Skip to content
Heptora Docs

Smart Classification

Smart Classification combines artificial intelligence with business rules to automatically categorize content, make intelligent routing decisions, and handle complex workflows. Process documents, analyze sentiment, detect intent, and route work items dynamically—all without manual intervention.

Overview

Section titled “Overview”

Smart Classification transforms unstructured data into actionable intelligence through a powerful hybrid engine that combines:

  1. AI-Based Classification: Machine learning models for automatic content understanding
  2. Business Rules Engine: Traditional rule-based logic for business-specific requirements
  3. Dynamic Routing: Intelligent workflow direction based on classification results

AI-Based Classification

Section titled “AI-Based Classification”

Heptora’s classification engine automatically understands and categorizes various types of content without manual configuration.

Classification Capabilities

Section titled “Classification Capabilities”

Document Classification

Section titled “Document Classification”
  • Automatically identifies document types (invoices, contracts, receipts, certificates)
  • Distinguishes between similar document formats
  • Extracts document-specific metadata
  • Confidence scoring for validation and escalation

Content Categorization

Section titled “Content Categorization”
  • Classifies emails by topic, urgency, or department
  • Tags support tickets by category and issue type
  • Organizes forms and requests by purpose
  • Segments data for targeted processing

Sentiment Analysis

Section titled “Sentiment Analysis”
  • Evaluates tone in customer communications
  • Identifies dissatisfaction signals
  • Prioritizes urgent or negative feedback
  • Triggers appropriate escalation protocols

Intent Detection

Section titled “Intent Detection”
  • Understands what action is requested in messages
  • Recognizes customer needs from natural language
  • Extracts key intents from complex communications
  • Maps intents to appropriate workflows

Anomaly Detection

Section titled “Anomaly Detection”
  • Identifies unusual patterns in data
  • Flags suspicious transactions or requests
  • Detects fraud indicators
  • Highlights exceptions requiring manual review

Continuous Learning

Section titled “Continuous Learning”
  • Improves accuracy from user feedback
  • Adapts to organizational changes
  • Learns from historical classifications
  • Refines models over time

Business Rules Engine

Section titled “Business Rules Engine”

Define precise business logic with a powerful, intuitive rules engine that handles simple conditions to complex workflows.

Rule Operators

Section titled “Rule Operators”

Logical Operators:

  • AND, OR, NOT for combining conditions
  • Nested conditions for complex logic
  • Truth table evaluation
  • Short-circuit optimization

Comparison Operators:

  • Equal, not equal
  • Greater than, less than, between ranges
  • Contains, starts with, ends with (string patterns)
  • Matches regex patterns
  • In list, not in list (set operations)

Temporal Rules

Section titled “Temporal Rules”
  • Date-based: Conditions on specific dates or date ranges
  • Time-based: Execution windows (business hours, days of week)
  • Duration-based: Rules based on time elapsed
  • Recurrence: Periodic rule evaluation
  • Seasonal: Rules that vary by season or period

Rule Examples

Section titled “Rule Examples”
Rule: Route High-Value Invoices
IF amount > 10000 AND approval_status = "pending"
AND supplier IN (list_of_critical_suppliers)
AND date WITHIN last_7_days
THEN priority = "high" AND route_to = "director_queue"
Rule: Escalate Unresolved Tickets
IF status = "open" AND created_date + 24_hours < now()
AND sentiment = "negative"
THEN escalate_to = "supervisor"

Dynamic Routing

Section titled “Dynamic Routing”

Automatically direct work items to the right destination based on intelligent classification and rule evaluation.

Conditional Flows

Section titled “Conditional Flows”
  • Bifurcation: Route to different processes based on conditions
  • Multi-way splits: Direct to multiple paths based on content
  • Nested conditions: Complex decision trees with multiple levels
  • Default paths: Fallback routing for unclassified items

Prioritization

Section titled “Prioritization”
  • Dynamic priority assignment: Based on classification results
  • Queue-based: Different queues for different priorities
  • SLA management: Automatic escalation based on priority
  • Load balancing: Distribute across available capacity

Intelligent Assignment

Section titled “Intelligent Assignment”
  • Round-robin: Distribute evenly across team members
  • Skill-based: Route to agents with matching expertise
  • Availability-based: Assign to available resources
  • Performance-based: Route to highest performers in category

Escalation Rules

Section titled “Escalation Rules”
  • Time-based escalation: Escalate after X time in queue
  • Complexity-based: Route complex cases to specialists
  • Attempt-based: Escalate after failed attempts
  • Manual escalation: Override rules when needed

Use Cases

Section titled “Use Cases”

Invoice Processing

Section titled “Invoice Processing”

Automatically classify and route invoices:

  • Identify invoice type and supplier
  • Validate amounts and required fields
  • Route to appropriate approval queue
  • Escalate exceptions or anomalies
  • Track through completion

Results: 80% reduction in manual sorting, faster approvals

Customer Service

Section titled “Customer Service”

Intelligently manage support tickets:

  • Analyze sentiment to identify urgent cases
  • Detect customer intent (complaint, question, request)
  • Route by expertise required
  • Prioritize negative sentiment cases
  • Escalate based on severity and wait time

Results: 40% faster resolution, improved satisfaction

Document Validation

Section titled “Document Validation”

Ensure document quality:

  • Classify document type automatically
  • Verify required fields are present
  • Flag incomplete or suspicious documents
  • Route for human review if needed
  • Validate against business rules

Results: 60% fewer processing errors, higher compliance

Approval Workflows

Section titled “Approval Workflows”

