Pricing an agency archive: published ranges against reported deals

Every founder wants one number. The market does not offer one, and anyone who gives you a confident figure for your archive before a buyer has seen a sample is guessing or selling something. What you can do is put the two kinds of evidence side by side and see where an agency plausibly sits.

The published ranges

These are the buyers’ own words, read on 7 November 2026.

Two things stand out. The floors cluster around $50K to $100K, and the ceilings are open-ended. A range that runs from $100K to “$3M+” tells you the buyer’s tiers, not your price.

The reported deals

Reported figures are rarer, because most licenses are private. The ones outlets have confirmed:

• A crypto founder closing his startup took home around $25,000 for its code, according to MIT Technology Review (October 2026).
• HVAC contractors were being paid about $150,000 each for their records by labeling startups, according to The Information (2026).
• Google offered $10 million for Spirit Airlines’ business records and Mercor $7.5 million, according to Reuters (August 2026); no judge had yet approved it.

At the corporate end, publishers have disclosed their totals: Wiley’s AI licensing income was $40 million in fiscal 2025 and $23 million in fiscal 2024. Those are catalogs of books and journals, licensed to several developers, and tell you nothing about a 50-person studio.

Where an agency sits

None of the reported deals is an agency. The nearest comparison is the HVAC report: an operating business with years of structured job records, roughly the kind of seller micro1 and Nyne describe. That sits at the bottom of their published ranges, not the middle.

The factors that move an agency up or down are the ones the buyers mention themselves: years of records, how structured they are (Scale AI wants mature systems of record), whether they are mainly in English, and how much is cleanly yours after the contract review. Then there is exclusivity. Prism starts every license as exclusive, and a grant that locks your records up deserves a higher price than one leaving you free to sell again.

Payment terms are part of the price

Two offers with the same headline can be worth very different amounts. At micro1, the whole sum is due inside 30 days once your data is accepted. Labelbox Prism splits it into a signing fee and a later revenue share, and says money can arrive less than three weeks after the opening call. Nyne fixes scope, rights and acceptance terms before work begins but gives no timing. Scale AI is silent on when it pays.

Acceptance is the word to watch. If payment depends on the buyer accepting the delivered data, the acceptance criteria decide when, and whether, you are paid in full. Ask for them in writing before you start preparing records, and ask what happens to a delivery that is partly rejected.

A revenue share can be worth more than a larger upfront fee, or nothing at all; it depends on reporting rights you will rarely have. Value the upfront part on its own.

Questions to put to any buyer about price

• Is the figure per company, per dataset or per year of records?
• Is the grant exclusive? If so, until when?
• What portion is paid at signing, and what depends on acceptance?
• Does the price assume a renewal or ongoing delivery?

How to use the ranges without fooling yourself

• Use the floor, not the ceiling, to decide whether a conversation is worth having.
• Ask what the range assumes: how many years, which systems, exclusive or not.
• Compare like with like: a per-company license against a per-company license, not against a publisher’s annual revenue.
• Remember Scale AI’s own label: illustrative. A range is an advertisement.

Treat any estimate as something short of an offer. Signed licenses frequently land well below the advertised bands, so never budget from the top of one.

Try it on your own numbers

The worth estimate filters the published ranges by your headcount, years trading and country, and shows which buyers’ rules you meet. It adds nothing of its own. For everything around the price (ownership, records, telling clients), start with the front page on how to sell agency data.

The right price for an archive is the one a buyer offers after reading a real sample, and that number is worth waiting for.