Turn reviews, support tickets and survey responses into voice of customer analysis your team can act on. Give your assistant redacted feedback, dates and business context; this Skill groups issues and needs, keeps representative customer quotes and helps prioritize product, messaging and service changes without losing the evidence behind them.
Customer feedback often arrives through several channels with different biases. This Skill brings it into a common review structure, connecting sentiment about specific product aspects with recurring themes, representative quotes and possible actions.
Use it when a team needs to decide which problems deserve attention across product and service work. Provide channel counts and collection context so a small or skewed sample is not mistaken for the whole customer base. The package organizes supplied feedback; it does not connect review or support platforms.
Aspect-level sentiment keeps the analysis specific: a customer may like the product but dislike its packaging or instructions. Preserve that distinction and the supporting quote so the resulting task targets the actual source of dissatisfaction.
Themes are connected to product, messaging and service actions, with prioritization factors made explicit. Keep channel bias and sample size visible when choosing what to address first, and treat proposed causes as questions to investigate rather than established facts.
Redacted reviews, support tickets, survey responses, channel counts, dates, and business context.
Use a compatible agent that can discover the installed Skill and read its supporting files. Any external service or creative runtime must be configured separately.
No feedback connectors or sentiment service are bundled. Channel bias and small samples need disclosure; prioritization factors are judgment inputs.
Send this request to the OpenClaw agent that will use the Skill. It installs the named publisher’s release and checks the workspace configuration.
Install @leooooooow/voice-of-customer version 1.1.0 in my active OpenClaw workspace. Use the original ClawHub release, not a same-named substitute. First check whether it is already installed; do not force-overwrite local changes. Run:
openclaw skills install @leooooooow/voice-of-customer --version 1.1.0
openclaw skills info voice-of-customer
Read SKILL.md and all setup references, identify required tools, dependencies and credentials, and report the actual version, path and readiness. Start a new session if needed. Do not describe package installation as a successful business-task test.This is the unchanged source SKILL.md for the reviewed version. Install the complete package to retain referenced scripts, assets, and instructions. Starter prompts below are EcomAgentTools examples.
---
name: Voice of Customer
description: Extract actionable product and experience insights from customer feedback across reviews, support tickets, surveys, and social mentions.
---
# Voice of Customer
Build a structured Voice of Customer analysis framework that transforms scattered feedback from reviews, support tickets, surveys, and social media into prioritized product improvements, messaging refinements, and customer experience optimizations — with clear methodology for categorization, sentiment scoring, and insight extraction.
## Quick Reference
| Decision | Strong | Acceptable | Weak|
|---|---|---|---|
| Data collection | Aggregates 5+ feedback channels with volume weighting | Covers reviews and support tickets | Analyzes single channel only |
| Categorization | Hierarchical taxonomy with 3 levels and cross-category tagging | Flat category list with 10+ categories | Ad hoc grouping without consistent framework |
| Sentiment analysis | Aspect-level sentiment with intensity scoring per feature/theme | Document-level positive/neutral/negative | Binary thumbs up/down |
| Insight extraction | Quantified themes with verbatim evidence, trend direction, and business impact | Theme identification with representative quotes | Vague summaries without supporting data |
| Prioritization | Scores by frequency × severity × revenue impact × effort | Ranks by mention frequency | Lists issues without prioritization |
| Action mapping | Specific recommendations tied to product, marketing, CX, and ops teams with owners | General improvement suggestions | Identifies problems without solutions |
## Solves
- Product teams guessing at what customers actually want instead of using systematic feedback analysis
- Marketing using internal language instead of actual customer vocabulary and phrases
- Support teams firefighting repeat issues that could be prevented with upstream fixes
- Review responses that are generic and miss the specific concern raised
- Customer churn caused by unaddressed experience friction points
- New product launches that miss critical requirements customers have already articulated
- Competitor positioning gaps that customers openly discuss but nobody tracks
## Workflow
### Step 1 — Define Scope and Objectives
Establish what you are analyzing and why:
- Product or product line to analyze
- Time period (last 30/60/90 days, or specific launch window)
- Feedback channels to include (reviews, support, surveys, social, community forums)
- Primary questions to answer (product issues, feature requests, competitive comparison, messaging fit)
- Stakeholder audience (product, marketing, CX, leadership)
### Step 2 — Collect and Normalize Feedback
Gather feedback from all channels into a unified format:
- Product reviews (Amazon, Shopify, app stores, marketplace listings)
- Support tickets and live chat transcripts (categorized by resolution)
- Post-purchase survveys (NPS, CSAT, open-ended responses)
- Social media mentions (brand mentions, hashtags, competitor comparisons)
- Community forums and Reddit threads
- Normalize: assign source, date, rating if available, and customer segment
### Step 3 — Build Categorization Taxonomy
Create a hierarchical theme structure:
- Level 1: Major categories (Product Quality, Shipping/Delivery, Customer Service, Pricing/Value, Usability)
- Level 2: Sub-categories (e.g., Product Quality → Durability, Sizing/Fit, Material, Color Accuracy)
- Level 3: Specific attributes (e.g., Sizing/Fit → Runs Small, Runs Large, Inconsistent Between Styles)
- Allow multi-tagging — one piece of feedback can touch multiple themes
### Step 4 — Analyze Sentiment and Extract Themes
For each feedback item:
- Assign aspect-level sentiment (a review can be positive about quality but negative about shipping)
- Score sentiment intensity (1-5 scale: strongly negative to strongly positive)
- Extract verbatim quotes that best represent each theme
- Identify emotional triggers (frustration language, delight language, comparison language)
- Flag competitive mentions and switching intent signals
### Step 5 — Quantify and Prioritize Findings
Turn qualitative themes into quantified, ranked insights:
- Frequency: What percentage of feedback mentions this theme?
