Review Panel

The Review Panel is the interface where reviewers examine requests, make decisions, and provide corrections. Each request type has a specialized review interface optimized for efficient decision-making.

Review Interface Layout

All review panels share a common structure:

• Left Side — Document preview with zoom, pan, and navigation
• Right Side — Review form with context and decision options
• Header — Request metadata, status, and timeout warnings
• Footer — Action buttons (approve, reject, escalate, save)

Document Preview

Full-featured document viewer:

Features:

• Multi-page navigation (thumbnails or arrows)
• Zoom controls (fit width, fit page, custom zoom)
• Pan and scroll for large documents
• Rotation for scanned documents
• Annotations and highlights

Toolbar:

• Zoom in/out buttons
• Fit to width/height
• Page navigation (previous, next, jump to page)
• Rotate clockwise/counterclockwise
• Download original document

Approval Review

For approval request types.

Review Context

What’s Being Approved:

• Workflow step description
• Previous processing results
• Business context and rationale

Approval Criteria:

• Required checks before approving
• Rejection reasons (if applicable)
• Business rules to verify

Decision Options

Common Approval Types

Financial Approvals:

• Verify transaction amount and recipient
• Check against approval limits and policies
• Confirm account balances or budgets
• Review supporting documentation

Contract Approvals:

• Verify contract terms match expectations
• Check for red flags or unusual clauses
• Confirm parties and signatures
• Review legal compliance

Process Approvals:

• Verify workflow execution was correct
• Check quality of automated processing
• Confirm exception handling
• Approve continuation to next phase

Correction Review

For correction request types (verifying and correcting AI predictions).

Prediction Display

Original Prediction:

• AI-predicted values with confidence scores
• Highlighted regions in document showing extraction
• Confidence breakdown per field

Correction Interface:

• Editable fields showing current predictions
• Side-by-side comparison (prediction vs document)
• Quick accept/reject buttons per field

Making Corrections

Field-Level Corrections

Text Fields:

• Click field to edit
• Type or paste correct value
• Validation shows format errors

Date Fields:

• Date picker for easy selection
• Manual entry if picker insufficient
• Format validation (MM/DD/YYYY, etc.)

Number Fields:

• Numeric input with validation
• Currency formatting for money fields
• Range validation if configured

Checkbox/Radio Fields:

• Toggle selections
• Multi-select support

Table Fields:

• Edit individual cells
• Add or remove rows
• Reorder columns

Correction Patterns

High Confidence, Correct:

• Quick review and accept all
• Fastest workflow (seconds per document)

High Confidence, Incorrect:

• Edge case or model failure
• Correct specific fields
• Flag for training data

Low Confidence, Correct:

• Model correctly uncertain
• Accept predictions
• May add to training data

Low Confidence, Incorrect:

• Expected failure case
• Correct as needed
• Definitely add to training data

Router Review

For router request types (choosing workflow path).

Routing Context

Available Paths:

• List of possible next steps
• Description of each path
• When to choose each option

Decision Context:

• Document content summary
• Previous processing results
• Routing criteria

Selecting Path

Common Routing Scenarios

Document Type Routing:

• Route invoices to accounts payable workflow
• Route contracts to legal review
• Route forms to data entry queue

Escalation Routing:

• Route complex cases to specialist team
• Route high-value items to management
• Route errors to investigation queue

Quality-Based Routing:

• Route high-quality extractions to automatic processing
• Route low-quality to enhanced review
• Route partial failures to correction queue

Annotation Tools

Enhance reviews with annotation capabilities:

Highlighting

• Select text — Click and drag to highlight
• Color coding — Different colors for field types
• Notes — Add comment to highlight
• Share — Highlights visible to other reviewers

Drawing Tools

• Bounding boxes — Draw rectangles around regions
• Arrows — Point to specific areas
• Text annotations — Add labels or notes
• Freehand — Draw custom shapes

Markup Use Cases

• Circle incorrect extractions
• Point to missed fields
• Highlight ambiguous text
• Mark damaged or unclear regions

Annotations are saved with review for audit and training purposes.

Review Actions

Primary Actions

Submit:

• Complete review with decision
• Workflow resumes or branches based on decision
• Request marked completed

Save Draft:

• Save progress without submitting
• Return to review later
• Request remains in review state

Escalate:

• Send to specialist or manager
• Add escalation reason and notes
• Request assigned to escalation queue

Reassign:

• Assign to different reviewer
• Add reassignment reason
• Request returns to pending state

Secondary Actions

Add to Training:

• Flag review for inclusion in training dataset
• Include corrected values as ground truth
• Improves model on similar cases

Report Issue:

• Flag problems with document or workflow
• Alert administrators
• Does not complete review

Request Context:

• Request additional information
• Pause review until context received
• Notify context provider

Keyboard Shortcuts

Efficient review with keyboard shortcuts:

Review Quality

Quality Checks

Built-in validation before submission:

• Required fields — All mandatory fields completed
• Format validation — Dates, numbers, emails formatted correctly
• Business rules — Custom validation rules passed
• Consistency — Cross-field validation (e.g., line items sum to total)

Errors shown inline with clear messaging.

Confidence Scoring

Optionally indicate your confidence in review:

• High confidence — Certain of decision/correction
• Medium confidence — Reasonably sure but some uncertainty
• Low confidence — Uncertain, may need specialist review

Low-confidence reviews may be flagged for secondary review.

Review Feedback

After submission, see:

• Agreement with AI — How often you agreed with predictions
• Correction frequency — Percentage of fields corrected
• Review time — How long review took
• Quality score — Based on validation and consistency

Use feedback to improve review speed and accuracy.

Multi-Reviewer Workflows

When multiple reviewers are required:

Unanimous Approval

• All reviewers must approve
• Each reviewer sees others’ decisions (after completing own)
• Last reviewer to approve completes request
• Any rejection blocks approval

Majority Vote

• Configurable threshold (e.g., 2 of 3 must approve)
• Reviewers can’t see others’ decisions until voting complete
• Request completes when threshold reached
• Ties route to tiebreaker reviewer

Sequential Review

• Reviewers assigned in order
• Each sees previous reviewer’s decision and notes
• Can approve, reject, or escalate
• Chain ends when decision made

Parallel Review with Reconciliation

• All reviewers complete independently
• Disagreements highlighted
• Reconciliation reviewer resolves conflicts

Audit Trail

All review actions are logged:

Logged Events:

• Who claimed request
• When review started/completed
• Decision made
• Corrections applied
• Escalations and reassignments
• Time spent on review

Audit Information Used For:

• Compliance and regulatory requirements
• Performance analytics
• Training data quality
• Dispute resolution

Access audit trail in request details after completion.

Review Best Practices

Speed and Accuracy

• Focus on high-value corrections — Don’t fix minor formatting
• Use keyboard shortcuts — 3x faster than mouse
• Batch similar requests — Reduce context switching
• Time-box complex reviews — Escalate if exceeding 5 minutes
• Trust high-confidence AI — Spot-check rather than re-verify everything

Quality

• Read full context — Understand why review requested
• Check against source — Verify predictions against document
• Be consistent — Apply same standards to all requests
• Document ambiguity — Add notes for unclear cases
• Flag training data — Mark good examples for model improvement

Collaboration

• Add helpful notes — Explain reasoning for next reviewer
• Escalate appropriately — Don’t guess on uncertain items
• Share knowledge — Document common patterns in notes
• Communicate blockers — Alert team to systemic issues

Next Steps

• Manage your queue in Review Queue
• Submit documents via Submissions
• Build HITL workflows in Workflows
• Monitor review metrics in Monitoring