What is my company's data worth? How AI buyers price archives
There's no price list. Buyers price business archives on headcount, years of history, which tools hold the data, how much there is and the industry, then discount for anything that's hard to clear. Reported deals for small startups run about $10,000 to $100,000 per company; very large archives have drawn bids in the millions.
"What is our data worth?" is the first question every founder, trustee and assignee asks. The honest answer is a range, and the range is wide. This guide explains how buyers get to a number, what the public deals tell us, and what you can do to land at the top of the range instead of the bottom.
What the public deals say
There is no exchange and no price list, but enough deals have been reported to sketch the market:
| Deal | Reported value | Source |
|---|---|---|
| Startup wind-downs through SimpleClosure (nearly 100 deals in a year) | $10,000 to $100,000 per company | Forbes, via Gizmodo[2] |
| One crypto startup's codebase | about $25,000 | MIT Technology Review[1] |
| The same startup's Google, Notion, Figma and Slack data | about $15,000, pending | MIT Technology Review[1] |
| Spirit Airlines' de-identified enterprise data (about 100 million emails, 500 million Teams messages) | $10 million winning bid by Google; $12.5 million competing bid later noticed by micro1 | MIT Technology Review[1]; Stretto[3] |
Two things stand out. Most deals are small-company deals in the tens of thousands of dollars. And the spread is huge, because buyers themselves disagree: a buyer quoted by MIT Technology Review said valuations rest on "research theses" about how many useful training tasks a dataset will yield, which is why quotes vary so much.[1]
The five factors buyers price on
Buyers and wind-down firms describe the same handful of inputs. One firm told MIT Technology Review it values a company's data by headcount, years in operation, volume of data and how the team used its tools.[1] Here's how each one moves the number.
1. Headcount
More people means more conversations, more handoffs and more variety. Peak headcount (everyone who ever used your systems) usually matters more than how many people are left. Very small teams can still sell, but the per-company number tends to be lower, and code is often the most valuable part.
2. Years of history
Longer history captures more situations: launches, incidents, customer escalations, changes in process. Buyers generally look for several years of records. Under about three years, expect the number to drop.
3. Which tools hold the data
Buyers want the systems where work actually happens: email, team chat, ticketing, docs and SOPs, CRM, code, and recordings of multi-step tasks. A company whose work is spread across several of these is worth more than one with only a mailbox. Workflow-heavy data, such as tickets and screen recordings, is in particular demand from companies building AI agents.
4. Volume and richness
Volume helps, but variety and depth matter more. The buyer quoted above said one company's Slack history, if it captures conversations with real variety and depth, can be worth more than another's Slack, email and calendars combined.[1]
5. Industry
Buyers want coverage of work that public data barely shows: logistics, construction, accounting, legal, healthcare administration and similar process-heavy fields. Common, well-covered domains are worth less per company. This is our read of buyer demand, not a published price table. unsourced estimate
What pulls a number down
The headline factors set a ceiling. These set how much of it you keep:
- Data that's gone. Cancelled accounts are often deleted. An archive that only covers the last year of a ten-year company is priced as a one-year archive.
- Personal and regulated data. Heavy consumer data, health data or financial data either gets excluded or raises the buyer's cost and risk.
- Employee data. It's the most contested part of these deals. At Spirit, four labor filings were among the objections to the sale.[3]
- Other people's material. Client-confidential documents and data under NDAs often have to come out.
- How chat was exported. Slack's API terms bar apps offered outside your organization from using API data to train a large language model.[4] Buyers want native workspace exports.
- Unclear authority. If it's not obvious who can sign, buyers wait or walk.
How our estimate works
FileFunding's free estimate uses the same factors. It starts from a base range by peak headcount, then applies multipliers for years of history, the tools you used, volume and industry. The base ranges are our assumptions, anchored to price ranges that buyer programs and a buyer directory published on their own pages, as collected in October 2026 (not independently verified for this article):
| Peak headcount | Base range we use | Basis |
|---|---|---|
| 10–19 | $30K–$100K | a buyer program's published range |
| 20–49 | $200K–$500K | an indicative range cited in a buyer directory |
| 50–99 | $350K–$650K | a buyer's published deal examples |
| 250+ | up to about $2M | buyer programs advertising "$100K–$2M+" |
| 1–9 and 100–249 | in between | our interpolation unsourced estimate |
Treat any number from public information, ours included, as a reason to look closer rather than a price. These program ranges are higher than the per-company payouts reported for small startup wind-downs ($10,000 to $100,000[2]), which is one more reason to get several offers. A firmer number needs a look at real counts (mailboxes, messages, repositories, tickets), which we can do with read-only statistics, without copying any data.
How to land at the top of the range
- Keep everything until a deal closes. Don't cancel tools until you have full exports. See the shutdown data checklist.
- Know what you have. A rough count per system (mailboxes, messages, repos, tickets, call hours) turns a vague pitch into a priced one.
- Clean up rights early. Flag client-confidential and NDA material, and confirm who can sign.
- Get more than one offer. Buyers price differently. A single quote tells you little about the market.
Ready for a number? Enter your website and see an indicative range in about ten seconds. For the full process, read how to sell your company's data after a shutdown.
Frequently asked questions
Is there a standard price per email or per message?
No. Buyers don't publish per-unit prices for business archives, and they price on the whole package: who used the systems, for how long, and what the work was about. That's why two companies with similar volumes can get very different offers.
Why are estimates shown as ranges?
Because the real number depends on what's inside the data and how much of it can be cleared, which nobody knows until the archive is reviewed. An estimate from public information can tell you whether it's worth pursuing, not what you'll be paid.
Does a bigger archive always mean more money?
Not always. Variety and depth matter more than raw size. A buyer told MIT Technology Review that one company's Slack history, if it's rich enough, can be worth more than another's Slack, email and calendars combined.
What lowers the value?
Missing history (deleted accounts), heavy personal or regulated data, client-confidential content, chat pulled through an app's API instead of a native export, and unclear authority to sign. Each either shrinks what can be included or adds risk the buyer prices in.
Sources
- MIT Technology Review, "AI labs are buying data from dead startups", Oct 13, 2026 www.technologyreview.com/2026/10/13/1145755/ai-labs-buying-d…
- Gizmodo (summarizing Forbes), "Failed Companies Are Selling Old Slack Chats and Email Archives to Train AI", Apr 17, 2026 gizmodo.com/failed-companies-are-selling-old-slack-chats-and…
- Stretto Research Suite, "The Spirit Airlines deidentified data sale: auction results, objections, and the September 30 hearing", Sept 13, 2026 chapter11cases.com/blogs/news/the-spirit-airlines-deidentifi…
- Slack, API Terms of Service slack.com/terms-of-service/api
Figures marked unsourced estimate are our own judgment, not from a published source.
The FileFunding team
FileFunding works with companies that hold old business data and the AI labs that license it. Articles are researched from public sources, cited inline, and reviewed before publishing.