Skip to main content
Your metrics dashboard transforms raw conversation data into actionable performance insights. Instead of guessing how well your AI is performing, you get concrete numbers that show what’s working, what needs improvement, and how changes affect your customer experience. This page shows you how to interpret your metrics, track trends over time, and use data to drive continuous improvement.

​
Why Metrics Matter

Performance metrics give you the visibility and confidence to make data-driven decisions about your AI:
  • Measure customer satisfaction - Track how users rate their AI interactions and identify dissatisfaction patterns
  • Quantify automation impact - See how much workload your AI handles autonomously vs. requiring human help
  • Identify improvement opportunities - Spot trends in escalations, unresolved conversations, and low-rated interactions
  • Validate changes - Confirm that knowledge updates and guidance refinements actually improve outcomes
  • Demonstrate ROI - Show stakeholders concrete evidence of reduced support burden and improved customer experience
  • Track weekend coverage - Measure how well your AI handles conversations when your team is offline
Effective teams review their metrics dashboard at least once per week, comparing current performance to the previous period to catch negative trends early and validate improvements.

​
Accessing the Metrics Dashboard

Navigate to Analyze → Metrics to view your performance analytics. The dashboard provides two distinct views optimized for different use cases: General View - Comprehensive overview of all AI activity, customer satisfaction, and conversation outcomes Ticketing View - Specialized metrics for support teams managing ticketed conversations with AI involvement Switch between views using the tabs at the top of the page.

​
Understanding Filters

All metrics respect your filter selections, allowing you to analyze specific segments of your data.

​
Time Range Filter

Location: Top right corner of the page The date range filter controls which conversations are included in all metrics and charts. You can select:
  • Last 7 days (default for quick overviews)
  • Last 30 days (recommended for trend analysis)
  • Last 90 days (for long-term patterns)
  • Custom range (specify exact start and end dates)
How trends work: When viewing “Last 30 days,” metric cards automatically compare to the previous 30 days to show you whether performance is improving or declining. Trend indicators appear as green (positive change) or red (negative change) badges showing the percentage point difference.
Trend comparisons require enough historical data. If you just started using botBrains, trends won’t appear until you have data from two equivalent time periods.

​
Channel Filter

Location: Below the view tabs, left side Filter metrics to specific communication channels to understand performance differences:
  • Web - Website chat widget conversations
  • Zendesk - Support ticket conversations from Zendesk integration
  • Salesforce - Case conversations from Salesforce integration
  • Slack - Messages from Slack workspace integration
  • WhatsApp - Conversations through WhatsApp Business integration
Why filter by channel: Different channels have different user expectations and conversation patterns. Web chat users expect instant, concise responses. Email users tolerate longer wait times but want thorough answers. Analyzing channels separately helps you optimize your AI’s behavior for each context. Multi-select: You can select multiple channels to compare specific subsets, or leave all unselected to view all channels combined.

​
Label Filter

Location: Below the view tabs, center Filter metrics to conversations tagged with specific labels. Labels help you segment by:
  • Customer tier (Enterprise, Professional, Free)
  • Product area (Billing, Technical Support, Sales)
  • Issue type (Bug Report, Feature Request, Question)
  • Quality markers (Training Example, Needs Review)
Include labels (positive filter): Show only conversations with the selected labels. Exclude labels (negative filter): Hide conversations with the selected labels. You can combine both modes. For example, include “Enterprise” and exclude “Spam” to analyze only enterprise customer conversations that aren’t spam.
Label filters only work for conversations that have been labeled. Unlabeled conversations won’t appear when filtering by labels, which may make your metrics seem lower than reality.

​
General View Metrics

The General view provides a comprehensive overview of your AI’s overall performance across all conversation types.

​
Overview Cards

The top row displays five key performance indicators with trend comparisons: Messages Total number of messages exchanged across all conversations in your selected timeframe. Includes both user messages and AI responses. 
Why it matters:
- Shows overall conversation volume
- Helps capacity planning
- Indicates growth or decline in support activity
- Rising messages with stable conversations = longer discussions
Conversations Total number of unique conversation threads started during the timeframe. Each conversation represents one customer inquiry, regardless of how many messages it contains. 
Use cases:
- Track support workload
- Measure seasonal fluctuations
- Identify unexpected spikes (product issues, marketing campaigns)
- Compare to previous period to detect trends
Unique Users Count of distinct users who started conversations. A single user may have multiple conversations, but they’re counted once. 
Insights from this metric:
- New vs. returning user ratio
- Customer retention indicators
- Impact of product changes on support volume
- Effectiveness of self-service improvements
CSAT Score Customer Satisfaction score calculated as the percentage of satisfied customers (4-5 star ratings) out of all rated conversations. 
CSAT = (Good + Amazing ratings) / (All 1-5 ratings) × 100%