- Sentiment: What is the average sentiment score for this theme?
- Trend: Is this theme increasing, stable, or decreasing over time?
- Severity: How much does this impact purchase decisions, satisfaction, or retention?
- Revenue impact: Estimate effect on conversion, repeat purchase rate, or AOV
- Create a priority matrix: frequency × severity × revenue impact
### Step 6 — Map Insights to Actions
Convert each prioritized insight into specific recommendations:
- Product team: Feature fixes, quality improvements, new feature development
- Marketing team: Messaging adjustments, ad copy using customer language, review response templates
- CX team: Process improvements, FAQ updates, proactive communication triggers
- Operations: Shipping, packaging, fulfillment improvements
- Assign owners, timelines, and success metrics for each action
### Step 7 — Build Monitoring Framework
Establish ongoing VoC tracking:
- Define KPIs to track (NPS trend, theme frequency shifts, sentiment scores)
- Set alert thresholds (e.g., if "defective" mentions exceed 5% of reviews)
- Create a review cadence (weekly pulse, monthly deep dive, quarterly strategic review)
- Build feedback loop: track whether implemented changes move the targeted metrics
## Examples
### Example 1: DTC Skincare Brand — Product Line Review
**Context:** DTC skincare brand with 12 SKUs, selling primarily through own website and Amazon. 1,200 product reviews in last 90 days, 340 support tickets, 180 survey responses. NPS dropped from 52 to 41 last quarter.
**Step 1 — Scope:**
- All 12 SKUs, last 90 days
- Channels: Shopify reviews, Amazon reviews, Zendesk tickets, post-purchase survey
- Primary question: Why did NPS drop? What product and experience issues are driving dissatisfaction?
**Step 2 — Data Collected:**
| Source | Volume | Avg Rating |
|---|---|---|
| Shopify reviews | 380 | 4.1/5 |
| Amazon reviews | 620 | 3.8/5 |
| Support tickets | 340 | — |
| Post-purchase survey | 180 | NPS 41 |
| Instagram mentions | 95 | — |
| **Total** | **1,615** | — |
**Step 3 — Taxonomy Built:**
| L1 Category | L2 Sub-categories |
|---|---|
| Product Efficacy | Results timeline, Effectiveness, Skin reaction, Ingredient concerns |
| Product Quality | Texture/Consistency, Fragrance, Packaging integrity, Shelf life |
| Shipping & Delivery | Speed, Packaging damage, Tracking accuracy |
| Pricing & Value | Price vs. quantity, Subscription value, Discount expectations |
| Customer Service | Response time, Resolution quality, Return process |
**Step 4 — Top Themes by Frequency:**
| Theme | Frequency | Sentiment | Trend | Verbatim Example |
|---|---|---|---|---|
| New formula dissatisfaction | 18% | 1.8/5 | ↑ Increasing | "The new version of the serum feels completely different and broke me out" |
| Shipping damage | 12% | 2.1/5 | ↑ Increasing | "Bottle arrived cracked and leaked all over the box" |
| Results exceeded expectations | 11% | 4.8/5 | Stable | "After 3 weeks my dark spots have noticeably faded" |
| Subscription flexibility | 9% | 2.4/5 | Stable | "I can't skip a month without calling support" |
| Price increase frustration | 8% | 1.9/5 | ↑ New | "Went up $8 with no warning or explanation" |
**Step 5 — Priority Matrix:**
| Insight | Frequency | Severity | Revenue Impact | Priority Score |
|---|---|---|---|---|
| Formula change backlash | 18% | High | High (retention) | **Critical** |
| Shipping damage | 12% | High | Medium (returns) | **High** |
| Subscription UX friction | 9% | Medium | High (churn) | **High** |
| Price increase communication | 8% | Medium | Medium (perception) | **Medium** |
**Step 6 — Action Map:**
| Insight | Team | Action | Timeline |
|---|---|---|---|
| Formula backlash | Product | Bring back original formula as "Classic" option; A/B test new formula with subset | 2 weeks |
| Shipping damage | Ops | Switch to reinforced mailer boxes; add inner padding for glass bottles | 1 week |
| Subscription UX | Engineering | Add self-service skip/pause in account portal | 3 weeks |
| Price communication | Marketing | Email campaign explaining ingredient upgrade behind price change | 1 week |
---
### Example 2: B2B SaaS — Feature Request Analysis
**Context:** Project management SaaS tool with 2,000 active accounts. Analyzing feedback to inform Q3 product roadmap. Sources: in-app feedback widget, Intercom chats, G2/Capterra reviews, churned customer exit surveys.
