Analyze AI Performance

Analyze AI Performance #

Once your AI Agent is live and handling customer conversations, it's crucial to monitor its performance, identify improvement opportunities, and track key metrics. Support Unicorn provides comprehensive analytics to help you understand how your AI Agent is performing and where to focus your optimization efforts.

Conversation History #

View and search all conversations your AI Agent has handled.

Accessing Conversations #

  1. Navigate to your AI Agent's page
  2. Click the "Chats" tab
  3. See a chronological list of all conversations

Conversation List View #

Each conversation shows:

Status Indicators:
- Active: Ongoing conversation
- Waiting on Customer: Last message from agent/AI
- Waiting on Agent: Customer waiting for response
- Closed: Conversation ended

Key Information:
- Customer name or identifier
- Channel type (Live Chat, SMS, WhatsApp)
- First message preview
- Message count (customer vs agent)
- Duration
- Created/updated timestamps
- AI handoff status (if escalated)

Find specific conversations using filters:

By Status:
- Active conversations
- Closed conversations
- Escalated to humans
- AI-handled only

By Channel:
- Live Chat
- SMS
- WhatsApp
- All channels

By Date Range:
- Today
- Last 7 days
- Last 30 days
- Custom date range

By Customer:
- Search by email
- Search by phone number
- Search by name
- Search by customer ID

By Content:
- Keyword search in messages
- Search AI responses
- Search customer queries

Conversation Details #

Click any conversation to see complete details:

Message Thread:
- Full conversation history
- Timestamps for each message
- Sender identification (customer, AI, or human agent)
- Message content and formatting
- Rich media (images, attachments)

AI Performance Data:

For each AI response:
- Confidence Score: How confident the AI was (0-1)
- Response Time: How long it took to generate
- Source Attribution: Which knowledge base chunks were used
- Relevance Scores: How relevant each source was
- Token Usage: Prompt and completion tokens

Conversation Metadata:
- Session ID
- User agent (browser/device)
- IP address (if available)
- Custom attributes
- Tags and labels

Handoff Tracking:

If conversation was escalated:
- When escalation occurred
- Why it was escalated (confidence, sentiment, explicit request)
- Which human agent took over
- How long until human responded
- Resolution details

Conversation Actions #

From the conversation details page:

Close Conversation:
Mark resolved and close the conversation thread.

Reopen Conversation:
Reactivate a closed conversation if customer returns.

Disable/Enable AI:
Toggle AI responses on/off for this specific conversation.

Export Conversation:
Download full conversation history (JSON or PDF).

Add Tags:
Categorize for better organization and analytics.


Key Metrics Dashboard #

Track your AI Agent's performance with real-time metrics.

Overview Metrics #

Total Conversations:
- Total handled by AI
- Comparison to previous period
- Trend over time (line chart)

AI Automation Rate:
- Percentage of conversations handled entirely by AI
- No human intervention required
- Target: 60-80% for most use cases

Average Response Time:
- How fast AI responds to customer messages
- Typically 2-5 seconds
- Impacts customer satisfaction

Customer Satisfaction:
- Based on post-conversation ratings
- Scale of 1-5 stars
- CSAT (Customer Satisfaction Score)

AI Performance Metrics #

Confidence Score Distribution:
- Average confidence score across all responses
- Distribution chart (how many at each confidence level)
- Trend over time

High Confidence Rate:
- Percentage of responses with confidence > 0.7
- Indicates knowledge base coverage
- Target: 70%+ for well-trained agents

Escalation Rate:
- Percentage of conversations escalated to humans
- Why escalations occurred
- Target: 20-40% depending on AI mode

Knowledge Base Utilization:
- Which sources are used most frequently
- Which sources are never retrieved
- Gaps in knowledge base

Customer Engagement Metrics #

Message Volume:
- Total customer messages received
- Messages per conversation (avg)
- Peak hours and days

Conversation Length:
- Average number of message exchanges
- Duration of conversations
- Drop-off points

Return Customers:
- Customers with multiple conversations
- Frequency of repeat contacts
- Trends (increasing or decreasing)

Channel Distribution:
- Breakdown by channel (Live Chat, SMS, WhatsApp)
- Most popular channels
- Conversion rates per channel

Agent Efficiency Metrics #

First Contact Resolution (FCR):
- Conversations resolved without escalation
- Critical metric for AI success
- Target: 60%+ for common questions

Time to Resolution:
- How long from first message to resolution
- AI-only vs escalated conversations
- Benchmarks by topic/category

Credits Consumed:
- Total AI messages sent
- Credits used (1 per message default)
- Cost per conversation
- Trend over time


Handoff Analytics #

Understand when and why AI escalates to human agents.

