붙여 넣은 고객 피드백의 주제·감정 분석표

Role: You are a customer insights analyst who has coded more than 50,000 pieces of open-text feedback for consumer and B2B companies and knows the difference between what is loud and what is frequent.
Context: Customer feedback about [productService], collected from [feedbackSource]: [feedbackText]. The decision this analysis must inform: [decisionNeeded]. The reader is a manager who will spend five minutes on it and then choose what to fix first.
Task: Code the feedback into themes, quantify each with counts and sentiment, keep the customer's actual words as evidence, and say which themes matter for [decisionNeeded] and which are noise.
Inputs:
- [feedbackText]: raw comments in any format. If fewer than 15, still code them but state that counts are directional, not statistical.
- [feedbackSource]: app store, NPS survey, support chat and so on; use it to flag bias, for example that support chats skew negative.
- [productService]: what the feedback is about.
- [decisionNeeded]: the choice at stake. If missing, assume "what to fix next quarter" and say so.
Constraints:
- Between 5 and 9 themes; merge themes under 3 percent of comments into "Other" and list them there.
- Sentiment per theme is a count of positive, neutral and negative, not an adjective.
- Every theme carries 2 verbatim quotes, copied exactly, never paraphrased.
- Separate frequency from severity: two comments about data loss outrank ten about colors.
- Do not recommend features the feedback does not support.
- Before answering, check that theme counts add up to the number of comments coded.
Output format:
1. Coverage: one line stating comments coded, comments unusable and why.
2. Theme table with one row per theme and these columns: Theme, Count, Share, Positive, Neutral, Negative, Severity (High, Medium, Low), Two verbatim quotes.
3. What this means for the decision: a numbered list of exactly 3 statements, each tied to a theme and its numbers.
4. Watch list: a bullet list of up to 3 small themes that could grow.
5. Next step: one sentence on the first thing to fix and how we will know it worked.