**Step 1 — Scope:**
- Full platform, last 6 months
- Channels: In-app feedback (890), Intercom (1,200 relevant threads), G2/Capterra (210 reviews), Exit surveys (45)
- Primary question: What features should we build in Q3 to reduce churn and win competitive deals?
**Step 2 — Data Collected:**
| Source | Volume | Focus |
|---|---|---|
| In-app feedback widget | 890 | Feature requests, bug reports |
| Intercom conversations | 1,200 | Support questions, workaround requests |
| G2/Capterra reviews | 210 | Comparative analysis, strengths/weaknesses |
| Exit surveys | 45 | Churn reasons, competitor mentions |
| **Total** | **2,345** | — |
**Step 4 — Top Themes:**
| Theme | Frequency | Source Concentration | Competitor Mention |
|---|---|---|---|
| Time tracking integration | 22% | In-app (35%), G2 (28%) | "Asana and Monday both have this" |
| Custom reporting/dashboards | 18% | In-app (24%), Exit (40%) | "Switched to Monday for reporting" |
| Mobile app performance | 14% | Intercom (22%), G2 (15%) | "Mobile app crashes daily" |
| API and Zapier improvements | 11% | In-app (15%), Intercom (10%) | "Their API docs are terrible" |
| Guest/client collaboration | 9% | Intercom (12%), Exit (20%) | "Basecamp handles client access better" |
**Step 5 — Prioritization with Revenue Impact:**
| Feature | Request Freq | Churn Driver | Competitive Gap | Dev Effort | Priority |
|---|---|---|---|---|---|
| Custom reporting | 18% | #1 exit reason | Yes (Monday) | Large (8 weeks) | **Critical — Q3 P0** |
| Time tracking | 22% | Mentioned in 15% of exits | Yes (Asana, Monday) | Medium (4 weeks) | **High — Q3 P1** |
| Mobile stability | 14% | Low direct churn but high frustration | Parity issue | Medium (3 weeks) | **High — Q3 P1** |
| Guest collaboration | 9% | #2 exit reason by revenue | Yes (Basecamp) | Large (6 weeks) | **Medium — Q4** |
**Step 6 — Action Map:**
| Insight | Team | Action | Success Metric |
|---|---|---|---|
| Custom reporting gap | Product + Eng | Build dashboard builder with 5 default templates and export | Reduce "reporting" churn reason by 50% |
| Time tracking need | Product | Native timer + Toggl/Harvest integration | 30% adoption within 60 days of launch |
| Mobile crashes | Engineering | Performance sprint — crash-free rate from 94% to 99.5% | App store rating from 3.2 to 4.0+ |
| Client access | Product | Design spec for Q4 — interview 10 churned accounts for requirements | Validated spec ready by end of Q3 |
## Common Mistakes
1. **Analyzing only one feedback channel** — Reviews skew toward extremes, support tickets skew toward problems. You need all channels for a balanced picture. Weight by volume but include every source.
2. **Using document-level sentiment only** — A 3-star review that praises the product but hates the shipping is not "neutral." Aspect-level sentiment captures what is actually working and what is not.
3. **Reporting themes without quantification** — Saying "customers complain about shipping" is not actionable. Saying "17% of feedback mentions shipping damage, up from 8% last quarter, concentrated on glass bottle SKUs" is actionable.
4. **Ignoring positive feedback** — VoC is not just about problems. Positive themes tell you what to protect and amplify in marketing. If 25% of reviews mention "gentle on sensitive skin," that is a messaging goldmine.
5. **Letting recency bias drive priorities** — The loudest recent complaint is not always the most important issue. Use frequency, severity, and trend data together to avoid whiplash in priorities.