Handoff Metrics #

Total Handoffs:
- Number of conversations escalated
- Percentage of all conversations
- Trend over time

Handoff Reasons:

Breakdown by escalation trigger:
- Low Confidence: AI uncertain about answer (X%)
- Customer Request: Explicitly asked for human (X%)
- Sentiment: Negative emotion detected (X%)
- Topic-Based: Predefined topic requiring human (X%)
- Timeout: No resolution after N exchanges (X%)

Time to Handoff:
- How long into conversation before escalation
- Average message count before handoff
- Patterns (immediate vs after attempts)

Human Response Time:
- How long until human agent responds after handoff
- SLA compliance
- By time of day/day of week

Handoff Success Rate #

Successful Handoffs:
- Human agent responded and resolved
- Customer satisfaction after handoff
- Resolution time

Failed Handoffs:
- No human response
- Customer abandoned before human reply
- Handoff to wrong team/person

Improving Handoff Flow #

Use handoff analytics to:

Reduce Unnecessary Escalations:
- Identify low-confidence topics → add to knowledge base
- Refine escalation thresholds
- Train AI with actual customer questions

Faster Human Response:
- Optimize routing rules
- Staff appropriately during peak times
- Use urgent notifications for high-priority handoffs

Better Context Transfer:
- Ensure humans see full conversation history
- Include customer metadata
- Provide AI's uncertainty reason


Content Performance #

Understand which knowledge base sources are working and which aren't.

Source Analytics #

Most Retrieved Sources:
- Documents or Q&As retrieved most frequently
- Indicates valuable content
- Ensure these are accurate and up-to-date

Underutilized Sources:
- Content that's never or rarely retrieved
- May be irrelevant, duplicate, or poorly formatted
- Consider removing or improving

Highest Confidence Sources:
- Which sources lead to high-confidence responses
- Model for creating new content
- Indicates good structure and clarity

Source Gaps:
- Common questions with no matching sources
- Low relevance scores across all sources
- Topics missing from knowledge base

Question Analysis #

Most Frequent Questions:
- What customers ask most often
- Ensure comprehensive answers in knowledge base
- Create dedicated Q&A pairs for top questions

Unanswered Questions:
- Questions AI couldn't answer confidently
- Explicitly said "I don't know"
- Priority for knowledge base expansion

Low Confidence Questions:
- Questions answered but with low confidence
- May need better content or clarification
- Review and improve

Question Categories:
Auto-categorization of questions by topic:
- Product information
- Billing and pricing
- Technical support
- Shipping and delivery
- Returns and refunds
- Account management

Content Recommendations #

Based on analytics, Support Unicorn suggests:

Add Content For:
- Top unanswered questions
- Frequent low-confidence responses
- Emerging topics from recent conversations

Update Content:
- Outdated information (detected from context)
- Sources with consistently low relevance
- Conflicting information across sources

Remove Content:
- Never-retrieved sources
- Duplicate content
- Irrelevant or outdated material


Customer Feedback #

Collect and analyze customer feedback on AI interactions.

Post-Conversation Surveys #

Automatic Survey Trigger:
After conversation closes, prompt customer:
- "How was your experience?" (1-5 stars)
- "Did the AI help solve your issue?" (Yes/No)
- "Any feedback?" (Optional text)

Survey Results:
- Overall satisfaction score
- Distribution of ratings
- Trends over time
- Correlation with AI confidence scores

Feedback Categories:

Analyze qualitative feedback:
- Positive: "Quick and helpful"
- Negative: "Couldn't understand my question"
- Suggestions: "Add information about X"

Thumbs Up/Down #

Allow customers to rate individual AI responses:

👍 Helpful responses:
- Mark in analytics
- Identify strong knowledge base areas
- Model for future responses

👎 Not helpful responses:
- Flag for review
- Identify gaps or inaccuracies
- Priority for improvement

Implicit Feedback #

Track behavioral signals:

Conversation Abandonment:
- Customer stops responding mid-conversation
- May indicate frustration or problem solved elsewhere
- Review abandoned conversation patterns

Repeat Questions:
- Customer rephrases same question multiple times
- AI response wasn't helpful or clear
- Knowledge base improvement needed

Escalation Requests:
- "Can I talk to a person?"
- Indicates AI couldn't meet need
- Review what led to request


Reports and Exports #

Generate custom reports and export data for deeper analysis.

Pre-Built Reports #

Daily Summary:
- Conversations handled
- AI automation rate
- Escalations
- Customer satisfaction
- Sent via email at 9 AM daily

Weekly Performance Report:
- Week-over-week comparisons
- Trending metrics
- Top questions
- Recommendations
- Sent via email every Monday

Monthly Business Review:
- Comprehensive monthly analysis
- ROI calculations (time saved, costs)
- Strategic recommendations
- Exported as PDF

Custom Reports #

Build your own reports with:

Metrics Selection:
Choose which metrics to include
- Conversation volume
- AI performance
- Customer satisfaction
- Channel breakdown
- etc.