Benchmarks:
- 80%+ = Excellent (industry-leading performance)
- 70-80% = Good (solid customer satisfaction)
- 60-70% = Fair (needs improvement)
- Below 60% = Poor (urgent attention required)
Click the CSAT card to filter conversations to only those with 4-5 star ratings, helping you identify what’s working well. Resolution Rate Percentage of conversations that were successfully resolved (neither escalated nor left unresolved). 
Resolution Rate = Resolved / (Resolved + Escalated + Unresolved) × 100%

What different rates mean:
- 80%+ = Strong autonomous performance
- 60-80% = Moderate effectiveness with room for improvement
- 40-60% = Significant knowledge gaps or guidance issues
- Below 40% = Major problems requiring immediate attention
Click the Resolution Rate card to view all resolved conversations.
Compare CSAT and Resolution Rate together: High resolution rate with low CSAT means the AI is answering quickly but not helpfully. High CSAT with low resolution rate means customers appreciate the effort even when escalation is needed.

​
Conversation Status Chart

Visual: Stacked area chart showing resolution status over time This chart breaks down your conversations by their final status across the selected date range: Resolved (Green) Conversations where the user’s question was successfully answered. This is your target outcome - the AI or support team fully addressed the customer’s need. Unresolved (Yellow) Conversations that ended without a satisfactory resolution. Common causes include knowledge gaps, unclear questions, or users abandoning conversations mid-way. Escalated (Purple) Conversations that were handed off to human agents, either automatically (based on escalation rules) or manually (user requested human help). Using this chart:
  • Track resolution trends over time - are you improving?
  • Identify specific dates with spikes in escalations or unresolved conversations
  • Correlate status changes with deployments or knowledge updates
  • Hover over any point to see exact counts for that day
A small percentage of escalations is normal and healthy - some questions genuinely require human expertise or judgment. Focus on reducing unnecessary escalations for routine questions.

​
Conversation Rating Chart

Visual: Histogram showing distribution of customer ratings This chart displays how customers rated their AI interactions: Rating scale:
  • 1 star (😠 Terrible) - Very dissatisfied
  • 2 stars (🙁 Bad) - Dissatisfied
  • 3 stars (😐 OK) - Neutral
  • 4 stars (😊 Good) - Satisfied
  • 5 stars (🤩 Amazing) - Very satisfied
  • Abandoned (0) - Survey offered but not completed
  • Unoffered - Survey not presented to user
Reading the distribution: Healthy distribution:
  • Majority of ratings at 4-5 stars
  • Small tail at 1-2 stars (under 10%)
  • Few abandonments (indicates survey timing is good)
Concerning patterns:
  • Bimodal distribution (peaks at both 1 and 5) - inconsistent experience
  • High 1-2 star ratings - serious quality issues
  • Many abandonments - survey timing or UX problems
  • High “unoffered” percentage - need to increase survey coverage
Click any bar to filter to conversations with that specific rating.

​
Message Volume Chart

Visual: Area chart showing messages and conversations over time This dual-metric chart helps you understand conversation patterns: Total Messages (Blue area) Shows the volume of all messages exchanged. Spikes indicate busy periods or particularly complex issues requiring many back-and-forth exchanges. Total Conversations (Orange area) Shows how many conversation threads were active. Helps distinguish between “many conversations” vs. “long conversations.” Key ratios to watch: 
Average messages per conversation = Total Messages / Total Conversations

Typical ranges:
- 2-4 messages = Quick, simple questions
- 5-8 messages = Moderate complexity or follow-ups
- 9+ messages = Complex issues or multiple related questions

Rising average = Questions getting more complex or AI struggling
Declining average = Improved AI effectiveness or better knowledge

