codexkit-csat-sentiment-analyzer

# CSAT Sentiment Analyzer ## When to Use - Analyzing CSAT, NPS, app reviews, support comments, or post-interaction feedback. - Finding recurring customer pain points and service improvement themes. - Preparing support, CX, product, or leadership feedback summaries. - Comparing sentiment across segments, channels, agents, products, or time periods. ## Procedure ### Step 1 - Normalize Feedback Identify source, date range, channel, score type, segment, and any metadata. Keep raw counts separate from percentages. ### Step 2 - Classify Sentiment Use a simple sentiment label: - positive - neutral - negative - mixed - unclear Include confidence when comments are short or ambiguous. ### Step 3 - Code Themes Group feedback into themes such as speed, quality, pricing, reliability, usability, billing, support tone, missing features, or documentation. ### Step 4 - Quantify Patterns Report counts, percentages, average score, trend direction, and representative examples. Avoid claiming statistical significance without enough data. ### Step 5 - Recommend Actions Link every action to a theme and owner group: support, product, docs, billing, success, operations, or leadership. ## Inputs | Input | Required | Format | |-------|----------|--------| | Feedback dataset | Yes | Comments, scores, reviews, tickets | | Date range | Recommended | Start and end date | | Segments | Optional | Plan, region, product, channel, agent | | Scoring system | Optional | CSAT 1-5, NPS, thumbs up/down | | Business context | Optional | Launch, outage, policy change | ## Output ```markdown ## CSAT Sentiment Analysis - [Period] ### Executive Summary [Key trend, sentiment, and action] ### Score Snapshot | Metric | Value | Notes | |--------|-------|-------| ### Theme Breakdown | Theme | Sentiment | Count | Percent | Representative Comment | Recommended Action | |-------|-----------|-------|---------|------------------------|--------------------| ### Segment Differences | Segment | Pattern | Confidence | |---------|---------|------------| ### Action Plan | Owner | Action | Evidence | Priority | |-------|--------|----------|----------| ``` ## Quality Criteria - [ ] Counts and percentages reconcile to the input size. - [ ] Themes are mutually understandable and not over-fragmented. - [ ] Representative comments are short and anonymized when needed. - [ ] Recommendations are tied to evidence. - [ ] Sensitive customer data is removed or masked. ## Verification (4C) | Check | Question | |-------|----------| | **Correctness** | Do sentiment labels and theme counts match the raw feedback? | | **Completeness** | Are scores, themes, segments, evidence, and actions included? | | **Context-fit** | Are the recommendations realistic for the support or CX team? | | **Consequence** | Could a small sample or biased feedback source lead to the wrong product or staffing decision? | ## Edge Cases - **Very small sample** - Mark findings as directional and avoid percentages that imply precision. - **Sarcasm or mixed sentiment** - Use "mixed" or "unclear" instead of forcing polarity. - **Personally identifiable information** - Mask names, emails, phone numbers, and account IDs. - **Outage or one-off incident** - Separate incident-driven feedback from baseline sentiment. ## Examples > **Prompt:** "Analyze these 300 CSAT comments from April. Break down sentiment, top pain themes, and the three actions support should take next." > **Good pattern:** "Billing confusion appears in 42 of 300 comments (14%). The most common evidence is customers not understanding prorated invoices after plan changes." ## Definition of Done - [ ] Sentiment and theme counts are transparent. - [ ] Representative examples support the themes. - [ ] Actions are owner-specific and prioritized. - [ ] Bias and sample limits are documented. ## Changelog - v1.0.0 - Initial release