6. **Creating a report nobody acts on** — Every insight needs an owner, a timeline, and a success metric. A VoC analysis that sits in a shared drive is wasted effort.
7. **Treating VoC as a one-time project** — Customer sentiment changes constantly. Build ongoing monitoring, not just periodic reports. Set alerts for emerging themes.
8. **Missing competitive intelligence** — Customers voluntarily compare you to competitors in reviews and exit surveys. This is free competitive intelligence. Track competitor mentions systematically.
## Resources
- [Output Template](references/output-template.md) — Structured VoC analysis deliverable format
- [Sentiment Scoring Guide](references/sentiment-scoring-guide.md) — Aspect-level sentiment methodology and calibration
- [Feedback Taxonomy Reference](references/feedback-taxonomy-reference.md) — Category frameworks for common ecommerce and SaaS verticals
- [Quality Checklist](assets/quality-checklist.md) — Comprehensive review checklist for VoC analysis deliverables
Starter prompts for the main use cases—copy and use them directly.
Group recurring product complaints with source quotes Use for: Group recurring product complaints with source quotes. Prepare: Redacted reviews, support tickets, survey responses, channel counts, dates, and business context.
Use the installed voice-of-customer Skill. Task: Group recurring product complaints with source quotes. Inputs: [INPUT]. Constraints: [CONSTRAINTS]. Check missing information first. Produce a reviewable result with evidence, assumptions, and next steps. No feedback connectors or sentiment service are bundled. Channel bias and small samples need disclosure; prioritization factors are judgment inputs.[INPUT][CONSTRAINTS]Extract customer language for a messaging brief Use for: Extract customer language for a messaging brief. Prepare: Redacted reviews, support tickets, survey responses, channel counts, dates, and business context.
Use the installed voice-of-customer Skill. Task: Extract customer language for a messaging brief. Inputs: [INPUT]. Constraints: [CONSTRAINTS]. Check missing information first. Produce a reviewable result with evidence, assumptions, and next steps. No feedback connectors or sentiment service are bundled. Channel bias and small samples need disclosure; prioritization factors are judgment inputs.[INPUT][CONSTRAINTS]Prioritize experience fixes with owners and validation steps Use for: Prioritize experience fixes with owners and validation steps. Prepare: Redacted reviews, support tickets, survey responses, channel counts, dates, and business context.
Use the installed voice-of-customer Skill. Task: Prioritize experience fixes with owners and validation steps. Inputs: [INPUT]. Constraints: [CONSTRAINTS]. Check missing information first. Produce a reviewable result with evidence, assumptions, and next steps. No feedback connectors or sentiment service are bundled. Channel bias and small samples need disclosure; prioritization factors are judgment inputs.[INPUT][CONSTRAINTS]120 synthetic reviews, support messages, survey responses, social posts and forum entries contain 24 records per channel across two periods. The merchant needs priorities across quality, shipping and copy.
Case input files: business-inputs.zip
Use the voice-of-customer Skill.
Combine the supplied support and feedback records into a voice-of-customer report. Deduplicate customers and repeated events, tag hierarchical themes and sentiment, retain evidence IDs, and compare frequency by channel and period. Recommend distinct actions for product quality, customer support and product-page content.
The attached business materials are:
- inputs/brief.json
- inputs/merchant.json
- inputs/methodology.md
Deliver a complete result.md report that the operations team can read, plus any finished files needed for this task. Keep supporting calculations and evidence with the deliverables. Write the report in English; use the requested target language for customer-facing copy. The business datasets are fictional examples unless the brief explicitly names a live public website.The primary file produced by this prompt is shown below in its original language.
Prioritize lid leakage investigation, while support and content teams remove avoidable uncertainty about dispatch, care and returns. Leakage is the most severe signal and doubles from 8.3% to 16.7% between the supplied periods. Customer support receives praise for timely replies; the return questions do not establish slow approval or poor service.
This analysis uses synthetic feedback and merchant facts; figures describe the supplied examples and are not evidence of real market prevalence or business results.
Analyze HarborMug HM-BLUE-001 for product, support and content operations. The previous snapshot is October 21, 2026 (60 records); the current snapshot is October 28, 2026 (60). These are two observed dates, not continuous weekly coverage. Questions: which issues threaten repeat purchase, which uncertainty can support resolve, and which facts must appear before purchase?
All 120 records represent distinct customers and retained events; zero duplicates were removed. Customer IDs were checked across channels, and customer/date/text combinations checked for repeated events. There are no order or incident linkage fields, so semantic incident matching beyond those supplied identities cannot be established. Repeated scripted wording across different customers is retained. Frequency means distinct retained records mentioning a theme divided by records in that channel/period, with each customer counted once per theme. Multi-tagging yields 121 aspect observations. Never sum theme percentages as exclusive shares.