​
AI Involvement Rate Chart

Visual: Pie chart showing AI participation levels This chart categorizes conversations by how the AI was involved: Fully Autonomous (Green) AI handled the entire conversation without any human operator involvement. This represents complete automation and maximum efficiency. 
Why this matters most:
- Zero human time required
- Scales infinitely
- Immediate response 24/7
- Lowest cost per conversation
- Goal: Maximize this category over time
Public Involvement (Blue) AI generated customer-visible responses, and a human operator also got involved. The AI started the conversation or assisted, but human expertise was ultimately needed. 
Common scenarios:
- AI provided initial answer, human added details
- AI maintained conversation until agent available
- Complex question requiring both AI knowledge and human judgment
Private Involvement (Purple) AI suggested responses internally to your support team (copilot mode), but all customer-facing messages came from human operators. 
Use cases:
- High-stakes conversations (legal, billing disputes)
- Training new support agents
- Maintaining human touch while getting AI assistance
- Specialized technical support
Not Involved (Gray) Zero AI involvement. These are typically imported historical tickets, outbound messages from operators, or conversations where AI was disabled.
A mature AI deployment typically sees 60-70% fully autonomous, 20-30% public involvement, and 5-10% private involvement. This mix provides automation efficiency while maintaining quality and human oversight where needed.

​
Handoff Chart

Visual: Visualization showing escalation patterns and reasons Shows when and why conversations were handed off from AI to human agents: Handoff triggers:
  • User explicitly requested human help
  • AI detected it couldn’t answer confidently
  • Automatic escalation rule triggered (based on topic, sentiment, or other criteria)
  • Support agent manually took over conversation
Use this chart to identify:
  • Peak handoff times (weekends, after hours, specific days)
  • Most common escalation reasons
  • Topics that frequently require human intervention
  • Opportunities to reduce unnecessary handoffs through better knowledge

​
Answer Completeness Chart

Visual: Pie chart showing response quality distribution Measures whether the AI provided complete answers or indicated missing information: Complete (Green) AI provided a full answer based on your knowledge base without caveats about missing information. Incomplete (Yellow) AI answered but indicated uncertainty or missing details (“I don’t have complete information about…”) No Answer (Red) AI explicitly stated it couldn’t answer the question due to missing knowledge. Why this matters:
  • Identifies knowledge gaps systematically
  • Tracks improvement as you add knowledge
  • Helps prioritize which missing information to add first
  • Distinguishes between “wrong answers” and “admitted gaps”
An honest “I don’t know” (No Answer) is better than a confident but incorrect response. Monitor the incomplete/no-answer percentages to guide your knowledge improvement efforts.

​
User Sentiment Chart

Visual: Bar chart showing emotional tone distribution Analyzes the sentiment of user messages using natural language processing: Positive (Green) User messages expressing satisfaction, gratitude, or positive emotions. Neutral (Gray) Factual questions or statements without emotional tone. Negative (Red) User messages expressing frustration, anger, or dissatisfaction. Using sentiment data: 
Sentiment trends to watch:
- Rising negative sentiment = Growing customer frustration
- High negative with high CSAT = Users frustrated with problem, not AI
- High negative with low CSAT = Users frustrated with AI's responses
- Positive sentiment shift after changes = Successful improvement
Filter to negative sentiment conversations to find the most frustrated users and understand what’s causing dissatisfaction.

​
User Rating Trend Chart

Visual: Line chart showing rating distribution over time Tracks how customer ratings evolve across your selected time period: Reading the trend lines:
  • Each colored line represents a rating level (1-5 stars)
  • Y-axis shows percentage of total rated conversations
  • Upward-sloping 4-5 star lines = improving satisfaction
  • Downward-sloping 1-2 star lines = fewer bad experiences
What to look for:
  • Sustained improvement in 4-5 star percentage
  • Reduction in 1-2 star percentage
  • Correlation with your deployments or knowledge updates
  • Day-of-week patterns (weekends often differ from weekdays)

​
User Language Chart

Visual: Horizontal bar chart showing language distribution Shows which languages your users communicate in: Why this matters:
  • Identify need for multilingual knowledge
  • Verify your AI handles non-English languages appropriately
  • Detect unexpected language patterns (could indicate spam or new markets)
  • Plan internationalization priorities
If you see significant non-English traffic but low satisfaction in those languages, this indicates a need for translated knowledge or multilingual support.

​
Usage by Page Chart

Visual: Horizontal bar chart showing conversation sources Displays which pages or entry points generated conversations (for web chat): Insights from this data:
  • High-traffic pages that need better self-service content
  • Product areas generating most support questions
  • Effectiveness of page-specific AI guidance
  • Opportunities for contextual knowledge (page-specific responses)
Example: If your pricing page generates many conversations, consider adding comprehensive pricing information to your knowledge base or embedding an FAQ directly on that page.

