AI Auto-Promote Settings
Admin knobs for the rate auto-promote pipeline — the kill switch, confidence tiers, Vision OCR caps, and SLA/lifecycle thresholds.
AI Auto-Promote Settings (/admin/tenant-settings/ai-auto-promote) is the admin screen to "Tune
the auto-promote pipeline thresholds, SLAs, vision confidence shift, and hard caps" — it governs when
an AI-extracted rate quote is promoted automatically instead of routed to manual review. Change a
value and choose Save changes (Reset re-loads the saved values).
Kill switch
Auto-promote enabled is the master toggle: "When off, every extracted RateQuote routes to manual review."
Confidence tiers
Three thresholds decide the landing state of an extracted quote, and must stay in descending order:
- ACTIVE tier — auto-promote straight to ACTIVE.
- PENDING tier — send to PENDING_APPROVAL.
- DRAFT tier — keep as DRAFT only.
Vision OCR
For quotes read from photos or images, Vision confidence shift is subtracted from the raw confidence (default -0.05), and Daily and Monthly Vision call caps bound how many image extractions run.
SLA and lifecycle
These keep auto-promoted cards honest over time:
- Verify SLA hours — raises a warning when an ACTIVE auto-card stays unverified too long.
- Auto-flip hours and Auto-flip max per run — after the window, an unverified ACTIVE card flips back to PENDING_APPROVAL, in bounded batches.
- Unverified hard cap — auto-promote switches off entirely when the count of ACTIVE-unverified cards exceeds this.