All records fall within 30 days of the supplied analysis date and receive weight 2. Channel weights follow observed volume: five channels each contribute 24/120 = 20%. Within each period they each contribute 12/60 = 20%; no channel-mix correction is needed. No population weights or significance tests are justified by scripted, non-independent examples. The two periods contain different customers, not a longitudinal panel.
| Channel | Previous | Current | Total | Rating coverage and mean |
|---|---|---|---|---|
| Reviews | 12 | 12 | 24 | 24/24; 2.62/5 |
| Support | 12 | 12 | 24 | None; not inferred |
| Survey | 12 | 12 | 24 | None; not inferred |
| Social | 12 | 12 | 24 | None; not inferred |
| Forum | 12 | 12 | 24 | None; not inferred |
Reviews may emphasize extremes, support may emphasize problems, and surveys/social/forums reflect different participation choices. Equal sample sizes here do not establish equal real-world customer reach. No NPS question, response population, resolution times or defect inspection results are supplied; do not infer NPS, SLA performance or a manufacturing defect rate.
Sentiment uses 1 strongly negative, 2 negative, 3 neutral, 4 positive, 5 strongly positive. Text governs aspect scoring rather than star rating. Supplied human-coded aspects were retained after text checks; this is not a new independent multi-rater coding exercise. The initial 20 records were reviewed for calibration, including mixed feedback F009. The neutral return question indicates uncertainty, not dissatisfaction. F009 retains thermal=5 and delivery=2 despite a five-star rating. “I won't buy again” supports leakage=1; concern/disappointment supports 2; satisfaction supports 4; love plus repurchase supports 5. No sarcasm was detected.
| Theme | Total /120 | Mean /5 | Previous /60 | Current /60 | Change (pp) | Evidence example |
|---|---|---|---|---|---|---|
| Leakage / no repurchase | 15 (12.5%) | 1.0 | 5 (8.3%) | 10 (16.7%) | +8.3 | F001 |
| Dispatch expectations | 11 (9.2%) | 2.0 | 6 (10.0%) | 5 (8.3%) | -1.7 | F003 |
| Arrival damage | 10 (8.3%) | 2.0 | 5 (8.3%) | 5 (8.3%) | +0.0 | F002 |
| Tracking uncertainty | 10 (8.3%) | 2.0 | 5 (8.3%) | 5 (8.3%) | +0.0 | F004 |
| Shipping cost surprise | 10 (8.3%) | 2.0 | 5 (8.3%) | 5 (8.3%) | +0.0 | F005 |
| Care disclosure | 10 (8.3%) | 2.0 | 5 (8.3%) | 5 (8.3%) | +0.0 | F006 |
| Return instructions | 10 (8.3%) | 3.0 | 5 (8.3%) | 5 (8.3%) | +0.0 | F007 |
| Timely support | 10 (8.3%) | 4.0 | 5 (8.3%) | 5 (8.3%) | +0.0 | F008 |
| Thermal satisfaction | 10 (8.3%) | 5.0 | 5 (8.3%) | 5 (8.3%) | +0.0 | F009 |
| Fair price | 10 (8.3%) | 4.0 | 5 (8.3%) | 5 (8.3%) | +0.0 | F011 |
| Lid choice / switching | 10 (8.3%) | 2.0 | 5 (8.3%) | 5 (8.3%) | +0.0 | F012 |
| Color accuracy | 5 (4.2%) | 4.0 | 5 (8.3%) | 0 (0.0%) | -8.3 | F010 |
No theme is new in the current period. Leakage frequency rises; color praise disappears from the current sample, which does not establish a color problem. Delivery falls by one mention because only the previous sample contains the extra mixed review. For themes present in both periods, sentiment means remain unchanged (0.0 point change); color has no current mean. No theme exceeds the guide’s 0.5-point shift threshold.
Each cell is mentions /24 (percent). Full channel-by-period counts, each with denominator 12, are in channel_period_counts.csv.
| Theme | Reviews | Support | Survey | Social | Forum |
|---|---|---|---|---|---|
| Leakage / no repurchase | 3 (12.5%) | 3 (12.5%) | 3 (12.5%) | 3 (12.5%) | 3 (12.5%) |
| Arrival damage | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) |
| Dispatch expectations | 3 (12.5%) | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) |
| Tracking uncertainty | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) |
| Shipping cost surprise | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) |
| Care disclosure | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) |
| Return instructions | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) |
| Timely support | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) |
| Thermal satisfaction | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) |
| Color accuracy | 1 (4.2%) | 1 (4.2%) | 1 (4.2%) | 1 (4.2%) | 1 (4.2%) |
| Fair price | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) |
| Lid choice / switching | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) | 2 (8.3%) |
In every channel, leakage changes from 1/12 (8.3%) to 2/12 (16.7%) and color from 1/12 (8.3%) to 0/12. All other themes stay at 1/12 in each channel/period, except review delivery: 2/12 to 1/12. This broad channel repetition is a property of the supplied example, not independent corroboration.