​
Knowledge Source Usage Chart

Visual: Horizontal bar chart showing data provider usage Shows which knowledge sources (data providers) your AI references most frequently: Knowledge sources:
  • PDFs (uploaded documentation)
  • Webpages (crawled URLs)
  • Snippets (manually created Q&A)
  • Tables (structured data)
  • Files (other uploaded content)
Using this data:
  • Identify which knowledge types are most valuable
  • Detect underutilized knowledge sources (might need better content)
  • Validate that important documentation is being referenced
  • Prioritize updates to frequently-used sources

​
Conversation Length Chart

Visual: Histogram showing message count distribution Displays how many messages typical conversations contain: Interpreting the distribution: Short conversations (1-3 messages):
  • Quick, simple questions
  • “Thank you and goodbye” interactions
  • Potentially unresolved issues where user gave up
Medium conversations (4-8 messages):
  • Normal back-and-forth for moderate complexity
  • Follow-up questions after initial answer
  • Multi-part inquiries
Long conversations (9+ messages):
  • Complex technical issues
  • Multiple related questions
  • Potential AI confusion (repeating itself or not understanding)
Red flags:
  • Many single-message conversations (users not engaging)
  • Very long conversations (AI not resolving efficiently)
  • Increasing average length over time (quality degradation)

​
Activity Heatmaps

Visual: Two calendar heatmaps showing conversation patterns Weekly Heatmap Shows conversation volume by day of week and hour of day for the most recent week: 
Use cases:
- Identify peak support hours
- Plan human agent staffing
- Optimize AI maintenance windows (deploy during quiet hours)
- Understand when weekend coverage matters
Yearly Heatmap Shows conversation volume by day of year for the past 365 days: 
Insights:
- Seasonal patterns (holidays, back-to-school, tax season)
- Impact of product launches or marketing campaigns
- Long-term growth trends
- Anomalies or unusual spikes
Darker cells indicate higher conversation volume. Click any cell to filter to conversations from that specific time period.

​
Hidden Conversations Chart

Visual: Pie chart showing spam and blocked conversations Displays conversations that were hidden from main views: Spam (Red) Conversations automatically or manually marked as spam. These are typically bot attacks, gibberish, or irrelevant messages. Blocked (Orange) Conversations from blocked users or domains. Used to prevent abusive users from consuming resources. Visible (Green) Normal, legitimate conversations that appear in your conversation list. Why monitor this:
  • Ensure spam detection isn’t too aggressive (legitimate users blocked)
  • Track abuse or attack patterns
  • Validate that your spam filters are working
  • Clean up spam manually if automatic detection missed it
High spam rates may indicate you need to implement CAPTCHA, rate limiting, or stricter content filtering on your chat widget.

​
Ticketing View Metrics

The Ticketing view focuses on AI involvement in support ticket workflows, providing specialized metrics for teams managing traditional ticketing systems with AI assistance.

​
Key Ticketing Metrics

The top row displays four critical ticketing performance indicators: Involvement Rate Percentage of tickets where the AI participated in some way (autonomous, public, or private involvement). 
Involvement Rate = (Autonomous + Public + Private) / Total Tickets × 100%

What this tells you:
- How much of your ticket volume AI is helping with
- Adoption of AI among support team
- Potential for further automation
- ROI of your AI investment

Target: 80%+ involvement rate means AI is assisting with most tickets
Involved Tickets Absolute count of tickets where AI was involved, with trend comparison to previous period. 
Why absolute numbers matter:
- Track actual workload assistance
- Calculate time saved (involved tickets × avg handling time)
- Demonstrate ROI in ticket reductions
- Measure growth in AI usage
Relative Autonomous Rate Percentage of involved tickets that were handled fully autonomously (no human intervention needed). 
Relative Autonomous Rate = Autonomous / (Autonomous + Public + Private) × 100%

This is different from "Autonomous Rate" because it only looks at tickets where AI was involved, excluding human-only tickets.

