Second Pair of Eyes - Document Anomaly Detection

Every document your team processes is automatically reviewed by Muin’s Second Pair of Eyes — five specialized AI detectors that run in parallel to catch issues humans might miss.

What Gets Detected

Muin runs five types of analysis on every processed document:

1. Duplicate Detection

Catches documents that have been submitted before:

Exact duplicates — Identical content (e.g., same invoice uploaded twice)
Fuzzy duplicates — Nearly identical with minor differences (e.g., reformatted PDF)
Partial duplicates — Significant content overlap across documents

2. Amount Anomalies

Identifies unusual financial patterns:

Outliers — Amounts significantly above or below historical averages for a vendor
Mismatches — Line items that don’t add up to the stated total

3. Contract Mismatches

Compares documents against your contract terms:

Rates exceeding contracted prices
Quantities beyond agreed limits
Terms that don’t match the active agreement

4. Policy Violations

Checks against your organization’s rules:

Missing required approvals for high-value purchases
Spend category violations
Compliance requirement gaps

5. Cross-Reference Checks

Verifies consistency across related documents:

Invoice references to non-existent POs
Delivery receipt quantities vs. PO quantities
Payment terms mismatches between documents

How It Works

Document processed → 5 detectors run in parallel → Findings appear in Anomalies tab
                                                  → Dashboard widget updated
                                                  → Insight generated

Detection is fully automatic — it runs every time a document reaches “Ready” status. You can also trigger a manual re-analysis from the document’s Anomalies tab.

Reviewing Anomalies

Each finding includes:

Severity level — Alert (red), Warning (yellow), or Info (blue)
Confidence score — How certain the detector is (0-100%)
Impact value — Estimated financial impact when applicable
Detection type — Which detector found the issue

Resolving Findings

When you review an anomaly, you choose a resolution:

Resolution	When to use
Approved	The flagged item is acceptable
Rejected	The item should not be processed
False Positive	The detection was incorrect
Corrected	The issue has been fixed
Escalated	Needs someone else’s review
Deferred	Will review later

You can add a note with each resolution for audit trail purposes.

Dashboard Overview

The Second Pair of Eyes widget on your dashboard shows:

Total detections in the last 30 days
Pending review count
Resolved count
Alert-level findings
Resolution rate progress
Breakdown by detection type

Configuration

Admins can enable or disable individual detector groups from Insights → Settings:

Toggle entire detector groups on/off
Disable specific detection sub-types
Changes apply to future document processing (existing findings are preserved)

FAQ

Does this slow down document processing? No. Detection runs asynchronously after a document is processed. Your document is available immediately — anomaly results appear shortly after.

Can I re-run detection on old documents? Yes. Open the document, go to the Anomalies tab, and click “Run Analysis” to re-process.

Are detections tenant-isolated? Absolutely. All analysis is scoped to your organization. Your documents are never compared against other tenants’ data.

What happens if I mark something as a false positive? The finding is resolved and hidden from the default view. This data helps improve detection accuracy over time. You can still see resolved findings by toggling “Show resolved” in the Anomalies tab.