Skip to main content

The Sentiment Analysis Action

Analyze Sentiment: AI-powered sentiment analysis for customer feedback. Classifies text as positive/neutral/negative with confidence scores, polarity, and key highlights. Automate feedback escalation, prioritize support, and track satisfaction.

Jerome Bajou avatar
Written by Jerome Bajou
Updated yesterday

Sentiment Analysis Action

Overview

The Analyze Sentiment action uses AI to analyze the emotional tone and sentiment of text. This powerful tool helps you understand customer feedback, reviews, comments, and other text-based communications automatically.

What It Does

The Sentiment Analysis action:

  • Classifies sentiment as positive, neutral, or negative

  • Measures confidence in the classification (0-1 scale)

  • Calculates polarity from -1.0 (strongly negative) to +1.0 (strongly positive)

  • Provides probability scores for each sentiment category

  • Identifies key highlights - specific phrases that influenced the sentiment

  • Detects language of the input text

Key Benefits

  • Automate feedback analysis - Understand customer sentiment without manual review

  • Trigger actions based on sentiment (e.g., escalate negative feedback)

  • Track customer satisfaction over time

  • Prioritize responses to negative feedback

  • Analyze reviews and comments at scale

How to Use

Adding the Action

  1. In your workflow builder, click Add Action

  2. Navigate to the AI category

  3. Select Analyze Sentiment

  4. Configure the text input (see Configuration below)

  5. Use the output in subsequent workflow steps

Configuration

Text Input (Required)

The text to analyze for sentiment. You can:

  • Enter text directly: Type or paste the text you want to analyze

  • Use variables: Reference data from your workflow context

    • {customer.feedback} - Customer feedback text

    • {booking.notes} - Booking notes

    • {review.comment} - Review comments

    • Any text field from previous workflow steps

Example:

{customer.feedback}

Or combine variables with text:

Customer said: "{customer.feedback}" about booking {booking.id}

Variable Picker

Click the Use variable button below the text input to:

  • Browse available variables from your workflow context

  • Insert variables with proper syntax

  • See variable descriptions and data types

Output Data

After execution, the action adds a sentiment_analysis object to your workflow context. You can access this data in subsequent steps.

Output Structure

{   "sentiment_analysis": {     "sentiment": "positive",     "confidence": 0.95,     "polarity": 0.8,     "scores": {       "positive": 0.85,       "neutral": 0.10,       "negative": 0.05     },     "highlights": [       {         "text": "excellent service",         "sentiment": "positive"       },       {         "text": "highly recommend",         "sentiment": "positive"       }     ],     "language": "en"   } }

Output Fields Explained

sentiment

  • Type: String

  • Values: "positive", "neutral", or "negative"

  • Description: The primary sentiment classification

  • Example: "positive"

confidence

  • Type: Float

  • Range: 0.0 to 1.0

  • Description: How confident the model is in the sentiment classification

  • Example: 0.95 (95% confident)

polarity

  • Type: Float

  • Range: -1.0 to +1.0

  • Description: Sentiment intensity

    • 1.0 = Strongly negative

    • 0.0 = Neutral

    • +1.0 = Strongly positive

  • Example: 0.8 (strongly positive)

scores

  • Type: Object

  • Description: Probability distribution across all sentiment categories

  • Properties:

    • positive: Probability the text is positive (0.0-1.0)

    • neutral: Probability the text is neutral (0.0-1.0)

    • negative: Probability the text is negative (0.0-1.0)

  • Note: Scores typically sum to approximately 1.0

  • Example:

    {   "positive": 0.85,   "neutral": 0.10,   "negative": 0.05 }

highlights

  • Type: Array of objects

  • Description: Key phrases from the text that contributed to the sentiment decision

  • Structure: Each highlight contains:

    • text: The relevant phrase

    • sentiment: The sentiment of that phrase

  • Count: Typically 2-5 highlights

  • Example:

    [   {     "text": "excellent service",     "sentiment": "positive"   },   {     "text": "would not recommend",     "sentiment": "negative"   } ]

language

  • Type: String

  • Format: ISO 639-1 language code

  • Description: Detected or assumed language of the input text

  • Examples: "en" (English), "es" (Spanish), "fr" (French)

Using the Output

Accessing Sentiment Data

Use the output in subsequent workflow steps:

In Conditions

Check sentiment to branch your workflow:

{sentiment_analysis.sentiment} equals "negative"

Or use polarity for more nuanced checks:

{sentiment_analysis.polarity} less than -0.5

In Notifications

Include sentiment in messages:

Customer feedback sentiment: {sentiment_analysis.sentiment} Confidence: {sentiment_analysis.confidence}

In Webhooks

Send sentiment data to external systems:

{   "feedback": "{customer.feedback}",   "sentiment": "{sentiment_analysis.sentiment}",   "polarity": "{sentiment_analysis.polarity}",   "confidence": "{sentiment_analysis.confidence}" }

In Variables

Store sentiment for later use:

Variable name: feedback_sentiment Value: {sentiment_analysis.sentiment}

Use Cases

These are fictive uses cases to help you understand how you could use this action.

