Our support team reads between the lines of every ticket, looking for signs of urgency or satisfaction. Our sales team analyzes prospect replies, trying to gauge genuine interest versus polite deflection. We're good at this - but we needed a more systematic approach to understanding how our customers really feel, both in the moment and over time.
The Solution: AI-Powered Message Analysis
Support Analysis: Beyond Surface-Level Reading
We built a nuanced categorization system based on real support patterns we've seen. Instead of simply flagging tickets as "urgent" or "normal," we analyze the full context and tone. Here's how we break it down:
Positive Interactions:
- Upgrade Interest: When customers want to expand their usage
- General Positive Inquiries: Questions that come with a clear positive tone
Neutral Communications:
- Standard Questions: Straightforward "how-to" inquiries
- Billing Clarifications: Simple questions about payments or plans
Issues Needing Attention:
- Service Disruptions: When core functionality isn't working as expected
- Access Problems: Authentication or login issues
- Billing Concerns: Payment disputes or charges that need investigation
- Support Timeline Issues: Follow-ups on pending responses
- Product Feature Feedback: Specific functionality concerns
- Other Issues: Unique situations that need special handling
Each ticket receives two key pieces of information:
- Primary sentiment category (from the 10 options above)
- Brief explanation of why it was categorized that way
Our prompt instructs the AI to analyze support tickets using the following structure:
You are a customer support quality analyst reviewing support requests submitted to a B2B SaaS company. Please assign a sentiment category for each customer request based on the following criteria:
Sentiment Categories (select one only from the list below):
- Positive - Upgrade: The customer expresses interest in upgrading their plan or adding features.
- Positive - General Inquiry: The customer has a general question about the service, showing a positive tone.
- Neutral - General Inquiry: The customer has a straightforward question without a positive or negative tone.
- Neutral - Billing: The customer is inquiring about billing or payment details without expressing frustration.
- Negative - Service Issue: The customer indicates a problem with the service or functionality.
- Negative - Access Issue: The customer is having trouble accessing their account or logging in.
- Negative - Billing Concern: The customer expresses a concern or dissatisfaction regarding billing.
- Negative - Delayed Support: The customer mentions a delay in support or response times.
- Complaint - Product Feature: The customer expresses dissatisfaction with a specific product feature.
- Complaint - Other: The customer provides general complaints or dissatisfaction not covered above.
Important: You must select only one category from the list above. Do not create additional categories or variations. For each request, provide the chosen sentiment category first, followed by a brief explanation.
Inputs:
- Subject:
- Body:
Example Message for Analysis:
Subject: Unable to access my account
Body: “I am trying to log into my account, but I keep getting an error message. Can someone help me resolve this issue as soon as possible?”
Sentiment: Negative - Access Issue
Explanation: The message indicates a problem with accessing the account, suggesting the customer is frustrated due to login issues.
Another Example Message for Analysis:
Subject: Interested in upgrading to a higher plan
Body: “Hello, I’d like to learn more about the premium plan options and see if they’d be a better fit for our team.”
Sentiment: Positive - Upgrade
Explanation: The message reflects interest in an upgrade and shows a positive outlook toward exploring premium options.
Check out this quick demo showing how our support team uses sentiment analysis to prioritize urgent issues, spot patterns, and provide more contextual responses - all without leaving Zendesk.
Sales Response Analysis: Finding Signal in Noise
For sales emails, we focus on three critical data points:
- Relevance Check: Is this a human response versus an auto-reply? (This alone saves hours of follow-up time)
- Sentiment Analysis:
- Positive: Clear interest or forward movement ("Let's schedule that demo")
- Negative: No progress indicators or automated content
- Confidence Level: How certain we are about this classification, helping our sales team prioritize their follow-ups.
Our AI analyzes sales responses using this prompt:
Step 1: Review Email Content
- Subject:
- Description:
Carefully analyze the provided email content with a focus on the most recent response:
- Focus Area: Only review the latest reply in the thread.
- Ignore: Earlier replies unless explicitly referenced in the most recent response.
- Contextual Independence: Treat each email as a standalone entry and avoid summarizing the full thread.
Step 2: Classify Email Sentiment
Each email must be classified individually, regardless of any associated task or context.
Sentiment Categories:
- Positive Sentiment: The email demonstrates progress, interest, or a clear intent to take actionable steps.