Benchmarks:
- 60%+ = Strong autonomous performance
- 40-60% = Moderate autonomy with significant human assistance
- Below 40% = AI mostly acting as copilot, limited full automation
Relative Autonomous Rate removes “not involved” tickets from the calculation, giving you a clearer picture of how effective the AI is when it does participate.
Better Monday Score Percentage of weekend tickets (Saturday and Sunday) where the AI provided at least one customer-visible response. 
Better Monday Score = Weekend AI-Responded Tickets / Total Weekend Tickets × 100%

Why this matters:
- Fewer tickets pile up for Monday morning
- Customers get responses even when team is offline
- Reduces Monday workload spike
- Improves customer experience with 24/7 availability

The tooltip provides the full calculation:
(Autonomous + Public weekend tickets) / Total weekend tickets × 100%
A high Better Monday Score (70%+) means your AI is effectively providing weekend coverage, reducing the traditional Monday morning ticket backlog.
Better Monday Score is one of the most tangible ROI metrics for support teams. If your AI handles 50 tickets autonomously each weekend, that’s 50 tickets your team doesn’t face Monday morning.

​
Involvement Flow (Sankey Diagram)

Visual: Sankey flow diagram showing ticket paths through AI involvement levels This powerful visualization shows how tickets flow from different involvement categories to resolution outcomes: Flow structure:
  • Left side: AI involvement level (Autonomous, Public, Private, Not Involved)
  • Right side: Resolution outcome (Resolved, Escalated, Unresolved)
  • Flow width: Number of tickets following that path
Reading the flows: Autonomous → Resolved (wide green flow): Ideal outcome - AI fully handled these tickets without human help and successfully resolved them. This is your highest-efficiency scenario. Public → Resolved (blue-green flow): Good outcome - AI started the conversation, human finished it, ticket resolved. Shows effective human-AI collaboration. Private → Resolved (purple-green flow): Copilot success - AI suggested responses, human sent them, ticket resolved. Human maintained control but AI assisted. Any → Escalated (red flows): Tickets that required human intervention regardless of AI involvement level. Study these to understand what triggers escalations. Any → Unresolved (yellow flows): Problematic tickets that weren’t fully resolved. These represent knowledge gaps or process issues requiring attention. Using this chart strategically:
  1. Maximize Autonomous → Resolved: This is pure automation. Focus improvements here.
  2. Minimize Not Involved → Any: If AI isn’t even attempting many tickets, investigate why.
  3. Analyze Autonomous → Escalated: These are tickets where AI tried to help but had to give up. Key improvement opportunity.
  4. Review Public → Unresolved: Even with human help, these weren’t resolved. Product or process issues?

​
AI Involvement vs Success Pivot Table

Visual: Interactive pivot table showing success rates across involvement and outcome dimensions This table provides a detailed breakdown of how different AI involvement levels correlate with resolution outcomes: Rows (AI Involvement):
  • Fully Autonomous
  • Public Involvement
  • Private Involvement
  • Not Involved
Columns (Outcomes):
  • Resolved
  • Escalated
  • Unresolved
  • Total
Cells: Show count and percentage for each combination How to read this table: 
Example interpretation:
Fully Autonomous: 450 tickets (75%)
  Resolved: 360 (80% of autonomous tickets)
  Escalated: 45 (10% of autonomous tickets)
  Unresolved: 45 (10% of autonomous tickets)

This tells you:
- AI attempts 450 tickets fully autonomously
- 80% success rate when it tries
- 10% escalation rate (these are knowledge gaps to address)
- 10% unresolved (users may have abandoned or question was unclear)
Strategic uses: Compare resolution rates across involvement types: If autonomous tickets have 80% resolution but public has 90%, humans are adding significant value. If both are 80%, humans aren’t improving outcomes much - consider increasing autonomous handling. Identify escalation patterns: High escalation rates in autonomous tickets indicate topics where AI is correctly recognizing its limitations. Review these to add knowledge or create escalation rules. Measure copilot effectiveness: If private involvement has high resolution rates, your support team is successfully using AI suggestions. If low, they may not trust the AI or suggestions are poor quality.

​
Involvement Rate Over Time Chart

Visual: Stacked bar chart showing involvement distribution across time periods Tracks how AI involvement levels evolve over your selected date range: Bars represent:
  • Green: Fully Autonomous tickets
  • Blue: Public Involvement tickets
  • Purple: Private Involvement tickets
  • Gray: Not Involved tickets
Trend patterns to watch: Growing green (autonomous): Your AI is successfully automating more tickets over time. This indicates improving knowledge and guidance. Shrinking gray (not involved): More tickets are getting AI assistance. Good for ROI and coverage. Growing purple (private): Support team is increasingly using AI as copilot. This suggests trust in AI suggestions and effective internal adoption. Stable or growing blue (public): AI is involved but still requires human finishing. If this grows while autonomous shrinks, investigate why - it could indicate knowledge degradation. Using this for planning: 
Calculate automation trajectory:
- Week 1: 40% autonomous
- Week 4: 55% autonomous
- Growth: +15 percentage points in 3 weeks
- Projection: Could reach 70% autonomous in 2 months

This helps set realistic automation goals and demonstrate progress.

