the value as recurring, scaling with how often data is delivered Claim.
Mercor
Mercor's program covers tools such as Slack, Teams, Gmail, Outlook, Confluence, Salesforce and GitHub, and says more tools shared means a higher payout. It publishes no size floor, though its leadership has publicly targeted companies with 100 or more employees Claim 9. Payment by wire is reported at two to four weeks after signing, the fastest stated among large buyers Claim 9.
Polyshares
Polyshares says it buys licenses directly and resells to AI labs, from companies in any sector with real operating history. Its published examples, a roofing company, a general contractor, a solar company and a marketing agency, were all perpetual exclusive licenses Claim 10. A perpetual exclusive license means the business can never license those records to anyone else.
Grepped, Endless Data, Labelbox Prism and Troveo
Grepped runs a marketplace and says recent manufacturing deals were two-year licenses Claim 11. Endless Data bundles records from many companies, excludes client personal data, health data and privileged legal content, and is unusual in offering non-exclusive terms Claim 12. Labelbox Prism pays an upfront fee plus royalties and is exclusive by default Claim. Troveo started with video archives and now licenses business records on a revenue share Claim 13.
Brokers and wind-down services
Ventra is a broker that shops a company's records to several labs. It asks for 30 or more staff and three years in business, and currently excludes legal, healthcare, financial services and accounting companies Claim 14. SimpleClosure and Sunset help closing startups sell code and workspace data. SimpleClosure reports more than 150 deals, with 47 percent now including code or data Verified 2.
What to compare before you sign
- Exclusivity. Perpetual or multi-year exclusive deals stop you from licensing the same records again.
- Scope. Which systems and date ranges, and who decides what is excluded.
- De-identification. How names and identifying details are removed, and whether you can check results first.
- Payment trigger. Paid on signing, on delivery, or only after the buyer approves the data.
- Permission. Whether your client contracts, professional rules and privacy laws allow it. Have your own lawyer review the agreement.
Get the next update
Terms in this market change monthly. We'll email you when this guide changes and when we publish new reports.
How we did this
Between October and November 2026 we reviewed each company's public seller pages, published terms and press coverage, plus independent reporting from Forbes and MIT Technology Review. We did not sign up for any program or receive any company's private terms. A fact is labeled verified only when a company's formal terms or an independent news outlet confirms it. Everything else, including every payout range, is labeled as a company claim. Terms in this market change often, so check each company's current terms before relying on them.
Cite this report
Mark Kessler. "Who buys business data for AI? A buyer-by-buyer guide." The Data Licensing Report, November 6, 2026. https://thedatalicensingreport.com/reports/who-buys-business-data-for-ai
Download the data: ai-data-buyers-2026-11.csv (13 rows, with source links and the date each was checked).
Sources
- Forbes, "AI's new training data: your old work Slacks and emails," April 16, 2026. https://www.forbes.com/sites/annatong/2026/04/16/ais-new-training-data-your-old-work-slacks-and-emails/
- MIT Technology Review, "AI labs are buying data from dead startups to train models," October 13, 2026. https://www.technologyreview.com/2026/10/13/1145755/ai-labs-buying-data-from-dead-startups-to-train-models/
- Mode Inc. press release via Business Wire, October 13, 2026. https://finance.yahoo.com/technology/ai/articles/mode-inc-launches-5-billion-151600765.html
- Mode BizBuyData website. https://data.mode.inc
- American Bar Association, Formal Opinion 512, "Generative Artificial Intelligence Tools," July 29, 2024. https://www.americanbar.org/content/dam/aba/administrative/professional_responsibility/ethics-opinions/aba-formal-opinion-512.pdf; and Journal of Accountancy (AICPA), "Asked to sell firm data for AI training? Be very cautious," November 2, 2026. https://www.journalofaccountancy.com/issues/2026/nov/asked-to-sell-firm-data-for-ai-training-be-very-cautious/
- Handshake AI data partnerships. https://joinhandshake.com/ai/data-partnerships/license/
- micro1 data partnerships. https://www.micro1.ai/data-partnerships
- Scale AI data partnership. https://scale.com/data-partnership
- Regulr buyer profile of Mercor (an advisory site that may earn fees). https://regulr.ai/data-licensing/buyers/mercor
- Polyshares track record. https://polyshares.com/track-record
- FourWeekMBA summary of a Grepped founder post, October 2026. https://fourweekmba.com/ai-the-data-wars-ai-data-broker-says-deals-now-run-1-1m-1-3m/
- Endless Data website. https://endless-data.com/
- Troveo, licensing company data for AI. https://www.troveo.ai/resources/license-company-data-for-ai
- Ventra company data and partner pages. https://joinventra.com/company-data
Corrections: none so far. If you spot an error, email editor@thedatalicensingreport.com and we'll fix it and note the change here.