Post Analysis

Post Analysis automatically processes conversations after they end, extracting structured insights, summaries, and metrics. This powerful feature turns raw conversations into actionable business data without manual review.

What is Post Analysis?

When a conversation ends, Post Analysis runs a separate AI analysis pass that can:

Capability	Example Output
Summarize conversations	"Customer inquired about billing, resolved by explaining payment due date"
Extract structured data	`{ "topic": "billing", "resolved": true, "sentiment": "satisfied" }`
Score interactions	Customer satisfaction: 4/5, Agent performance: 92%
Classify intents	Primary: "billing_inquiry", Secondary: "plan_upgrade_interest"
Flag follow-ups	"Customer requested callback about enterprise pricing"

Additional Token Cost

Post Analysis runs a separate LLM call after each conversation, which incurs additional token costs. The cost depends on conversation length and analysis complexity.

How It Works

Conversation Ends
       ↓
Post Analysis Triggered
       ↓
┌─────────────────────────────────────┐
│  LLM analyzes full conversation:    │
│  - Applies your analysis prompts    │
│  - Extracts structured data         │
│  - Generates summaries              │
└─────────────────────────────────────┘
       ↓
Results stored and/or sent via webhook

Trigger timing:

Voice calls: When call ends
Chat: When thread is marked complete or times out
API threads: When explicitly closed

Configuration

Enabling Post Analysis

Navigate to Agent Configuration > Advanced
Find Post Analysis section
Toggle Enable Post Analysis
Configure your analysis schemas

Analysis Schema

Define what data to extract using JSON Schema:

json

{
  "type": "object",
  "properties": {
    "summary": {
      "type": "string",
      "description": "2-3 sentence summary of the conversation"
    },
    "topic": {
      "type": "string",
      "enum": ["billing", "technical_support", "sales", "general", "complaint"],
      "description": "Primary topic of the conversation"
    },
    "resolved": {
      "type": "boolean",
      "description": "Whether the customer's issue was resolved"
    },
    "sentiment": {
      "type": "string",
      "enum": ["positive", "neutral", "negative"],
      "description": "Customer's overall sentiment"
    },
    "follow_up_needed": {
      "type": "boolean",
      "description": "Whether follow-up action is required"
    },
    "follow_up_notes": {
      "type": "string",
      "description": "Details about required follow-up, if any"
    }
  },
  "required": ["summary", "topic", "resolved", "sentiment"]
}

Analysis Prompt

Customize instructions for the analysis:

markdown

Analyze this customer support conversation and extract the following:

1. **Summary**: Provide a brief, factual summary of what was discussed and the outcome.

2. **Topic Classification**: Categorize into one of these topics:
   - billing: Payment, invoices, charges, refunds
   - technical_support: Product issues, bugs, how-to questions
   - sales: Pricing, plans, upgrades, new purchases
   - general: General inquiries, account management
   - complaint: Dissatisfaction, escalation requests

3. **Resolution**: Was the customer's primary question/issue resolved in this conversation?

4. **Sentiment**: Based on the customer's language and responses, gauge their overall sentiment.

5. **Follow-up**: Note if any action items or callbacks were promised.

Be objective and base your analysis on the actual conversation content.

Example Schemas

Customer Support Analysis

json

{
  "type": "object",
  "properties": {
    "summary": {
      "type": "string",
      "description": "Concise summary of the support interaction"
    },
    "issue_category": {
      "type": "string",
      "enum": ["account_access", "billing", "product_bug", "feature_request", "how_to", "other"]
    },
    "resolution_status": {
      "type": "string",
      "enum": ["resolved", "escalated", "pending", "unresolved"]
    },
    "customer_effort_score": {
      "type": "integer",
      "minimum": 1,
      "maximum": 5,
      "description": "Estimated effort customer had to exert (1=easy, 5=difficult)"
    },
    "escalation_reason": {
      "type": "string",
      "description": "If escalated, explain why"
    }
  }
}