​
Involvement Rate Evolution Chart

Visual: Multi-line chart showing involvement category trends Similar to the stacked bar chart but with separate lines for each involvement type, making it easier to see individual trends: Lines:
  • Green line: Autonomous rate trend
  • Blue line: Public involvement trend
  • Purple line: Private involvement trend
  • Gray line: Not involved trend
Cross-reference points: When autonomous line rises while public line falls: AI is successfully taking over tickets that previously required human finishing. When private line rises while not-involved falls: Support team is adopting AI copilot features for tickets they previously handled alone. When all involvement lines rise while not-involved falls: Overall AI adoption is increasing across all use cases.

​
Interpreting Metrics Together

Individual metrics tell part of the story. Combining metrics reveals deeper insights:

​
Healthy Performance Pattern

CSAT Score: 80%+ (users are satisfied)
Resolution Rate: 75%+ (most questions answered)
Autonomous Rate: 60%+ (high automation)
Better Monday Score: 70%+ (weekend coverage working)
Low 1-2 star ratings: <10% (few bad experiences)
Conversation Length: 4-6 messages avg (efficient resolution)
This pattern indicates a well-tuned AI providing effective, automated support with high customer satisfaction.

​
Knowledge Gap Pattern

Resolution Rate: 50-60% (moderate success)
Answer Completeness: 40% "No Answer" (missing knowledge)
Unresolved Status: 30%+ (many unresolved)
CSAT: 60-70% (users frustrated with lack of answers)
Escalation Rate: 25%+ (AI frequently gives up)
This pattern indicates systematic knowledge gaps. Priority: Review unresolved conversations and add missing information to data providers.

​
Quality Problem Pattern

CSAT: Below 60% (users dissatisfied)
Resolution Rate: 70%+ (high resolution claim)
High 1-2 star ratings: 20%+ (many complaints)
Negative Sentiment: 40%+ (frustrated users)
Conversation Length: 8+ messages avg (inefficient)
This pattern suggests the AI is marking conversations “resolved” but users disagree. The AI may be providing incorrect answers confidently or not understanding questions. Priority: Review low-rated conversations and refine guidance.

​
Adoption Problem Pattern

Involvement Rate: Below 50% (AI underutilized)
Not Involved: 50%+ of tickets (many human-only)
Private Involvement: Very low (team not using copilot)
Autonomous: Low but with high resolution (AI works when used)
This pattern indicates the AI performs well but isn’t being used enough. Priority: Team training, better escalation rules, or integration improvements to increase AI participation.

​
Weekend Coverage Success Pattern

Better Monday Score: 75%+ (strong weekend handling)
Weekend conversations: High volume
Weekend autonomous rate: 65%+ (mostly automated)
Weekend CSAT: Similar to weekday (quality maintained)
Monday morning ticket backlog: Reduced vs. baseline
This pattern shows successful 24/7 AI coverage, particularly valuable for reducing Monday morning workload.
Screenshot your metrics dashboard weekly and compare month-over-month. Visual trends are easier to spot than numbers alone, and historical screenshots help you remember what changed when.

​
Filtering and Segmentation Strategies

Use filters to uncover insights hidden in aggregate numbers:

​
Compare Time Periods

Strategy: Analyze same metrics for different time ranges 
Example:
1. View metrics for "Week after deployment" (Nov 1-7)
2. Note CSAT, Resolution Rate, Autonomous Rate
3. Change date range to "Week before deployment" (Oct 25-31)
4. Compare metrics to measure deployment impact

Result: "Deployment increased autonomous rate from 45% to 62% while maintaining 82% CSAT"

​
Segment by Channel

Strategy: Analyze each channel separately to optimize channel-specific performance 
Example workflow:
1. Filter to "Web" channel only
   - Note: CSAT 78%, Resolution 82%, Avg length 3 messages
2. Filter to "Email" channel only
   - Note: CSAT 85%, Resolution 75%, Avg length 6 messages
3. Insight: Email users are more satisfied but ask harder questions
4. Action: Optimize Web for speed, Email for thoroughness

