Price the permitted use, not the annotation count
A million labels is an inventory measure, not a valuation. An engineering archive with resolved failure causes, a practitioner-written answer set and a collection of generic boxes are different products. Begin by describing the decision the labels support and the gap a buyer could test with them.
The same corpus may be proposed for model training, specialist fine-tuning or held-out evaluation. Each use needs a defined licence. The comparison is not simply cost per label: it is useful coverage, rights and the effort required to make the asset usable in that particular setting.
What a public enterprise-data example actually tells you
On November 8, 2026, micro1’s official data-estimate page advertised $100K–$500K for a category called “Qualified Enterprise Data Partnerships.” This is an illustrative vendor marketing category for operational enterprise data—not a verified completed transaction or an observed market median. (micro1, enterprise data-estimate marketing page; checked November 8, 2026)
The page does not establish a price per annotation, the likelihood of acceptance or typical seller earnings. Broader advertised payout claims are not evidence of realized payments, and this category cannot be applied as a rate card for annotated datasets.
For a labelled-data owner, the useful lesson is to define the proposed partnership: a finite corpus, an ongoing stream or additional expert work. A marketing range cannot substitute for a scoped offer and a buyer’s validation.
Make replacement effort legible
Explain what recreating the labels would require, with your own records rather than assumed market rates. Keep source acquisition, specialist judgment, schema development, adjudication and delivery preparation separate. Past spending is evidence of your effort; it is not a floor a buyer must pay.
| Work behind the asset | Evidence to prepare |
|---|---|
| Specialist decisions | Contributor qualifications, task instructions and recorded effort |
| Resolved ambiguity | Disagreement cases, reviewer rationales and guideline changes |
| Unusual coverage | Class inventory, rare cases and known gaps |
| Reusable delivery | Schema, file formats, version history and validation checks |
Show why the labels can be trusted
Give a prospective buyer the review method, not a single headline accuracy figure. State how reference answers were established, what was sampled, which classes were checked and where errors remain. Dataset documentation frameworks support keeping composition, collection and intended use visible. (Gebru et al., arXiv:1803.09010)
An agreement statistic measures agreement under a particular task and sampling design. It does not prove rights, eliminate systematic mistakes or promise improvement to a buyer’s model. Describe the method and limitations beside the result. (K. Krippendorff, University of Pennsylvania (ScholarlyCommons))
Negotiate a defined licensing unit
- Scope: identify the corpus version or subset, label schema, size and delivery format.
- Use: distinguish training from evaluation, and define any field-of-use restrictions.
- Exclusivity: specify which uses, parties and duration would be restricted; consider the opportunities you give up.
- IP and consent: establish permissions for underlying content and contributor work before pricing access.
- Refresh: separate existing inventory from updates, new expert work and review obligations.
- Buyer fit: test whether unique coverage reduces a real gap; volume alone is not proof of utility.
Separate the corpus from the ongoing work
Consider a hypothetical offer comprising a fixed image corpus with specialist labels, plus quarterly annotation updates. Keep the existing-corpus licence, update delivery and any additional adjudication work as separate items. No dollar rate is implied: the example illustrates contract structure, not earnings.
This makes it easier to discuss what changes when a buyer asks for a new schema, more review or exclusivity. A larger commitment is a different offer—not merely the same labels at a different price.
Prepare a commercial evidence note
Use this to support a licensing conversation, not to manufacture a precise appraisal. Protect raw files and sensitive information while permissions and access terms are being agreed.
- One paragraph on the task and why your coverage is difficult to reproduce.
- A versioned inventory with content type, label type, size, format and refresh cadence.
- A concise account of source permissions, contributor rights and consent constraints.
- A quality note with review design, limitations and known failure cases.
- Your proposed use, exclusivity boundaries, delivery and update commitments.
Have an annotated dataset or specialist expertise?
Tell us what you own, how the labels were reviewed and which rights you can license. A short overview of your dataset or annotation business is a useful place to start.
Prefer to write directly? hello@callwavemedia.com