
You can sell Box data, de-identified, to AI builders who need real business documents to train models. Years of contracts, SOPs, proposals and version history show how your company actually works, and that operational record has value right now.
Box tends to hold the formal side of a business: the signed agreement, the approved template, the final deck. That makes it one of the cleaner document sources a machine learning team can license, provided the personal details are stripped out first.
What lives in a typical Box account
Most companies adopt Box because they need governed, shareable storage. Over the years that produces a layered archive of how the organization operates. The useful material for model training is less about any single file and more about the patterns across thousands of them.
- Document types and templates: proposals, statements of work, quotes, onboarding packets, policy manuals and checklists.
- Folder structure: how teams organize clients, projects, departments and fiscal years.
- Version history: how a draft changes from first pass to approved final, and how many rounds it takes.
- Metadata and activity: file types, sizes, created and modified dates, upload and preview counts, collaboration roles.
- Box Notes and comments: short-form review notes that show how people critique and approve work.
Taken together, this is a record of operational data that rarely leaves a company's walls.
Why AI and ML teams want Box archives
Model builders have already consumed most of the public web. What they lack is the private, everyday paperwork of business: the real SOP for closing a job, the actual escalation memo, the proposal that won versus the one that lost. Box archives fill that gap.
Buyers use document corpora like yours to:
- train agents that draft contracts, proposals and reports in a believable house style;
- teach retrieval systems how real folders are organized so they find the right file;
- model document lifecycles, from draft to review to approval, for workflow automation;
- build classifiers that sort and route incoming paperwork by type and urgency.
Founder Jeff Cline puts it bluntly: "Every industry is a geek away from being uberized." The firms whose documents teach those systems are the ones in a position to be paid for it.
What is never included, and how de-identification works
We never take HIPAA/PHI data or personal customer data. Everything is de-identified before it leaves your control.
For a Box library, that means names, signatures, addresses, phone numbers, account numbers, tax IDs and similar identifiers are removed or replaced with consistent placeholders. Folders that hold HR files, medical records or customer personal data are excluded from scope entirely, not just scrubbed. You decide which top-level folders are in and which are out before anything is processed.
The goal is to keep the structure and the language of the work while removing who it was about. A buyer learns how a services agreement is drafted and revised, not who signed it. If you have contractual confidentiality obligations to clients, review those with your attorney before scoping; our de-identified data guide walks through the common questions.
How to find your Box numbers
You only need two figures to start: total storage and the age of your oldest content. Box shows these in one place.
Where to look: Box admin console > Reports
Run a storage or user activity report and note the total GB in use. Then sort a top-level folder by date to find the earliest files. A rough count of active seats helps too, since more contributors usually means a richer variety of documents. If you are not sure, the valuation calculator accepts ranges.
What a Box header sample looks like
A header sample is a list of field names only. No file contents, no records, nothing that identifies a person. For Box, it usually describes the file inventory rather than the documents themselves. A typical set of column headers might be:
- file_id
- file_extension
- folder_path_depth
- parent_folder_category
- size_bytes
- created_at
- modified_at
- version_count
- collaborator_count
- comment_count
- shared_link_type
- retention_policy_flag
That list is enough for buyers to judge breadth, age and structure. Our header sample how-to shows how to export one in a few minutes.
What makes one Box library worth more than another
Value comes from depth and distinctiveness, not raw gigabytes. A terabyte of duplicate images is worth far less than a few hundred GB of well-organized working documents.
- Years of history: a long archive shows how processes evolved, which models cannot learn from a snapshot.
- Document diversity: contracts, SOPs, estimates and reports together beat one repeated form.
- Version depth: files with many revisions teach editing and approval behavior.
- Seats and departments: more authors means more varied writing and workflows.
- Industry: specialized fields such as construction or regulated professional services are scarcer, so buyers pay attention.
According to founder Jeff Cline, small companies have gotten over $100,000 for their data, and offers arrive from multiple buyers so you can see the highest bid. Results vary; not a guarantee.
From Box export to signed license in four steps
- Describe your archive. Tell us your storage size, years of history, seat count and industry through the free bid request.
- Share column names. Send a header sample of your file inventory, field names only.
- Compare offers. Buyers review the sample and bid, and you see the competing numbers side by side.
- Clean, license, deliver. Data is de-identified inside your environment, licensed under a written agreement, delivered, and you are paid.
In Jeff Cline's experience, bids come back within 72 hours when the data is correct and a header sample is provided, and deals typically close in about 90 days. Results vary; not a guarantee. See how it works for the full timeline.
Why waiting costs you
The window for selling document data is not permanent. Once the large models have absorbed enough business paperwork from enough sources, the next archive adds less, and prices fall. Unique data is king today because few companies have offered theirs yet.
If your Box account holds a decade of real work, the strongest time to find out what it is worth is before the market fills up. Start with a no-obligation valuation.
How to request a bid for your Box data
- Pull your storage report. Open the Box admin console, go to Reports, and note total storage used and your seat count.
- Find your oldest content. Sort a top-level folder by created date to see how many years of history you hold.
- Submit the bid form. Enter tool, years, seats, storage and industry on the get-a-bid page.
- Send a header sample. Export the column names of your file inventory, with no records or file contents.
- Review competing bids. Compare buyer offers, choose one, and complete de-identification and licensing before delivery.
Frequently asked questions
Can I sell my company's Box data?
Yes, in most cases you can license de-identified Box content you own to AI builders. You choose which folders are in scope, personal and sensitive material is removed or excluded, and the license spells out how buyers may use it. Check client contracts with your attorney first.
Do buyers get our actual documents?
They receive de-identified documents and metadata, never raw files with names or identifiers. We never take HIPAA/PHI data or personal customer data. Everything is de-identified before it leaves your control.
How much storage do I need before it is worth selling?
There is no fixed minimum. A smaller archive with many years of varied working documents can be more interesting than a large one full of media files, so share your numbers and let buyers decide.
Will this disrupt how my team uses Box?
No. The export is read-only and runs on a copy, so files, permissions and sharing stay exactly as they are for your users.
What should I send first?
Send your total GB, oldest file date and seat count through the bid form, then a header sample of column names only. That is enough for buyers to respond.
How fast can I get a number?
According to founder Jeff Cline, bids typically come back within 72 hours once the data is correct and a header sample is provided. Results vary; not a guarantee.
Updated November 19, 2026.
OpenDataPolicies.org connects businesses with buyers of de-identified operational data. We are not a law firm, broker-dealer or investment adviser, and nothing on this site is legal, tax or financial advice. Statements about amounts, timing and outcomes are Jeff Cline's statements based on past transactions. Results vary; nothing here is a guarantee of any bid, price, buyer or closing. Every deal depends on the data, its quality, its rights and buyer demand at the time.