​
Analyze High-Value Customers

Strategy: Use labels to focus on important customer segments 
Example:
1. Filter conversations with "Enterprise" label
2. Review CSAT and escalation rate for enterprise customers
3. Compare to overall metrics
4. If enterprise CSAT is lower, prioritize their common topics
5. Consider creating enterprise-specific guidance or escalation rules

​
Identify Problematic Topics

Strategy: Combine topic filters with low metrics 
Example process:
1. Navigate to Topics page, identify topics with low resolution
2. Click topic to filter conversations
3. Navigate to Metrics, review CSAT and sentiment for that topic
4. If metrics are poor, dive into conversations to find root cause
5. Add knowledge or refine guidance for that specific topic

​
Track Improvement Over Time

Strategy: Monitor same segment across multiple time periods 
Example tracking:
Week 1: "Billing" topic - 45% resolution, 65% CSAT
Week 2: Added 10 billing snippets
Week 3: "Billing" topic - 58% resolution, 72% CSAT (+13pp, +7pp)
Week 4: "Billing" topic - 67% resolution, 78% CSAT (+9pp, +6pp)

Conclusion: Knowledge additions are working, improvements continuing

​
Exporting Metrics Data

Export your metrics for reporting, external analysis, or compliance purposes.

​
Quick Export

  1. Click the Export button in the top right of the metrics page
  2. Select data type (conversations, messages, users)
  3. Choose format (CSV for spreadsheets, JSON for programmatic analysis)
  4. Current filters are automatically applied to the export
  5. Download begins immediately
Exported data includes:
  • All visible metrics and their values
  • Timestamps for the data range
  • Filter parameters applied
  • Trend comparisons to previous period

​
Scheduled Exports

For regular reporting, set up automatic recurring exports:
  1. Navigate to Settings → Data Exports
  2. Click Create Scheduled Export
  3. Configure:
    • Frequency (daily, weekly, monthly)
    • Data type (metrics summary, raw conversations, etc.)
    • Filters to apply
    • Delivery method (email, webhook, cloud storage)
  4. Save the schedule
Use cases for scheduled exports:
  • Weekly executive reports
  • Monthly board presentations
  • Quarterly business reviews
  • Compliance archiving
  • External analytics tools (feed to Tableau, PowerBI, etc.)

​
Troubleshooting Metrics

​
Metrics seem incorrect or incomplete

Issue: Numbers don’t match expectations or seem too low Solutions:
  • Verify date range is set correctly (not accidentally in future)
  • Check if filters are applied (channel, label) that limit the data
  • Ensure conversations have finished (ongoing conversations may not have final status)
  • Confirm data sync has completed for integrated channels (Zendesk, Salesforce)
  • Refresh the page to clear any cached stale data
Issue: Trend indicators are missing or always show 0% change Solutions:
  • Ensure you have data from at least two equivalent time periods (e.g., 60 days of history for 30-day trend)
  • Check that you just started using botBrains (trends require historical comparison data)
  • Verify filters haven’t changed between periods (comparing “Web” to “All Channels” shows false trends)
  • Confirm date range allows trend calculation (custom ranges too short may not have comparison data)

​
Better Monday Score is 0% or not calculating

Issue: Weekend coverage metric shows 0% or doesn’t appear Solutions:
  • Verify your date range includes at least one Saturday or Sunday
  • Check that you actually had conversations on weekends (no weekend traffic = no score)
  • Ensure conversations are not filtered out by channel or label restrictions
  • Confirm AI is deployed and active on weekends (check deployment schedule)

​
Charts loading slowly or timing out

Issue: Dashboard takes a long time to load or shows errors Solutions:
  • Reduce date range to analyze shorter time periods (30 days instead of 90)
  • Remove channel and label filters temporarily to reduce query complexity
  • Refresh page to clear any stuck queries
  • Try using General view instead of Ticketing view (simpler calculations)
  • Contact support if issue persists (may indicate data optimization needed)

​
Exported data doesn’t match dashboard

Issue: CSV export has different numbers than what’s shown on screen Solutions:
  • Verify export was completed after applying filters (don’t change filters mid-export)
  • Check export timestamp vs. dashboard timestamp (data may have updated between views)
  • Ensure export format settings match expected structure
  • Confirm you’re comparing the same time range and filters
  • Re-export with explicit filter documentation to verify consistency

