Years of company records you never thought of as an asset.
AI labs and the firms that supply them license real operating records from established businesses: the email, documents, CRM notes, invoices and project history that show how work actually gets done. If your company has 20 or more staff and several years of records, they may have a price.
About 4 minutes. You answer questions about your systems; nothing is connected or uploaded.
A freight company’s records, year by year
- 2025
- 2024
- 2023
- 2022
- 2021
- 2020
- 2019
- 2018
- 2017
- 2016
- Shared drives
- TMS records
- Invoices
- Call recordings: left out
What AI labs are buying, in plain terms
Labs are training AI agents to do real office work: answer the customer, update the file, raise the invoice, close the job. To practise, an agent needs realistic workplaces. The industry calls these RL environments: simulated companies built from real companies’ records.
Public text is everywhere. What is scarce is the private trail of how a business runs: who asked for what, what went wrong, how it was fixed, and what it cost. An established company produces exactly this trail, across many customers and many years.
Demand is public. One buyer, micro1, says it will spend $1B over 12 months licensing companies’ operational data. And connections matter: a firm that resells company data told MIT Technology Review that linked records are worth more than isolated documents.
The records that count
- Email and calendarGoogle Workspace or Microsoft 365, with years of customer and supplier threads.
- Documents, chat and projectsShared drives, Slack or Teams, and tools like Asana, Monday or Jira.
- CRM and accountingHubSpot or Salesforce, QuickBooks, Xero or NetSuite.
- Years of itLonger history means more complete examples for a buyer to build from.
Companies whose records buyers look at
Examples, not a promise. Size, years of records and how many systems hold them matter more than the industry itself.
- Accounting and bookkeeping firmsClient correspondence, workpapers in shared drives, billing and engagement history.
- Logistics, freight and truckingLoad emails, TMS records, claims, carrier calls and invoices.
- Engineering, architecture and constructionRFIs, project correspondence, drawings registers, change orders and site reports.
- Real estate and property managementTenant requests, work orders, leasing threads and owner statements.
- Manufacturing and distributionERP orders, quality reports, supplier email and purchasing history.
- Marketing, consulting and digital firmsBriefs, proposals, project tickets, call recordings and, for software teams, code history.
Who this is for
- 20 or more full-time staffPublished buyer minimums start at 20 to 40 employees.
- Three or more years of recordsMost buyers look for an established operating history.
- The owner or a senior executiveOwner, CEO, president, partner or director: someone who can sign.
- Records the company owns, at least in partYour factual answer; the buyer and your own lawyer assess the rights.
Who it isn’t for, for now
We don’t take on regulated industries where records are mostly other people’s protected information. You can still fill in the form; we record your answers but won’t show a range or make introductions.
- Medical, dental or healthcareMedical, dental and healthcare records are full of patient information, which is tightly regulated.
- InsuranceInsurance records are mostly policyholders’ and claimants’ personal and financial details, which are tightly regulated.
- Staffing or recruitment agencyA staffing agency’s records are mostly candidates’ and workers’ personal data.
- Title or escrowTitle and escrow files are mostly consumers’ identity and financial details, which are tightly regulated.
- Law firm or legal servicesLaw firm records are covered by client privilege.
What buyers say they pay, and what deals have been worth
Public figures only, each linked to where it was published. Most are buyers’ own headline ranges, which are claims, not deals. None of these are our deals: we list them so you can judge our estimate against the market.
| What was sold or offered | Published figure | Source |
|---|---|---|
| Company records, priced by staff sizeReplay’s headline bands; it says head count is only a proxy. A buyer’s claim, not a reported deal. | $10K–$100K for 20–50 employees; $100K–$1M for 50–250; $1M+ for 250+ | Replay’s own page, read 2026-10-18 |
| Data programme for operating companiesDescribed by the buyer as indicative, subject to a full review. | $100K–$5M | Mode’s own page, read 2026-10-18 |
| Company data licenceCorpus’s own headline figures, subject to its diligence. A buyer’s claim, not a reported deal. | Deals “start at $100K”; largest to date “$2M” | Corpus’s own page, read 2026-10-18 |
| Data licence from a companyTelegraph Lab’s headline range, subject to data review. A buyer’s claim, not a reported deal. | $100K–$4M | Telegraph Lab’s own page, read 2026-10-18 |
| Data licence from a companyHandshake’s headline range. A buyer’s claim, not a reported deal. | $100K to $4M+ | Handshake AI’s own page, read 2026-10-18 |
| Company data from qualifying businessesPublished by the buyer; it asks for 20+ employees and 3+ years in business. | “Six figures” (in dollars) for qualifying businesses | Nyne’s own page, read 2026-10-17 |
| Licence of an operating company’s dataPublished by the buyer. The top tier is for “highly unique” data. | Tiers of $100k+, $500k+ and $1M+ | micro1’s own page, read 2026-10-17 |
| Data licence from an operating companyDescribed by the buyer as illustrative. | “$10K–$1M+” per agreement | Scale AI’s own page, read 2026-10-17 |
| 13 years of Slack, email and Jira from one company | “Hundreds of thousands” of dollars | Forbes, via Slashdot |
| One data point: code and workspace data of shut-down startupsTypical payouts across 150+ deals by a wind-down service. | $10k–$100k per company | SimpleClosure, reported by MIT Technology Review |
See the full price index, with every published range and reported deal, and the market log of launches and changes.
Offers are not guaranteed. Many companies never get one. One code marketplace reports that only a low single-digit share of listed codebases find a buyer. (the marketplace’s own page, read 2026-10-17) Buyers price on their own review of what the records can teach, so a final offer can be lower than any estimate, or there may be none.
How we work
We match established companies with established AI-data buyers. You keep control of your data throughout, and we never hold it.
Read the full process
Read our data-handling pledge
Read the guide to selling company data to AI
How the valuation is calculated
Free valuation
Answer questions about your industry, size, systems, years of records and who owns client work. You get an indicative range on the spot.
Call and NDA
A 20-minute call with our Partnerships team to check fit. If it makes sense, we sign a mutual NDA.
Introduction to buyers
We describe your records to buyers who fit, without sharing any data. You decide whether to take any offer further.
See your indicative range
Free, about 4 minutes, and nothing leaves your systems.