Distribution below is across 121 aspects, not 120 star ratings or customer-level overall scores.
Previous: 61 aspects; weighted mean 2.738/5. Current: 60 aspects; weighted mean 2.500/5. The aggregate movement reflects theme mix, not worsening within-theme intensity. Overall aspect mean is 2.620/5.
Quotes are supplied Chinese verbatim; English glosses explain their meaning. These are evidence, not publishable testimonials.
| Theme / evidence | Verbatim | English interpretation / emotional trigger |
|---|---|---|
| Leakage / no repurchase / F001 | 漏得到处都是，我不会再买。 | Leaks everywhere; won't buy again. Explicit rejection and retention risk. |
| Arrival damage / F002 | 杯子到货有凹痕，很失望。 | Arrived dented; disappointed. Arrival condition matters. |
| Dispatch expectations / F003 | 等得很烦，页面没说发货还要两天。 | Frustrated waiting; page omitted two-day dispatch. Expectation mismatch. |
| Tracking uncertainty / F004 | 物流两天没更新，我有点担心。 | No tracking update for two days; worried. Anxiety. |
| Shipping cost surprise / F005 | 24美元杯子还要5美元运费，有些失望。 | $24 cup plus $5 shipping; disappointed. Total-price surprise. |
| Care disclosure / F006 | 买完才看到只能手洗，很失望。 | Only discovered hand-wash restriction after purchase; disappointed. Disclosure gap. |
| Return instructions / F007 | 未开封想退货，下一步怎么做？ | Unopened return: what next? Neutral request for instructions. |
| Timely support / F008 | Mia回复很及时。 | Reply was timely. Preserve responsiveness. |
| Thermal satisfaction / F009 | 我爱这只杯子的保温体验，还会买。 等得很烦，页面没说发货还要两天。 | Love thermal experience; will buy again. Delight and repeat intent. |
| Color accuracy / F010 | 蓝色和照片一样，很满意。 | Blue matches photos; satisfied. Visual accuracy. |
| Fair price / F011 | 24美元这个价格我觉得合理。 | $24 feels reasonable. Price acceptance. |
| Lid choice / switching / F012 | 我正考虑换去FictionCup，因为想要另一种杯盖。 | Considering FictionCup for another lid type. Explicit switching consideration. |
Use “保温体验” (thermal experience), “只能手洗” (hand-wash only), and “发货还要两天” (two days before dispatch) rather than vague performance, maintenance or fulfillment language. Positive copy can discuss thermal experience and blue color without promising tested hours or universal color fidelity.
FictionCup appears in 10/120 records (8.3%), five per period, one per channel per period. All ten express switching consideration over lid choice; none confirms a completed switch or verified competitor superiority. They represent 10/76 negative-bearing records (13.2%). Leakage adds 15 explicit no-repurchase signals; thermal praise adds 10 positive repurchase signals. Do not combine these into a churn forecast. Protect thermal performance, timely replies and price-value perception while validating demand for another lid.
Scores are decision assumptions, not measured financial outcomes. Frequency is percent of all 120 records; severity (1–5) reflects functional failure and stated intent, revenue risk (1–5) reflects plausible retention/conversion exposure, and effort (1 small, 2 medium, 3 larger) is provisional. Risk = frequency × severity × revenue risk; delivery score = risk / effort. Effort accelerates quick wins without displacing leakage as the highest underlying risk. No stakeholder approval or implementation has occurred.
| Action theme | Frequency % | Severity | Revenue risk | Effort | Risk | Delivery score |
|---|---|---|---|---|---|---|
| Leakage / no repurchase | 12.5 | 5.0 | 5.0 | 3.0 | 312.5 | 104.2 |
| Dispatch expectations | 9.2 | 3.0 | 3.0 | 1.0 | 82.5 | 82.5 |
| Care disclosure | 8.3 | 3.0 | 3.0 | 1.0 | 75.0 | 75.0 |
| Arrival damage | 8.3 | 4.0 | 4.0 | 2.0 | 133.3 | 66.7 |
| Shipping cost surprise | 8.3 | 2.0 | 3.0 | 1.0 | 50.0 | 50.0 |
| Tracking uncertainty | 8.3 | 2.0 | 2.0 | 1.0 | 33.3 | 33.3 |
| Return instructions | 8.3 | 2.0 | 2.0 | 1.0 | 33.3 | 33.3 |
| Lid choice / switching | 8.3 | 3.0 | 4.0 | 3.0 | 100.0 | 33.3 |
Leakage severity=5 and revenue risk=5 follow functional failure plus explicit no-repurchase statements. Arrival damage scores 4/4 because it can require replacement; dispatch and care 3/3 because purchase expectations are affected; shipping surprise 2/3 reflects conversion friction; tracking and neutral returns 2/2 reflect information needs; lid choice 3/4 reflects explicit alternative consideration with unvalidated development effort.