Streamline approval processes:

  • Classify requests by type and amount
  • Route to appropriate approver level
  • Apply business rules (amount limits, department, etc.)
  • Escalate complex or edge cases
  • Track approval status

Results: Automated routing, consistent policies, audit trail

Exception Management

Section titled “Exception Management”

Handle edge cases intelligently:

  • Detect anomalies in data or transactions
  • Flag items requiring human judgment
  • Route exceptions to specialists
  • Apply special processing rules
  • Maintain audit trail

Results: Reduced fraud, improved control, transparent exceptions

Configuration

Section titled “Configuration”

Visual Flow Editor

Section titled “Visual Flow Editor”

Design classification and routing logic graphically:

  • Drag-and-drop nodes: Add conditions, routes, and actions
  • Visual connectors: Link decision paths intuitively
  • Real-time preview: See how data flows through rules
  • Collaborative editing: Multiple team members contribute

Condition Library

Section titled “Condition Library”

Pre-built conditions for common scenarios:

  • Document conditions: Type, format, completeness
  • Amount conditions: Value ranges, thresholds
  • Temporal conditions: Date, time, duration ranges
  • Text conditions: Keywords, patterns, sentiment
  • Custom conditions: Create organization-specific conditions

Simulation and Testing

Section titled “Simulation and Testing”

Validate rules before deployment:

  • Test with sample data: Run classifications on examples
  • Batch testing: Test multiple records at once
  • Coverage analysis: Verify all paths are covered
  • Performance testing: Check execution speed
  • Debug mode: Step through rules for troubleshooting

Versioning and Rollback

Section titled “Versioning and Rollback”

Manage rule changes safely:

  • Version history: Track all rule changes
  • Change descriptions: Document why rules changed
  • Rollback capability: Return to previous versions
  • Scheduled deployment: Deploy changes at specific times
  • Canary testing: Test rules on small percentage first

A/B Testing

Section titled “A/B Testing”

Compare different rule strategies:

  • Split classification: Route percentage to each variant
  • Performance comparison: Compare metrics between variants
  • Statistical significance: Verify results are meaningful
  • Winner selection: Automatically promote better variant
  • Gradual rollout: Increase percentage over time

Best Practices

Section titled “Best Practices”

Rule Design

Section titled “Rule Design”

  1. Start Simple

    • Begin with obvious rules
    • Add complexity gradually
    • Test each addition

  2. Be Specific

    • Define clear conditions
    • Avoid ambiguous logic
    • Use consistent naming

  3. Plan for Exceptions

    • Include default/fallback paths
    • Handle edge cases explicitly
    • Test boundary conditions

  4. Document Decisions

    • Explain why rules exist
    • Note business requirements
    • Track changes over time

Classification Quality

Section titled “Classification Quality”

  1. Training Data

    • Use representative samples
    • Include edge cases
    • Maintain class balance
    • Update periodically

  2. Feedback Loop

    • Review misclassifications
    • Correct incorrect results
    • Retrain models regularly
    • Monitor for drift

  3. Validation

    • Manually verify results
    • Spot-check classifications
    • Track accuracy metrics
    • Monitor confidence scores

  4. Escalation

    • Set appropriate thresholds
    • Route low-confidence items
    • Enable manual override
    • Learn from reviews

Performance

Section titled “Performance”

  1. Optimization

    • Order rules by frequency
    • Avoid redundant conditions
    • Cache common lookups
    • Monitor execution time

  2. Scalability

    • Test with large volumes
    • Plan for growth
    • Optimize expensive operations
    • Monitor resource usage

  3. Monitoring

    • Track classification rates
    • Monitor routing distribution
    • Alert on anomalies
    • Analyze trends

Frequently Asked Questions

Section titled “Frequently Asked Questions”

How accurate is the AI classification?

Section titled “How accurate is the AI classification?”

Accuracy depends on the data and classification type. Document classification typically achieves 90%+ accuracy. We recommend starting with AI suggestions and using human feedback to improve accuracy over time.

Can I combine AI classification with business rules?

Section titled “Can I combine AI classification with business rules?”

Yes, that’s the core strength of Smart Classification. Use AI for initial categorization, then apply business rules for routing and processing decisions.

What happens with low-confidence classifications?

Section titled “What happens with low-confidence classifications?”

Configure confidence thresholds in your rules. Low-confidence items can be routed to a human queue for verification, keeping high-confidence items in automated workflows.

Can I use multiple classifications for the same item?

Section titled “Can I use multiple classifications for the same item?”

Yes, an item can receive multiple classifications (e.g., document type AND sentiment). Use combined conditions in your rules.

How often should I update my rules?

Section titled “How often should I update my rules?”

Review rules quarterly initially, then as-needed. Update when business processes change, when you notice misclassifications, or when new scenarios emerge.

Can I A/B test different routing strategies?

Section titled “Can I A/B test different routing strategies?”

Yes, use the A/B testing feature to test different rule sets on a percentage of traffic before full deployment.

What data is stored for audit purposes?

Section titled “What data is stored for audit purposes?”

Classification results, routing decisions, and confidence scores are logged. Original content can be stored based on your retention policies.

Can I override automated classifications?

Section titled “Can I override automated classifications?”

Yes, users can manually override classifications at any point. Overrides are logged and can feed back into model retraining.

Support Section

Section titled “Support Section”

Need help with Smart Classification?

When contacting support, include:

  • Description of your classification use case
  • Sample data (sanitized if needed)
  • Current rule configuration
  • Expected vs. actual results
Section titled “Related Resources”