Support tickets & chats
Problem, back-and-forth, fix and outcome, in order.
Zendesk, Intercom, Freshdesk, ServiceNow, Jira Service Management
AI labs license real business records to train and test their models: email and chat, support tickets, CRM history, documents, code, finance and operations. Talkbase values what you have, finds the buyers and runs the deal. Personal details come out, your name stays private until you approve, and you sign off on every term.
To act like a good employee, an AI model needs examples of real work: how a request came in, what people tried, what they decided and how it ended. That record lives inside companies like yours, in systems nobody outside can scrape.
Labs now buy it directly, and the companies that sell first set the terms. Most businesses with 25 or more people at desks and a few years on the same tools already hold enough to interest a buyer. They just store it.
What makes company data valuable →Buyers pay for records that show reasoning and results. Tap any card to add it to your estimate. It updates everywhere on the site.
Problem, back-and-forth, fix and outcome, in order.
Zendesk, Intercom, Freshdesk, ServiceNow, Jira Service Management
Threads where requests, pushback and decisions happen.
Gmail, Outlook, Slack, Microsoft Teams, Google Chat
Deals through every stage, with notes on why they were won or lost.
Salesforce, HubSpot, Pipedrive, Zoho CRM, Microsoft Dynamics
How your processes run, with versions and comments.
Google Drive, SharePoint, Confluence, Notion, Dropbox, Box
Commits, pull requests and reviews that explain why code looks the way it does.
GitHub, GitLab, Bitbucket, Azure DevOps
Orders, dispatch, work orders, quotes and exceptions with timestamps.
Shopify, ServiceTitan, Procore, Samsara, Epicor, a TMS or WMS
Move anything and every number moves with it. The free valuation turns this into a real figure.
60 staff · 5 yrs · Other / mixed
Illustrative only. Anchored to ranges buyers and brokers published in 2026; the weights for team size, history, data type and industry are our own assumptions. Your picks carry over to the valuation form, the rights checker and every data card on this site. An estimate is not an offer.
Most company deals stay private under confidentiality clauses. These are the ones that were reported.
| Seller → buyer | What was licensed | Reported value | Source |
|---|---|---|---|
| Reddit → Google | Public posts and comments | About $60M a year | Reuters, Feb 2024 (reported, not confirmed by the companies) |
| News Corp → OpenAI | News archive and current journalism | Reported at more than $250M over five years, partly in credits | Wall Street Journal, May 2024 (reported) |
| Shutterstock | Image and video library, several buyers | $104M AI licensing revenue in 2023 | Bloomberg, 2024 (reported) |
| Taylor & Francis → Microsoft | Academic books and journals | $10M initial fee plus recurring payments | Company disclosure, 2024 |
| Spirit Airlines estate → Google | About 100M internal emails, 500M Teams messages, operations records and code, de-identified | $10M winning bid | Bankruptcy court notice, Aug 2026, via Bloomberg Law and TIME. Court approval pending; sale hearing set for Nov 6, 2026. |
| Closed startups → AI labs | Slack, email, Jira and code archives | $10K to $100K each, about 100 deals in a year | Forbes, Apr 2026, citing wind-down firm SimpleClosure |
Company names appear only to report public deals. They do not imply partnership, endorsement or current buyer interest. These are mostly publishers and platforms with clean rights at huge scale, so they are the top of the market, not a benchmark. A single company is priced on how hard its records are to find elsewhere. The rules are the same: rarity, history, outcomes and clean rights.
About ten minutes from you to start, and a few hours over the whole deal. We do the rest.
Team size, systems, years and who owns what. Counts only, no files.
~10 minutesWe separate what you own, read your terms and agree what stays out.
Week 1–2An anonymous one-page summary goes to buyers under NDA. Never your data.
Week 2–6You pick the offer. Data is de-identified, delivered and paid on milestones.
Typically 30–90 daysCould the buyer get something similar elsewhere? A decade of one specialist workflow beats a bigger generic archive.
Years on the same tools. A software switch resets the clock, so five unbroken years can beat ten split ones.
Solved or not, won or lost, approved or rejected. Results turn records into lessons a model can learn from.
A ticket tied to the chat, the code change and the CRM note is worth more than any of them alone.
Enough data gets a buyer to look. After that, price follows quality and scarcity, not gigabytes.
Data you can't show you may license is worth little. Clear ownership speeds the deal and protects the price.
Buyers see an anonymous summary first. Your name is released only after you approve.
Names, numbers and accounts become consistent placeholders, checked by people before delivery.
You approve the buyer, price, scope and use. Turn down any offer at any time.
You license a copy for a defined use. Ownership and your own use never change.
For established US companies with 25+ people at desks and years of records on the same tools, in any industry. Software, finance, legal, manufacturing, logistics, construction, healthcare admin, field service and more.
Licensing. A buyer gets the right to use an agreed, de-identified copy for agreed purposes. Your company keeps ownership and keeps using its data.
Licensing your own operating data is an ordinary commercial agreement. Three checks decide each dataset: you own it, your contracts allow de-identified use, and personal information can be removed. See how we check.
Nothing upfront. We're paid a success fee only when a deal closes, at a rate agreed in writing before we start.
No. The valuation asks for counts and years, not files. Later you export only the agreed data, after you've approved a buyer.
Deals like this typically take 30 to 90 days from first call to signed license, depending on how much data there is and how much testing the buyer wants. That's an estimate, not a promise.
A free valuation takes about ten minutes. No files, no logins, no obligation.