At $24 price and $8 unit cost, product-only contribution is $16 before shipping, refunds, fees and advertising. A single-unit domestic order below $50 collects $5 shipping and incurs $4 shipping cost, giving $17 before those other costs. A one-unit replacement would consume at least $12 in product plus outbound shipping, excluding handling. This is unit economics context, not a savings estimate: feedback cannot be mapped to orders, replacements or refunds. Do not reduce price merely because shipping is disliked; price fairness is praised in 10 records and the problem can first be addressed through disclosure. Weekly merchant totals cannot establish which themes drove refunds or sales.
Targets below are proposed operating goals, not achieved results. Assign role-based owners where the merchant supplies no named product/content lead.
| Timing | Workstream and owner | Concrete action and evidence | Acceptance / success measure |
|---|---|---|---|
| 0–2 weeks | Product quality lead; Alex coordinates escalation | Investigate lid seals, fit and assembly across batches; log lid/batch identifiers and reproduce leaks under intended use. Keep arrival dents separate from leakage (F001, F002, F070). Establish approved replacement criteria. | Every new leak case records lid/batch and use conditions; document reproduction and disposition before declaring a root cause. Goal: leak feedback share below 10% for four consecutive adequately sampled weeks. |
| 0–2 weeks | Support: Mia; approvals/escalation: Alex | Adopt separate leak/damage, tracking and unused-return triage. Collect relevant photos or details, avoid repeated requests, assign the next owner, and seek merchant approval for final refunds (F004, F007, F008). | 100% of return cases have eligibility check and approval status; 90% of cases get a clear next step within one business day, measured prospectively. No claim that current reply speed is poor. |
| 0–2 weeks | Product-page content lead | Put dispatch, transit, US shipping fee/threshold, hand washing and microwave restriction beside purchase controls; add return steps and the drafts below (F003, F005, F006, F007). | Check all five facts on desktop/mobile before release. Goal: dispatch/care confusion each below 5% of feedback over four adequately sampled weeks. |
| 2–8 weeks | Fulfillment/packaging lead | Inspect dent locations and packing protection; trial protection changes with controlled shipment checks (F002). | Record damage cases per delivered order; reduce from a newly measured baseline. Do not presume packaging is the proven cause. |
| 2–8 weeks | Product quality lead | Implement and verify the validated lid correction; compare matched batches/use conditions. Preserve thermal performance (F009). | Agreed leak-test acceptance standard and documented results before production release; track thermal complaints as a guardrail. |
| 2–8 weeks | Mia and fulfillment lead | Define no-scan investigation trigger, carrier check and proactive update workflow; preserve timely responses (F004, F008). | Review tracking stagnant for two business days; communicate known facts and next check, without guaranteeing delivery. Measure reopen and repeat-contact rates. |
| 8+ weeks | Product lead; Alex sponsors | Validate alternate-lid demand and compatibility before tooling, using switching themes (F012). | Approved requirements, compatibility and cost review before a build decision; do not promise a new lid publicly. |
Chinese is used for customer-facing drafts to match the supplied customer language; no separate target-language field was supplied. These are ready for merchant review and placement, with no claim that they have been sent or published.
