AI Policy
Repertoire Review uses AI to speed up catalogue work that would otherwise be manual reading and reconciliation. This page sets out exactly where it is used, what data is sent, and the review step that keeps a person accountable for every change.
Principles
- Grounded, not generative about facts: assistant answers and briefs are built from your workspace records, not from open-web guesses.
- Human in the loop: AI proposes, a person with a write role decides. No AI output edits your repertoire without confirmation.
- Traceable: accepted suggestions and applied rights changes are written to the audit trail with the field, old value and new value.
- Confidential: your documents and queries are sent to model providers on a no-training basis and are not used to improve public models.
- Bounded scope: AI is not used to decide access, roles or billing, and is never the sole basis for a legal or valuation conclusion.
Where AI is used
Reads uploaded agreements, statements and registration documents to suggest a document type, the writer or work it belongs to, and key terms such as territory, term dates, advances and royalty rates.
Review: Suggestions appear with a confidence score and are only written to your records when a user accepts them.
Drafts an agreement record and its royalty-rate table from a document, and reads amendments or letters of clarification to propose changes to live terms.
Review: You see the proposed before/after values and apply them explicitly; applied amendments are logged and revertible.
Suggests how spreadsheet columns map to repertoire fields, including multi-writer split roster columns, and flags likely duplicates.
Review: The mapping and detected writer rosters are shown for confirmation before any rows are committed.
Compares splits, identifiers, publishers and society data across records to surface conflicts and gaps.
Review: Output is advisory. It never edits records on its own.
Answers questions grounded in your own workspace data, and can turn an instruction such as a publisher change into a proposed update across matching works and agreements.
Review: Instructions always produce a preview first. Nothing is written until a user with a write role presses Apply, and every field change is recorded in the audit trail.
Summarises how sale-ready a transaction or catalogue looks based on completeness, diligence status and evidence.
Review: Narrative summary only; figures should be checked against the underlying records.
What is sent to model providers
- The text of the specific document, spreadsheet rows, records or question involved in the request you triggered.
- Structured context from your workspace needed to answer, such as work titles, writer names, splits and completeness status.
- No credentials, no other organisation's data, and no bulk export of your catalogue.
Known limitations
- Scanned or handwritten documents extract less reliably than digital text.
- Ambiguous writer or work names may match the wrong record — check the suggested link before accepting.
- Royalty rates expressed in unusual bases, side letters referenced but not attached, and non-English contracts may be partially captured.
- Percentages and totals should always be checked against the split sheet; the platform flags totals that do not reach 100%.
Turning AI off and reporting problems
AI features are triggered by user action, so you can simply not use them and enter records manually. If you would like AI features disabled at workspace level, or you want to report a bad extraction or an incorrect rights suggestion, contact support@repertoirereview.app. AI output is decision support and is not legal, accounting, valuation or tax advice.
AI improvement telemetry consent
When you first sign in you are asked to accept this policy and to choose your cookie preferences. One of the optional categories, AI feature improvement, is off by default. Leaving it off means the outcomes of AI classification and extraction in your workspace are not reviewed for accuracy improvement. Turning it on never permits your repertoire, documents or royalty data to be used to train third-party models. You can change this at any time from the account menu.