Why price is hard to pin down
There is no public price list for operating data. MIT Technology Review reported in October 2026 that buyers' quotes vary widely because valuations rest on estimates of how many useful training tasks the data could produce. What can be pinned down is what pushes value up or down.
The eight drivers
| Driver | Higher value | Lower value |
|---|---|---|
| Demand | Domains labs are building agents for: support, legal, finance, software, sales | Domains with plenty of public data |
| Decisions and outcomes | Records of why something was done and what happened next | Final documents with no history |
| Connected systems | Tickets linked to messages linked to outcomes | Isolated exports |
| History | Several years of continuous records | A few months |
| Scale | Many people and lots of activity | A small team's light usage |
| Uniqueness | Work nobody else records | Generic, widely available data |
| Accessibility | Mainstream systems with clean exports | Paper, scans and custom formats |
| Rights | Clear rights to license, consent where needed | Restrictive contracts or unclear consent |
Rights work like a multiplier. Restrictions can take an otherwise valuable dataset to zero until they are resolved. See the rights checklist.
What doesn't matter much
Company revenue and profitability barely move the price of data. A struggling company with ten years of well-documented support tickets can hold more valuable data than a profitable one that keeps few records. That is also why wind-downs have become a source of supply.
Frequently asked questions
Does more data always mean more money?
No. Connected, well-documented records of decisions and outcomes beat a larger volume of disconnected files.
Sources
- AI labs are buying data from dead startups to train their models · MIT Technology Review, Oct 13, 2026