> ## Documentation Index > Fetch the complete documentation index at: https://help.gostanna.com/llms.txt > Use this file to discover all available pages before exploring further. # Sentiment Analysis > How Stanna analyzes communication sentiment to track relationship health ## Overview Stanna analyzes the emotional tone of client communications across Gmail, HubSpot, and Intercom. Sarcasm detection and context-aware scoring provide insights into client satisfaction and relationship health. Sentiment analysis dashboard showing timeline, breakdown percentages, sarcasm detection, and key factors ## How Sentiment Analysis Works ### Sentiment Scale Stanna uses a **-2 to +2** numerical scale for precise sentiment tracking: | Score | Classification | Indicators | | ------ | --------------- | ------------------------------------------------ | | **+2** | Highly Positive | Enthusiastic, appreciative, excited language | | **+1** | Positive | Satisfied, collaborative, constructive tone | | **0** | Neutral | Informational, transactional, factual | | **-1** | Negative | Frustrated, concerned, disappointed | | **-2** | Highly Negative | Angry, threatening churn, severe dissatisfaction | ### Analysis Methods **Gmail Integration:** * Subject line analysis for urgency and tone * Message content evaluation using natural language processing * Context-aware analysis considering email threads * Professional vs. personal communication tone detection * **Sarcasm detection** to correctly interpret ironic or sarcastic messages **Factors Considered:** * Language choice and formality * Urgency indicators (ALL CAPS, urgent keywords) * Positive language (thanks, great, excellent) * Negative language (issues, problems, disappointed) * Sarcastic patterns (e.g., "Oh great!", "Just wonderful", "Thanks a bunch") **Intercom Integration:** * Customer message tone in support tickets * Resolution satisfaction indicators * Escalation language patterns * Follow-up communication sentiment **Special Considerations:** * Support context affects baseline expectations * Issue resolution often improves sentiment over conversation * Escalation patterns indicate deeper relationship concerns **HubSpot Integration:** * Meeting notes and call summaries * Deal progression indicators * Contact engagement levels * Account activity patterns **Business Context:** * Meeting attendance and participation * Response rates to outreach * Deal advancement or stagnation * Contract negotiation tone ## Sentiment Weighting & Impact ### Directional Weighting Communication direction affects sentiment impact: * **Inbound negative**: Higher weight (client expressing dissatisfaction) * **Outbound negative**: Lower weight (you addressing issues) * **Inbound positive**: High weight (client expressing satisfaction) * **Outbound positive**: Moderate weight (your positive communications) ### Temporal Weighting Recent sentiment carries more weight than historical sentiment: * **Last 7 days**: 100% weight * **8-30 days**: 70% weight * **31-90 days**: 40% weight * **Over 90 days**: 20% weight This ensures current relationship status is prioritized over past issues that may have been resolved. ## Sarcasm Detection ### How Sarcasm Detection Works Stanna uses AI models to detect sarcasm and irony in client communications, ensuring that seemingly positive messages with sarcastic intent are correctly identified as negative sentiment. **Detection Methods:** * **AI-Powered Analysis**: Uses AI models trained on sarcasm detection * **Pattern Recognition**: Identifies common sarcastic phrases and structures * **Context Clues**: Analyzes punctuation patterns, contradictory statements, and exaggerated language **Common Sarcasm Patterns Detected:** * Exaggerated positive phrases: "Oh great!", "Just wonderful", "Exactly what I needed" * Quotation marks around positive words: "This is 'great'" * Contradictory statements: "This is perfect... but it doesn't work" * Excessive punctuation: Multiple exclamation marks or mixed punctuation * Common sarcastic expressions: "Thanks a bunch", "Brilliant", "Couldn't be better" **Sentiment Adjustment:** When sarcasm is detected with high confidence (>70%), the apparent sentiment is flipped: * Apparent positive becomes negative * The adjusted sentiment reflects the true negative intent * Confidence levels are maintained for transparency Sarcasm detection helps prevent misclassification of frustrated clients whose negative feelings are expressed through ironic positive language. ## Content Analysis Details ### Subject Line Analysis Email subject lines provide immediate sentiment indicators: * **Urgent/Escalation**: "URGENT", "Problem", "Issue", "Help needed" * **Positive**: "Thank you", "Great meeting", "Success", "Appreciation" * **Neutral**: "Meeting notes", "Follow-up", "Update", "Question" * **Negative**: "Disappointed", "Concerned", "Problem", "Not working" ### Content Snippet Processing Stanna analyzes key portions of message content: * **Opening and closing phrases**: Tone-setting language * **Key sentence sentiment**: Main message emotional content * **Context clues**: References to satisfaction, issues, or future plans * **Professional formality**: Changes in communication style ## Sentiment Attribution ### Understanding Sentiment Changes Track what drives sentiment variations: * **Specific interactions**: Which emails or conversations affected scores * **Content triggers**: Particular words or phrases that influenced analysis * **Trend patterns**: How sentiment is changing over time * **Contextual factors**: Business events that correlate with sentiment changes ### Manual