Resources
How to sell egocentric data to AI companies
Prepare a first-person video dataset for buyer review: task coverage, activity hours, camera streams, annotations, permissions and a representative sample package.
Published 2026-10-27 · 6 min read
Sell egocentric data by establishing what the recordings contain, who can authorize the intended use and whether a buyer’s task requires those signals. Start with a concise inventory and representative sample review; an archive’s size alone does not establish demand or licensability.
Existing data and custom collection require different discussions. If files already exist, describe them accurately. If you plan to record, get the specification, permissions, pilot acceptance and commercial terms agreed before scaling.
Describe what actually exists
List unique activity hours, camera streams, scenarios, original formats, approximate capture regions and the labels or sensors available. State whether a stereo pair is calibrated and synchronized, rather than describing any two-camera setup as stereo data.
Mark estimated counts as estimates. Do not count the same activity multiple times because it appears in several views or exports. Preserve originals and explain which files are derivatives, previews or annotations.
Check source authority and restrictions
Identify the owner or authorized licensor and the records connecting files to the source. Review participant permissions, location access, confidential content, third-party works and any contractual restrictions for the intended use.
Keep raw identity and permission documents in a controlled review process. An anonymous public library description is not a rights-clearance certificate. Material with unresolved questions should not be offered as cleared inventory.
Make a representative evaluation package
Use an episode ID and manifest connecting original media, timestamps, annotations, calibration where applicable and quality notes. Select examples that show ordinary conditions and known limitations, not only the best-looking few seconds.
Agree sample access and evaluation scope before sharing. A sample is not automatically permission for production training. Identify measured and model-derived signals separately so buyers understand what they are evaluating.
Discuss acceptance and compensation together
The relevant unit may be an accepted activity hour, episode or defined dataset licence. Clarify included work, rejection reasons, preparation responsibilities and payment timing. A collector’s wage or another vendor’s catalog rate is not a guaranteed offer for your recordings.
Fiund’s standard terms do not authorize synthetic voice or likeness replication. Licensing authority, scope and compensation follow the applicable written agreement. Do not assume per-deal approval, exclusivity or recurring payouts that the agreement does not provide.
Start with a private assessment
The dataset assessment asks for a description before archive upload or payment details. Fiund reviews source quality, permissions and fit with relevant buyer needs. Additional samples or documentation are discussed when useful.
Submitting a dataset does not guarantee a sale. For collection teams without an existing library, use the capability assessment instead and describe devices, region, team capacity and the tasks or settings you can legitimately access.
Before the next step
- Create an honest inventory and identify estimates.
- Separate unique activity and camera-stream hours.
- Gather source and permission references.
- Agree evaluation scope before sharing samples.
Frequently asked questions
Does Fiund guarantee a purchase or paid assignment?
No. A description is reviewed for fit before a sample, collection scope or licence is agreed. Availability, acceptance, compensation and permitted uses follow the applicable written terms.