Date Range:
- Custom start and end dates
- Compare two periods
- Year-over-year comparisons

Filters:
- Specific channels
- Customer segments
- AI agents (if multiple)
- Tags or labels

Visualization:
- Charts and graphs
- Tables
- Summary statistics

Schedule:
- One-time
- Daily, Weekly, Monthly
- Email recipients

Data Export #

Export raw data for external analysis:

Formats:
- CSV (spreadsheet)
- JSON (programmatic access)
- PDF (presentation-ready)

What Can Be Exported:
- Conversation histories
- Message-level data
- Metrics time-series
- Customer feedback
- Handoff logs

Use Cases:
- Import into business intelligence tools
- Share with stakeholders
- Compliance and record-keeping
- Custom analysis in Excel/Python


Continuous Improvement Workflow #

Use analytics to systematically improve your AI Agent.

Weekly Review Process #

1. Review Top Metrics (5 min)
- Automation rate
- Customer satisfaction
- Escalation rate
- Any concerning trends?

2. Identify Top Issues (10 min)
- Most common unanswered questions
- Lowest-confidence topics
- Highest escalation triggers
- Negative customer feedback

3. Prioritize Improvements (5 min)
- Quick wins (add 1-2 Q&As)
- Medium effort (update document)
- Large projects (new knowledge source)

4. Implement (throughout week)
- Add/update knowledge base content
- Adjust system prompts
- Refine escalation rules
- Test changes

5. Measure Impact (next review)
- Did metrics improve?
- Were specific issues resolved?
- Any new issues introduced?

Quarterly Deep Dive #

Every quarter, conduct comprehensive analysis:

Performance Review:
- Compare to previous quarter
- Benchmark against goals
- Identify major wins and misses

Knowledge Base Audit:
- Review all sources for accuracy
- Remove outdated content
- Consolidate duplicates
- Add missing information

Customer Journey Analysis:
- Map common conversation flows
- Identify friction points
- Optimize handoff experiences

ROI Calculation:
- Time saved by AI automation
- Cost per conversation
- Support team capacity increase
- Customer satisfaction improvement

Strategic Planning:
- Set goals for next quarter
- Plan major knowledge base projects
- Explore new channels or features
- Team training needs


Benchmarks and Goals #

Industry Benchmarks #

AI Automation Rate:
- Excellent: 70%+
- Good: 50-70%
- Needs Work: <50%

Customer Satisfaction (CSAT):
- Excellent: 4.5+/5
- Good: 4.0-4.5/5
- Needs Work: <4.0/5

First Contact Resolution:
- Excellent: 70%+
- Good: 50-70%
- Needs Work: <50%

Average Confidence Score:
- Excellent: 0.80+
- Good: 0.65-0.80
- Needs Work: <0.65

Setting Your Goals #

Consider your context when setting targets:

Industry:
- B2B SaaS: Higher complexity, lower automation
- E-commerce: High volume, higher automation potential
- Healthcare: Lower automation (safety/compliance)

Team Size:
- Small team (1-5): Target 60%+ automation
- Medium team (6-20): Target 50%+ automation
- Large team (20+): Target 40%+ automation (higher complexity)

AI Mode:
- Auto Response: Aim for 70%+ automation
- Hybrid: Aim for 90%+ suggestion acceptance
- Fallback: Aim for <10% fallback activation


FAQ #

How often should I review analytics? #

Daily: Quick glance at key metrics (5 min)
Weekly: Detailed review and minor adjustments (30 min)
Monthly: Comprehensive analysis and reporting (2 hours)
Quarterly: Deep dive and strategic planning (half day)

What's a good AI automation rate? #

Depends on your industry and complexity. For most businesses, 60-70% is excellent. Don't aim for 100% - some conversations genuinely need human judgment.

Why is my escalation rate so high? #

Common reasons:
- Knowledge base doesn't cover common questions
- Confidence threshold set too high (too conservative)
- Customers explicitly requesting humans (trust building needed)
- System prompt doesn't match customer language

How do I know which content to add? #

Use the "Unanswered Questions" report to see what customers are asking that the AI can't answer. Start with the most frequent questions first.

Can I track ROI of my AI Agent? #

Yes! Calculate:
- Time saved: Conversations automated × avg handling time × hourly rate
- Cost per conversation: Credits used / total conversations
- Capacity increase: % of team time freed up
- Customer satisfaction: Compare before/after AI deployment

What if customer satisfaction decreases after adding AI? #

This is uncommon but can happen. Check:
- Are responses accurate and helpful?
- Is AI tone appropriate for your brand?
- Are escalations handled smoothly?
- Are customers aware they're talking to AI?
- Consider starting in Hybrid mode for more control

How do I compare performance across multiple AI agents? #

If you have multiple agents (e.g., support vs sales), the dashboard shows comparative metrics side-by-side. You can also filter reports by specific agent.