Sentiment Overrides When automated analysis misinterprets context: 1. **Access client interaction history** 2. **Identify incorrectly analyzed interactions** 3. **Use manual attribution** to provide correct context 4. **Document reasoning** for future reference Manual overrides help train the AI system to better understand your specific business context and communication patterns. ## Integration-Specific Considerations ### Gmail Sentiment Patterns **Professional Email Norms:** * Formal business communication tends toward neutral baseline * Urgency doesn't always indicate negative relationship health * Thank you messages and positive feedback carry high positive weight * Issue reporting is expected and manageable if followed by resolution ### Intercom Support Context **Support Communication Expectations:** * Customers contacting support often start with neutral or slightly negative sentiment * Resolution quality significantly impacts final sentiment score * Multiple tickets about the same issue indicate deeper problems * Quick, effective resolution often results in positive sentiment ### HubSpot Business Intelligence **Deal and Meeting Context:** * Meeting attendance and engagement indicate relationship investment * Deal progression stagnation may reflect underlying concerns * Contract negotiations can temporarily affect sentiment * Long sales cycles require different sentiment interpretation ## Best Practices for Sentiment Management ### Proactive Monitoring Check for significant sentiment drops (≥1 point) in the past 24 hours Review the specific interactions that caused sentiment changes Reach out to clients showing declining sentiment trends before they become critical Acknowledge and build on positive sentiment interactions ### Communication Style Impact Use clear, professional language to avoid misinterpretation by AI analysis Acknowledge client concerns before addressing solutions to improve sentiment tracking Consistent follow-up communication helps maintain positive sentiment trends Document resolution of negative sentiment interactions to provide context ## Advanced Sentiment Features ### Sentiment Trending Track sentiment patterns over time: * **Rolling averages**: 7-day, 30-day, and 90-day sentiment trends * **Seasonal patterns**: Identify recurring sentiment cycles * **Intervention effectiveness**: Measure how actions affect sentiment recovery * **Predictive indicators**: Early warning signs of sentiment decline ### Stakeholder-Level Sentiment For clients with multiple contacts: * **Individual sentiment tracking**: Monitor each stakeholder separately * **Aggregate sentiment**: Overall account relationship health * **Influence weighting**: Key decision-maker sentiment carries more weight * **Sentiment divergence**: When different stakeholders show different sentiment patterns ### Sentiment-Based Automation Configure automated actions based on sentiment: * **Alert thresholds**: Notifications when sentiment drops below specified levels * **Task creation**: Automatic follow-up tasks for negative sentiment interactions * **Escalation triggers**: Alert management when sentiment reaches critical levels * **Success celebration**: Recognize and reinforce positive sentiment achievements ## Troubleshooting Sentiment Analysis **Common causes:** * Cultural or industry-specific communication styles * Complex sarcasm patterns not detected by AI models * Technical language interpreted as negative * Context missing from snippet analysis **Solutions:** 1. Use manual attribution to provide correct context 2. Review complete interaction context, not just snippets 3. Consider industry-specific communication norms 4. Document patterns for future reference **Check these factors:** * Are integrations syncing properly? * Is there sufficient interaction data for analysis? * Are interactions being correctly attributed to clients? **Troubleshooting:** 1. Verify integration sync status 2. Check for recent interactions in source systems 3. Confirm client domain matching is working correctly **Understanding volatility:** * Small interaction volumes can cause large sentiment swings * New clients may have unstable sentiment until more data is collected * Single negative interactions can disproportionately affect overall sentiment **Stabilization strategies:** 1. Focus on sentiment trends rather than individual values 2. Allow time for sentiment patterns to establish (30-60 days) 3. Use manual attribution to provide stabilizing context ## Privacy & Data Handling ### Sentiment Data Security * **Content analysis**: Only analyzes content for sentiment, doesn't store complete message text * **Encryption**: All sentiment data encrypted at rest and in transit * **Access control**: Sentiment details only available to authorized team members * **Data retention**: Sentiment scores retained according to workspace policies ### Compliance Considerations * **GDPR compliance**: Sentiment analysis respects data processing preferences * **Opt-out options**: Clients can request exclusion from automated sentiment analysis * **Data minimization**: Only processes content necessary for sentiment evaluation * **Transparency**: Clear documentation of how sentiment analysis affects client relationships **Cultural Sensitivity**: Sentiment analysis may interpret communication styles differently across cultures. Consider manual overrides for international clients with different communication norms. **Sentiment Training**: The AI system continuously learns from manual overrides and feedback. Regular use of manual attribution helps improve accuracy for your specific business context.