Preparing a dataset for licensing: a four-week work plan
Written for companies without a dedicated data team. Durations are indicative, not a promise.
Four weeks to a licensable version
- Inventory & rights Weeks 1–2: sources, agreements, exclusions.
- Technical freeze Week 3: version, hash, clean sample.
- Documentation Week 4: datasheet and provenance dossier.
Week 1 – Inventory
- List candidate datasets with an internal owner for each.
- Record source, volume, time span and content type.
- First pass on personal information: yes / no / unknown.
Week 2 – Rights review
- Pull customer, vendor and contractor agreements.
- Compare against what users were told at collection.
- Decide exclusions (sensitive categories, minors, biometrics).
Week 3 – Technical freeze
- Freeze a version and compute a hash.
- Deduplicate and remove corrupt records.
- Build a small sample free of personal information for demos.
Week 4 – Documentation
- Datasheet following Gebru et al.
- Provenance dossier.
- A one-paragraph, non-confidential description — the only thing you'd put in an inquiry form.
How this affects value
Preparation is the cheapest place to add value: a frozen, documented version is easier to scope and price.
See what shapes a dataset's value