1. Customer Feedback Escalation

Scenario: Automatically escalate negative feedback to management

Workflow:

  1. Trigger: Customer Created or Booking Confirmed

  2. Action: Analyze Sentiment (analyze {customer.feedback})

  3. Condition: If {sentiment_analysis.sentiment} equals "negative"

  4. Action (True branch): Send Notification to manager

  5. Action (False branch): Send standard thank you email

2. Review Monitoring

Scenario: Track sentiment of customer reviews

Workflow:

  1. Trigger: Webhook (when review is submitted)

  2. Action: Analyze Sentiment (analyze review text)

  3. Action: Update Customer (store sentiment in custom field)

  4. Condition: If {sentiment_analysis.polarity} less than -0.3

  5. Action (True branch): Create task for follow-up

3. Support Ticket Prioritization

Scenario: Prioritize support tickets based on sentiment

Workflow:

  1. Trigger: Customer Created (with support ticket)

  2. Action: Analyze Sentiment (analyze ticket description)

  3. Condition: If {sentiment_analysis.polarity} less than -0.5

  4. Action (True branch): Set ticket priority to "High"

  5. Action (False branch): Set ticket priority to "Normal"

4. Automated Response Selection

Scenario: Choose response template based on sentiment

Workflow:

  1. Trigger: Booking Confirmed

  2. Action: Analyze Sentiment (analyze {booking.notes})

  3. Condition: If {sentiment_analysis.sentiment} equals "positive"

  4. Action (True branch): Send Notification (positive template)

  5. Condition (False branch): If {sentiment_analysis.sentiment} equals "negative"

  6. Action (True branch): Send Notification (apology template)

  7. Action (False branch): Send Notification (neutral template)

5. Sentiment Tracking Dashboard

Scenario: Log sentiment data for analytics

Workflow:

  1. Trigger: Scheduled Time (daily)

  2. Action: Find Bookings (from last 24 hours)

  3. Action: Loop (for each booking)

  4. Action: Analyze Sentiment (analyze {current_item.notes})

  5. Action: Send Webhook (send to analytics system)

Examples

Example 1: Basic Sentiment Check

Input Text:

"Amazing experience! The staff was friendly and the service was excellent. Highly recommend!"

Output:

{   "sentiment_analysis": {     "sentiment": "positive",     "confidence": 0.98,     "polarity": 0.92,     "scores": {       "positive": 0.95,       "neutral": 0.04,       "negative": 0.01     },     "highlights": [       {"text": "Amazing experience", "sentiment": "positive"},       {"text": "excellent", "sentiment": "positive"},       {"text": "Highly recommend", "sentiment": "positive"}     ],     "language": "en"   } }

Example 2: Negative Feedback

Input Text:

"Very disappointed with the service. The booking was incorrect and staff was unhelpful. Will not return."

Output:

{   "sentiment_analysis": {     "sentiment": "negative",     "confidence": 0.96,     "polarity": -0.85,     "scores": {       "positive": 0.02,       "neutral": 0.12,       "negative": 0.86     },     "highlights": [       {"text": "Very disappointed", "sentiment": "negative"},       {"text": "incorrect", "sentiment": "negative"},       {"text": "unhelpful", "sentiment": "negative"},       {"text": "Will not return", "sentiment": "negative"}     ],     "language": "en"   } }

Example 3: Neutral/Mixed Sentiment

Input Text:

"The booking process was straightforward. The experience was okay, nothing special but no major issues."

Output:

{   "sentiment_analysis": {     "sentiment": "neutral",     "confidence": 0.78,     "polarity": 0.05,     "scores": {       "positive": 0.25,       "neutral": 0.65,       "negative": 0.10     },     "highlights": [       {"text": "straightforward", "sentiment": "positive"},       {"text": "okay", "sentiment": "neutral"},       {"text": "nothing special", "sentiment": "neutral"}     ],     "language": "en"   } }

Credits and Costs

Credit Usage

The Sentiment Analysis action uses AI credits based on actual token usage:

  • Prompt tokens: Text you send for analysis

  • Completion tokens: AI response tokens

Credits are charged after successful execution. The cost depends on:

  • Length of input text

  • Complexity of the analysis

Cost Optimization Tips

  1. Analyze only necessary text - Don't include unnecessary context

  2. Use filters - Only analyze feedback that meets certain criteria

  3. Batch processing - Consider analyzing multiple items in a loop efficiently

  4. Monitor usage - Check execution logs to see credit consumption

Best Practices

1. Input Text Quality

  • Provide complete context - Include enough text for accurate analysis

  • Avoid very short text - Single words may not provide reliable sentiment

  • Clean the text - Remove formatting that might interfere (if possible)

2. Interpreting Results

  • Use confidence scores - Low confidence (< 0.7) may indicate ambiguous text

  • Consider polarity - More nuanced than simple positive/negative

  • Review highlights - Understand what influenced the sentiment

  • Check language - Ensure detected language matches expectations

3. Workflow Design

  • Handle edge cases - What if sentiment is neutral?

  • Set thresholds - Use polarity for more nuanced decisions than just sentiment

  • Combine with other data - Don't rely solely on sentiment

  • Log results - Store sentiment for historical analysis

4. Error Handling

  • Validate input - Ensure text exists before analyzing

  • Handle failures - Set up notifications for analysis failures

  • Provide fallbacks - Default behavior if sentiment can't be determined

5. Privacy and Ethics

  • Respect privacy - Only analyze text you have permission to process

  • Use responsibly - Don't use sentiment to make unfair decisions

  • Be transparent - Inform customers if their feedback is analyzed

Troubleshooting

Common Issues

Empty or Invalid Text

Problem: Action fails with "Text is required" error

Causes:

  • Variable is empty or null

  • Variable doesn't exist in context

  • Text field is not properly configured

Solutions:

  • Check that the variable contains text before analyzing

  • Use a condition to verify text exists

  • Provide a fallback value: {customer.feedback||No feedback provided}

Low Confidence Scores

Problem: Sentiment analysis returns low confidence (< 0.5)

Causes:

  • Text is too short or ambiguous

  • Mixed sentiment in the text

  • Unclear or contradictory language

Solutions:

  • Ensure sufficient text length (at least a sentence)

  • Review the highlights to understand ambiguity

  • Consider using polarity instead of sentiment for nuanced decisions

Unexpected Sentiment Classification

Problem: Sentiment doesn't match your expectations

Causes:

  • Sarcasm or irony in text

  • Context-dependent meaning

  • Cultural or language nuances

Solutions:

  • Review the highlights to see what influenced the decision

  • Check the polarity score for more nuance

  • Consider manual review for critical decisions

  • Use confidence scores to identify uncertain classifications

API Errors

Problem: "API error" or "Failed to analyze sentiment"

Causes:

  • API service temporarily unavailable

  • Invalid API configuration

  • Rate limiting

Solutions:

  • Retry the workflow execution

  • Contact support if issues persist

  • Enable failure notifications to catch errors quickly

Debugging Tips

  1. Check execution logs - Review input and output data

  2. Test with sample text - Verify the action works with known examples

  3. Use test mode - Test workflows before activating

  4. Verify variables - Ensure text variables contain expected data

  5. Review highlights - Understand what influenced the sentiment

Technical Details

AI Model

The action uses GPT-4o-mini for sentiment analysis, optimized for:

  • Fast response times

  • Cost efficiency

  • Accurate sentiment classification

Response Format

The AI returns structured JSON with:

  • Sentiment classification

  • Confidence metrics

  • Detailed scores

  • Key highlights

  • Language detection

Language Support

The action can analyze text in multiple languages. The detected language is returned in the language field using ISO 639-1 codes.

Limitations

Text Length

  • Very short text (< 10 characters) may produce less reliable results

  • Very long text (> 10,000 characters) may be truncated or produce less focused analysis

  • Optimal length: A few sentences to a paragraph

Accuracy Considerations

  • Sarcasm and irony may be misclassified

  • Context-dependent meaning may not be fully captured

  • Mixed sentiment may result in neutral classification

  • Cultural nuances may affect interpretation

Processing Time

  • Analysis typically completes in 1-3 seconds

  • Longer text may take slightly more time

  • Network latency can affect response times

Related Actions

  • Send Notification - Use sentiment to personalize messages

  • Set Variable - Store sentiment for later use

  • Condition - Branch workflows based on sentiment

  • Send Webhook - Forward sentiment data to external systems

Conclusion

The Sentiment Analysis action is a powerful tool for understanding customer feedback and automating responses based on emotional tone. By integrating sentiment analysis into your workflows, you can:

  • Respond faster to negative feedback

  • Identify satisfied customers for testimonials

  • Track customer satisfaction trends

  • Automate personalized communications

  • Make data-driven decisions about customer service

Start with simple use cases and gradually build more complex workflows as you become comfortable with the sentiment data.

For more information about workflows in general, see the Workflows Documentation.

Did this answer your question?