Clari Copilot Conversation Intelligence Extraction
Overview
Treat Copilot as a separate data product rather than an extension of Revenue API exports. Build a metadata-first extraction that minimizes sensitive content, joins stable identifiers, and records the exact window and pagination boundary.
Prerequisites
- Copilot API key/password pair stored in an approved secret manager
- Authorized time window, user population, and data-classification decision
- Destination controls suitable for call and participant data
Instructions
Step 1: Inventory the required resources
Choose only the calls, call details, users, topics, scorecards, and templates necessary for the approved use case.
Step 2: Freeze the extraction window
Record workspace, start and end time, pagination order, and inclusion rules so reruns are deterministic.
Step 3: Fetch metadata first
List users, topics, scorecard templates, and call summaries before requesting detailed call content.
Step 4: Retrieve bounded details
Request call details only for selected call IDs, honor both documented rate ceilings, and checkpoint pagination without duplicates.
Step 5: Normalize with privacy controls
Separate identities, call metadata, scoring, and sensitive content; tokenize joins and restrict transcript or recording access.
Step 6: Reconcile and publish
Compare requested and received IDs, record missing or changed calls, and publish only the authorized dataset with lineage.
Authentication
Send both X-Api-Key and X-Api-Password to the Copilot host over HTTPS. Do not substitute the Revenue apikey; redact both Copilot credentials and minimize logs containing call titles, participant data, transcripts, or recording links.
Tool Discipline
Use Read and Grep to inspect configuration, provider contracts, fixtures, logs, schemas, and existing tests before proposing a change. Use Write or Edit only for the approved plan, implementation, test, or redacted receipt; do not issue, rotate, revoke, create, update, cancel, delete, export, ingest, or publish provider data without explicit operator approval.
Output
- Windowed extraction manifest and pagination checkpoints
- Governed call, user, topic, and scorecard datasets
- Reconciliation, privacy-classification, and publication receipt
Return the exact surface, environment, resource or job identifiers, contract fingerprint, evidence, unresolved risks, and final decision without exposing credentials or sensitive customer data.
Examples
A coaching dataset lists calls for one approved week, selects only calls for an authorized team, retrieves details under the documented limits, tokenizes participant joins, and records missing IDs before publication.
Error Handling
| Failure | Response |
|---|---|
| One credential header is missing | Stop and validate the Copilot credential pair; do not fall back to a Revenue token. |
| Rate ceiling is reached | Honor retry timing, checkpoint the cursor, and resume without widening concurrency. |
| Detailed content exceeds approval | Publish metadata only and quarantine or omit transcripts and recording references. |