​
Best Practices

​
Establish a Metrics Review Routine

Weekly Review (15 minutes):
  1. Check CSAT and Resolution Rate trends - are they improving?
  2. Review any sudden drops or spikes in key metrics
  3. Filter to 1-2 star ratings, scan recent poor experiences
  4. Note any growing topics or unusual patterns
  5. Track Better Monday Score to validate weekend coverage
Monthly Deep Dive (1 hour):
  1. Compare month-over-month metrics across all categories
  2. Segment analysis by channel to identify optimization opportunities
  3. Review involvement rate evolution - is automation increasing?
  4. Analyze topic-specific metrics for your top 10 topics
  5. Export data for stakeholder reports
  6. Document improvements made and their measured impact
Quarterly Review (2-3 hours):
  1. Comprehensive trend analysis across 90-day periods
  2. Calculate ROI metrics (tickets automated, time saved, costs reduced)
  3. Validate long-term strategic goals (automation %, CSAT targets)
  4. Identify seasonal patterns for future planning
  5. Present findings to leadership with recommendations

​
Set Realistic Goals and Track Progress

Define specific, measurable targets based on your current baseline: 
Example goal setting:
Current: 55% Autonomous Rate, 75% CSAT, 68% Resolution Rate
3-month goals: 65% Autonomous (+10pp), 80% CSAT (+5pp), 75% Resolution (+7pp)

Track weekly:
Week 1: 55% / 75% / 68% (baseline)
Week 4: 58% / 76% / 70% (+3pp / +1pp / +2pp) - on track
Week 8: 61% / 78% / 73% (+3pp / +2pp / +3pp) - on track
Week 12: 66% / 81% / 76% (+5pp / +3pp / +3pp) - exceeded goals

​
Correlate Metrics with Actions

Always connect changes to outcomes to understand what works: Create a change log: 
Nov 1: Added 15 billing snippets
Nov 8: Updated guidance for technical questions
Nov 15: Enabled weekend deployments
Nov 22: Integrated new help center articles
Review metrics with change dates: 
Nov 1-7: Resolution Rate 68%, Billing topic 45% resolved
Nov 8-14: Resolution Rate 71%, Billing topic 62% resolved (+17pp) ✓
Nov 15-21: Better Monday Score jumped from 55% to 78% (+23pp) ✓
Nov 22-28: Resolution Rate 74%, Answer completeness improved 12% ✓
This approach proves which changes drive improvement and which don’t.

​
Don’t Chase Perfect Metrics

Some important nuances to remember: 100% CSAT is not realistic or even desirable:
  • Some customers are dissatisfied with your product, not your AI
  • Honest “I don’t know” responses may get low ratings but are correct behavior
  • Controversial topics (pricing, policies) naturally have lower satisfaction
100% Autonomous Rate is not the goal:
  • Complex or sensitive issues legitimately require human judgment
  • Maintaining human touch for high-value customers adds strategic value
  • Some escalations prevent worse outcomes (wrong automated answers)
Focus on continuous improvement, not perfection:
  • 5 percentage point improvement in resolution rate per month is excellent
  • Sustained upward trends matter more than hitting arbitrary targets
  • Balance automation efficiency with answer quality and customer satisfaction

​
Combine Quantitative and Qualitative Analysis

Metrics show you what’s happening. Conversations show you why: Process:
  1. Metrics identify problem areas (e.g., low CSAT for “Refund” topic)
  2. Filter conversations to that segment (Topic: Refund, Rating: 1-2 stars)
  3. Read 10-20 conversations to understand root causes
  4. Make targeted improvements based on qualitative insights
  5. Track metrics to validate improvements worked
Don’t rely solely on numbers - always investigate the underlying conversations.

​
Next Steps

Now that you understand your performance metrics:
  • Review Conversations - Dive deep into individual conversations to understand metrics context
  • Analyze Topics - Segment metrics by topic to identify specific improvement areas
  • Search Messages - Find patterns in user questions and AI responses
  • Improve Answers - Use metric insights to guide knowledge and guidance refinement
  • Manage Labels - Create custom labels to segment metrics by business-specific categories
Remember: Metrics are a compass, not a destination. Use them to guide continuous improvement, validate that changes work, and demonstrate the value of your AI. Regular review, combined with conversation analysis and systematic refinement, compounds into dramatically better customer experiences and reduced support costs over time.