- Examples:
- Scheduling or confirming meetings, trials, or demos.
- Expressing enthusiasm for next steps.
- Indicating readiness to move forward in the process.
- Negative Sentiment: The email shows no progress, disinterest, or contains irrelevant information.
- Examples:
- Automated replies (e.g., out-of-office notifications).
- System-generated messages (e.g., “Delivery failed”).
- Responses unrelated to the primary topic (e.g., “I’m on holiday,” “Thanks”).
- Fail-Safe Rule:
- If the content is ambiguous and does not clearly fit into Positive or Negative categories, classify as Negative Sentiment with Low Confidence.
- Provide a short rationale only if ambiguity leads to this classification.
Step 3: Assign Confidence Level
Determine the confidence level based on the clarity and directness of the sentiment:
- High Confidence: The sentiment is explicit and unambiguous.
- Examples:
- Positive: “Looking forward to trying the product.”
- Negative: “We’re not interested at this time.”
- Medium Confidence: There is some uncertainty, but a classification is reasonably clear.
- Examples:
- Positive: “I’ll check and get back to you.”
- Negative: “I’m out of office.”
- Low Confidence: The sentiment is vague or ambiguous, requiring interpretation.
- Examples:
- Responses like “Thanks” or “Got it.”
Step 4: Determine Email Type
Classify the email’s purpose based on its content:
- Book a Meeting: Explicit mention of scheduling or confirming a meeting.
- Example: “Can we meet at 3 pm tomorrow?”
- Set Up a Trial: Explicit mention of initiating or confirming a trial.
- Example: “We’d like to start a trial next week.”
- Book a Demo: Explicit mention of scheduling or confirming a demo.
- Example: “Can we schedule a demo to explore this feature?”
- None of the Above: Content unrelated to the above categories.
Output Format
- Sentiment: Positive / Negative
- Confidence Level: High / Medium / Low
- Type: Book a Meeting / Set Up a Trial / Book a Demo / None of the Above
See it in action: Watch how our sales team uses sentiment analysis to prioritize their day and never miss a hot lead. In this 3-minute demo, we'll walk you through how the system automatically categorizes responses, displays confidence scores, and integrates seamlessly with Salesforce.
Putting Sentiment Data to Work Everywhere
While our demo shows the analysis flowing into Zendesk and Salesforce, those are just the beginning. Since this runs through Census, you can send these sentiment insights to any of our 200+ destinations. This opens up powerful possibilities:
- Send it to Braze or Intercom to personalize customer communications
- Use it in Slack for real-time team notifications
- Feed it into BI tools to track sentiment trends
- Trigger different workflows based on sentiment in your customer engagement platforms
The goal is to make this sentiment data actionable wherever your team works:
- Support team sees it in their ticketing system
- Sales team gets it in their CRM
- Marketing team can use it to tailor communications
- Product team can track it in their analytics tools
This creates a more cohesive, data-driven approach to customer experience across all channels.
Real Impact
For Sales Teams
The impact extends beyond efficient inbox management. First, we can instantly identify genuine interest and prioritize follow-ups accordingly. But the real power comes from the patterns we can now see:
- Track how different outreach approaches affect response sentiment
- Identify which message types and content generate the most engagement
- Learn from our most successful interactions to refine our approach
- Test new outreach theories based on what's actually working
It's like having a constant feedback loop on our sales communication. When we see a spike in positive responses, we can investigate what changed - whether it's the message timing, content style, or value proposition. This helps us continuously refine our outreach strategy based on real data rather than assumptions.
For Support Teams
The impact works on two levels - immediate and strategic.
- Immediate: Each ticket gets tagged with its sentiment (positive/neutral/negative) and specific category, plus a brief explanation of why. This context appears right alongside the ticket in Zendesk, giving our support team immediate insight into what they're handling.
- Strategic: We can track patterns in ticket types, spot trends, and proactively address potential systemic issues before they escalate. This approach helps our operations team prioritize fixes based on actual customer pain points and track how changes to our product or processes impact support requests.
These insights help our whole team understand what's driving customer questions and concerns. When we see patterns - like an increase in questions about a specific feature or type of billing issue - we can proactively address the root cause rather than just handling tickets one by one.
Want to Learn More?
Want to learn more about implementing this for your team? Contact us to see how AI-powered sentiment analysis can transform your customer interactions.