AI-Assisted Document Intake reads incoming faxed clinical documents, extracts structured data (member ID, diagnosis codes, requesting provider), and routes the request to the correct work queue. No manual data entry required for documents that meet the confidence threshold.
| Requirement | Details |
|---|---|
| Feature License | AI Document Intake license active on the tenant. Confirm in Admin > Licensing > Feature Entitlements. |
| Admin Role | System Configuration Administrator role or custom role with INTAKE_AI_CONFIG permission. |
| OCR Engine | Meridian OCR Engine 4.2+. Earlier versions do not support structured field extraction. Check in Admin > System > Component Versions. |
| Fax Integration | At least one active inbound fax channel in Admin > Integrations > Fax Channels. |
| Work Queues | Target queues must exist before routing rules can reference them. |
AI Intake is toggled per request type. This supports phased rollout: enable for prior auth first, add concurrent review later.
The configuration page displays with a list of available request types.
Status changes to Enabled (Pending Configuration). The feature is not active until you configure an intake job.
Enabled types show a green indicator.
Disabling AI Intake for a request type does not delete existing configuration. Documents already processed retain their extracted data.
Each enabled request type needs an intake job defining which fields to extract and where to route results.
The Intake Job Settings panel displays.
Rules evaluate top to bottom. First match wins. Documents matching no rule go to the default queue.
Required. The system will not process documents without a default queue set.
Status changes to Active. New documents on configured fax channels are now processed by the AI engine.
The engine assigns a confidence score (0 to 100) to each extracted value. Thresholds control what happens next.
The threshold page displays with sliders for each extraction field.
Scores below Review Required are rejected. The field is left blank and the document is flagged for full manual intake.
Applies to new documents only. Previously processed documents are not re-evaluated.
Start with Auto-Accept at 90+ during initial rollout. Lower it after reviewing accuracy data in Reports > AI Intake > Extraction Accuracy.
Cause: OCR Engine service unavailable or overloaded.
Fix: Go to Admin > System > Services. Confirm OCR Engine shows Running. If stopped, restart it. If Running but documents stuck 15+ minutes, check logs at Admin > System > Logs > OCR Engine for memory or timeout errors.
Cause: Member ID format mismatch. System expects one format, incoming documents use another.
Fix: Go to Admin > Intake > AI Document Intake > Field Mapping. Verify the Member ID regex matches what providers actually submit. Common patterns:
^\d{10}$ : 10-digit numeric^[A-Z]{2}\d{10}$ : 2-letter prefix + 10 digits^\d{3}-\d{7}-\d{2}$ : hyphenatedCause: Rule order. First match wins. A broad rule above a specific rule captures everything.
Fix: Open the request type in Admin > Intake > AI Document Intake. Drag rules so specific conditions appear first. Use Test Routing to simulate a document and confirm the result.
Cause: Engine extracted free-text descriptions instead of codes. Source document likely does not contain printed ICD-10 values.
Fix: Enable Diagnosis Code Lookup in Field Mapping > Primary Diagnosis Code. The system will match extracted text against the ICD-10 code set and return the best match. If no match meets threshold, the field goes to manual review.