Sales Qualification Analysis

json

{
  "type": "object",
  "properties": {
    "lead_quality": {
      "type": "string",
      "enum": ["hot", "warm", "cold", "unqualified"],
      "description": "Lead qualification score"
    },
    "budget_mentioned": {
      "type": "boolean"
    },
    "budget_range": {
      "type": "string",
      "description": "Estimated budget if mentioned"
    },
    "timeline": {
      "type": "string",
      "description": "Purchase timeline if discussed"
    },
    "decision_maker": {
      "type": "boolean",
      "description": "Is the contact a decision maker?"
    },
    "competitor_mentioned": {
      "type": "array",
      "items": { "type": "string" },
      "description": "Any competitors mentioned"
    },
    "next_step": {
      "type": "string",
      "description": "Recommended next action"
    }
  }
}

Quality Assurance Analysis

json

{
  "type": "object",
  "properties": {
    "greeting_proper": {
      "type": "boolean",
      "description": "Did the agent properly greet the customer?"
    },
    "identification_verified": {
      "type": "boolean",
      "description": "Was customer identity verified when required?"
    },
    "accurate_information": {
      "type": "boolean",
      "description": "Was information provided accurate?"
    },
    "professional_tone": {
      "type": "boolean",
      "description": "Did agent maintain professional tone throughout?"
    },
    "closing_proper": {
      "type": "boolean",
      "description": "Did agent properly close the conversation?"
    },
    "compliance_issues": {
      "type": "array",
      "items": { "type": "string" },
      "description": "Any compliance concerns noted"
    },
    "overall_score": {
      "type": "integer",
      "minimum": 0,
      "maximum": 100
    }
  }
}

Accessing Analysis Results

In the Platform

Navigate to Conversations or Thread History
Select a completed conversation
View the Analysis tab to see extracted data

Via Webhook

Configure a webhook to receive analysis results automatically:

json

{
  "event": "conversation.analyzed",
  "conversation_id": "conv_abc123",
  "agent_id": "agt_xyz",
  "timestamp": "2025-01-18T15:30:00Z",
  "analysis": {
    "summary": "Customer called about a billing discrepancy...",
    "topic": "billing",
    "resolved": true,
    "sentiment": "positive",
    "follow_up_needed": false
  }
}

See Webhooks for configuration details.

Via API

Retrieve analysis for a specific conversation:

bash

GET /api/conversations/{conversation_id}/analysis
Authorization: Bearer {api_key}

Best Practices

1. Keep Schemas Focused

Don't try to extract everything. Focus on data you'll actually use:

❌ Over-engineered:

json

{
  "properties": {
    "summary": {},
    "detailed_summary": {},
    "brief_summary": {},
    "executive_summary": {},
    // ... 30 more fields
  }
}

✅ Focused:

json

{
  "properties": {
    "summary": {},
    "topic": {},
    "resolved": {},
    "follow_up": {}
  }
}

2. Use Enums for Consistency

Structured values make reporting easier:

json

{
  "sentiment": {
    "type": "string",
    "enum": ["positive", "neutral", "negative"]
  }
}

This ensures you get consistent values rather than variations like "happy", "satisfied", "good", "positive".

3. Write Clear Descriptions

The description field guides the analysis:

json

{
  "customer_effort_score": {
    "type": "integer",
    "minimum": 1,
    "maximum": 5,
    "description": "Rate how much effort the customer had to exert: 1=issue resolved immediately, 5=customer had to repeat themselves multiple times or got transferred"
  }
}

4. Test with Real Conversations

Before deploying, test your schema against actual conversation transcripts to verify:

Fields are extractable from typical conversations
Categories/enums cover common cases
Analysis prompt produces accurate results

5. Consider Cost vs. Value

Post Analysis uses additional tokens. Balance:

Analyze all conversations for high-value insights
Sample conversations for quality monitoring
Skip analysis for very short interactions

Use Cases

Customer Success Dashboard

Track resolution rates, sentiment trends, and common issues:

Metric	Source Field
Resolution Rate	`resolved` (boolean)
Sentiment Trend	`sentiment` (enum)
Top Issues	`topic` (enum) count
Follow-up Queue	`follow_up_needed` filter

Sales Pipeline Integration

Push qualified leads to your CRM:

javascript

// Webhook handler example
if (analysis.lead_quality === 'hot' && analysis.decision_maker) {
  await crm.createOpportunity({
    contact: conversation.customer,
    source: 'AI Agent',
    budget: analysis.budget_range,
    timeline: analysis.timeline,
    notes: analysis.summary
  });
}

Quality Monitoring

Automatically flag conversations for review:

javascript

// Alert on low quality scores
if (analysis.overall_score < 70 || analysis.compliance_issues.length > 0) {
  await alertTeam('QA Review Needed', conversation.id);
}

Troubleshooting

Analysis Not Running

Verify Post Analysis is enabled
Check that conversations are being properly closed
Review schema for syntax errors

Incorrect Extractions

Improve description fields in schema
Add examples to your analysis prompt
Ensure enum values cover all cases
Test with representative conversation samples

High Costs

Reduce schema complexity
Use shorter analysis prompts
Consider sampling instead of analyzing all conversations
Use smaller models for analysis if available

Webhooks - Receive analysis results automatically
Structured Analysis - Broader structured output capabilities
Agent Configuration - Complete agent setup

Agents

Configuration

Tools

Advanced

Deploying Agents

Settings

Post Analysis

What is Post Analysis?

How It Works

Configuration

Enabling Post Analysis

Analysis Schema

Analysis Prompt

Example Schemas

Customer Support Analysis

Sales Qualification Analysis

Quality Assurance Analysis

Accessing Analysis Results

In the Platform

Via Webhook

Via API

Best Practices

1. Keep Schemas Focused

2. Use Enums for Consistency

3. Write Clear Descriptions

4. Test with Real Conversations

5. Consider Cost vs. Value

Use Cases

Customer Success Dashboard

Sales Pipeline Integration

Quality Monitoring

Troubleshooting

Analysis Not Running

Incorrect Extractions

High Costs

Configuration

Tools

Advanced

Deploying Agents

Post Analysis ​

What is Post Analysis? ​

How It Works ​

Configuration ​

Enabling Post Analysis ​

Analysis Schema ​

Analysis Prompt ​

Example Schemas ​

Customer Support Analysis ​

Sales Qualification Analysis ​

Quality Assurance Analysis ​

Accessing Analysis Results ​

In the Platform ​

Via Webhook ​

Via API ​

Best Practices ​

1. Keep Schemas Focused ​

2. Use Enums for Consistency ​

3. Write Clear Descriptions ​

4. Test with Real Conversations ​

5. Consider Cost vs. Value ​

Use Cases ​

Customer Success Dashboard ​

Sales Pipeline Integration ​

Quality Monitoring ​

Troubleshooting ​

Analysis Not Running ​

Incorrect Extractions ​

High Costs ​

Related Documentation ​

Post Analysis

What is Post Analysis?

How It Works

Configuration

Enabling Post Analysis

Analysis Schema

Analysis Prompt

Example Schemas

Customer Support Analysis

Sales Qualification Analysis

Quality Assurance Analysis

Accessing Analysis Results

In the Platform

Via Webhook

Via API

Best Practices

1. Keep Schemas Focused

2. Use Enums for Consistency

3. Write Clear Descriptions

4. Test with Real Conversations

5. Consider Cost vs. Value

Use Cases

Customer Success Dashboard

Sales Pipeline Integration

Quality Monitoring

Troubleshooting

Analysis Not Running

Incorrect Extractions

High Costs

Related Documentation