Operating ProcedureExecution layer

Client Document Intake Gate (Onboarding)

A checklist with 7 steps: Inventory every source system the client will feed into retrieval before any connector is authorized.

By InnovaAI ResearchPublished

What are the steps?

checklist

Client Document Intake Gate (Onboarding)

  1. 01

    Inventory every source system the client will feed into retrieval before any connector is authorized

    List Google Drive folders, Notion workspaces, Slack channels, and shared inboxes by name and owner. A managed context engine API such as Ragie can connect to these directly, but the agency still owns the access map.

  2. 02

    Classify each source as public, client-confidential, or regulated before ingestion begins

    Regulated material (loan files, health records, legal exhibits) changes which retrieval architecture is acceptable. A rule-grounded system like ai·rete·rag exists precisely because some decisions need a deterministic audit trail rather than a similarity score.

  3. 03

    Confirm file-type coverage against the client's actual document mix

    Count PDFs, scanned images, audio recordings, and video assets separately. Multimodal parsing is not uniform across vendors, and a gap discovered in week six costs more than a gap discovered at intake.

  4. 04

    Set a chunking and metadata convention the client's team can maintain without agency help

    Agree on folder naming, document titles, and effective-date fields. Retrieval quality degrades quietly when metadata drifts, and the client will add documents after the retainer kickoff.

  5. 05

    Run a 20-question retrieval smoke test using real client queries, not synthetic ones

    Pull questions from the client's support inbox or sales calls. Score each answer for source accuracy and note which documents the system failed to surface.

  6. 06

    Document the connector credentials owner and rotation schedule in the client handoff file

    Name the individual at the client who controls each OAuth grant. When that person leaves, retrieval breaks and nobody knows why.

  7. 07

    Record the baseline retrieval accuracy score as the contract reference point

    Store the smoke-test result in the account file. Every later accuracy conversation, including a provider swap, gets measured against this number.