HarborMug 蓝色杯｜24美元
下单前请确认：本商品仅支持美国境内配送。订单将在2个工作日内发货，运输通常需要3–5个工作日；发货处理时间与运输时间分开计算。
订单金额低于50美元，运费为5美元；订单金额满50美元免运费。单买一只24美元的杯子，商品与运费合计29美元（如有适用税费，另计）。
清洁与使用：仅限手洗，不可放入微波炉。
**退货说明：**未使用的商品可在30天内申请退货。请先联系客服说明商品状态与退货需求；最终退款须经商家审核批准。请先获取退货步骤，再安排寄回。
您好，感谢您告诉我们这个情况。很抱歉杯子出现渗漏或到货损坏，影响了您的使用。请说明具体情况，并提供能看清杯盖渗漏位置或杯身损坏的照片；如已提供，无需重复发送。若方便，也请补充杯盖或包装上的批次标识，帮助我们排查。我们会根据具体情况确认处理方案，并告知您下一步；如涉及退款，最终须经商家审核批准。
您好，未使用的商品可在30天内申请退货。请告知收货日期及商品目前是否开封、使用，我们会核对申请条件，并向您说明退货步骤。请先不要自行寄回；最终退款须经商家审核批准。
您好，理解物流暂未更新会让您担心。订单发货处理时间为2个工作日，运输通常需要3–5个工作日。我们会核对订单是否已发货以及最新物流记录，并向您说明当前已确认的状态和下一步查询安排。物流未更新本身并不代表包裹已经丢失。
很抱歉渗漏影响了您的使用体验。请通过客服渠道联系我们，并说明杯盖渗漏的具体位置；我们会协助核查并确认下一步处理方案。请勿在公开评论中提供订单或联系方式。
Mia maintains a weekly pulse; product quality and content owners review monthly; Alex reviews priorities quarterly. Track unique customers/events, source/date/segment, three-level theme, aspect score, evidence ID, batch if available, resolution/approval status and action version. Link repeated contacts to an incident ID; retain all evidence while counting one incident per theme, with cross-channel attribution shown separately.
Dashboard: theme count and denominator by channel/period, weighted and unweighted aspect sentiment/distribution, missing-data rates, duplicate rate, and unique customer reach. Separately measure inspected defects per inspected unit, damage per delivered order, first reply time, resolution time, repeated contact and approval turnaround. Feedback prevalence must never stand in for order defect rate.
Proposed alerts: any leak report with damage beyond the cup escalates immediately to Alex; leak share ≥10% triggers weekly quality review (current sample 16.7%); dispatch/care confusion ≥10% or +5 percentage points triggers content review. Require at least 30 feedback records per evaluated window, otherwise show counts and extend the window. Flag theme sentiment drops >0.5 points and any new explicit switching signal. The supplied 12-record channel windows are descriptive, too small for threshold-driven channel conclusions.
For each change, retain a dated change log and compare at least four weeks before and after with channel mix, batch and order volume shown. Targets are hypotheses: confirm reduced confusion without lower satisfaction or additional contacts; investigate concurrent changes before attributing improvement. Maintain a 5% monthly text/scoring spot check (at least six of a 120-record collection); obtain independent calibration when a second reviewer is available.
normalized_feedback.csv: all 120 retained records with source, date, rating, segment and original text.coded_evidence.csv: 121 aspect rows, three-level English taxonomy, original Chinese paths, sentiment, intent, lifecycle and business signals; evidence IDs preserve traceability.deduplication.csv: retain/merge decision for every supplied record.theme_summary.csv: counts, denominators, weighted-equivalent means, trends and full evidence IDs.channel_period_counts.csv: every theme × channel × total/period combination, including zeros.priority_scores.csv: editable scoring assumptions and calculated rankings.verification.json: arithmetic and coverage checks.The coded evidence file supplies all 12 leaf definitions through their three-level paths (11 distinct level-two categories). No records are uncategorized. The methodology mentions a separate coding file not included in the materials; this report uses the aspect codes embedded in the supplied records and checks them against text. Root causes, actual customer churn, competitor performance and realized revenue effects remain unmeasured; those gaps make investigation and validated trials preferable to an immediate redesign or price change.
Merchant business data is synthetic. Website observations, when present, come from the public URLs listed in the report.
No feedback connectors or sentiment service are bundled. Channel bias and small samples need disclosure; prioritization factors are judgment inputs.
Source and package files reviewed: ClawHub 1.1.0. This is a static review, not an installation or business-outcome test.
Use authorized inputs and keep secrets out of prompts. Review customer-facing claims, calculations, code changes, and any proposed external action before applying them.
Organize customer feedback into themes, evidence, priorities, and team actions. The framework connects aspect-level sentiment and representative quotes to product, messaging, and service improvements.
Choose the method for your agent and install the complete package, not just the displayed SKILL.md. Redacted reviews, support tickets, survey responses, channel counts, dates, and business context. Start a new agent session and explicitly ask to use voice-of-customer. Try one use case below with a small sample and review the output.
Group recurring product complaints with source quotes; Extract customer language for a messaging brief; Prioritize experience fixes with owners and validation steps
No feedback connectors or sentiment service are bundled. Channel bias and small samples need disclosure; prioritization factors are judgment inputs.
Source and package files reviewed: ClawHub 1.1.0. This is a static review, not an installation or business-outcome test.
Source and package files reviewed: ClawHub 1.1.0. This is a static review, not an installation or business-outcome test.
Source and package files reviewed: ClawHub 1.1.0. This is a static review, not an installation or business-outcome test.
No reviews yet.