Three steps, no mystery.

One side needs training data it can actually defend using. The other side owns it. fiund is the licensing layer between them — and the process is the same three steps every time.

01 · Tell us what you need

A buyer sends a brief, in plain terms: modality, languages and accents, recording conditions, volume, and the licence terms the use requires — training versus evaluation, exclusive or not, term and territory. Consent requirements too, if counsel has them. The brief doesn't have to be final; sourcing to a loose brief and tightening it together is normal.

Or you're on the other side: you own an archive. Send what you have — even a rough inventory of hours and formats — and we tell you honestly whether there's a market for it, before you spend a weekend organising files. The sell pages show what preparation looks like per asset type.

02 · We source and clear it

We go to the people who hold the material — creators, studios, archives, contact centres — not to crawls. Clearing means three specific things. First, a signed licence in which AI-training rights are explicit, not implied. Second, voice and likeness consent papered separately wherever a person is identifiable. Third, chain of title checked, so the party granting rights actually holds them.

Third-party material — music beds, stock inserts, footage the owner doesn't control — is carved out rather than waved through. Nothing moves until the paperwork exists. The full methodology is on how we source.

03 · Delivery and payout

The dataset arrives with its documentation attached: the licence, the consent records behind it, a provenance record covering source and ownership, and a manifest of the files delivered. That's what lets a buyer's legal team trace any file back to a signed grant.

The owner is paid under the contributor agreement whenever fiund licenses their material — how payouts work covers the supplier side. Completed transactions are documented so both sides can see exactly what was licensed and on what terms.