voice-of-customer-miner
About
This skill mines public reviews and forums to extract unmet customer needs and competitor weaknesses, providing quoted evidence. It's used to gather customer insights without interviews, feeding into discovery tools like JTBD canvases. Developers can leverage it to generate evidence-backed hypotheses for competitive intelligence and product discovery.
Quick Install
Claude Code
Recommendednpx skills add deanpeters/Product-Manager-Skills -a claude-code/plugin add https://github.com/deanpeters/Product-Manager-Skillsgit clone https://github.com/deanpeters/Product-Manager-Skills.git ~/.claude/skills/voice-of-customer-minerCopy and paste this command in Claude Code to install this skill
Documentation
Voice-of-Customer Miner
Purpose
Mine public customer voice — review sites, app stores, Reddit and practitioner forums, community boards — for unmet needs, competitor weaknesses, and switching triggers: search plan → source sweep → verbatim capture → need themes → so what → next-step options. This bridges competitive intelligence and discovery: it delivers customers' exact words without waiting on an interview cycle. But public voice skews toward the angry and the vocal, so every theme it surfaces is a hypothesis to validate, never a verdict — the output's last stop is always a real conversation.
Input
Works best with: the product(s) or competitor(s) to mine — yours, a rival's, or a set — and the decision this should inform. Also useful: a theme to focus on (onboarding, pricing, reliability) if you have one; otherwise the sweep runs open.
Input supplied inline with the invocation — text after the skill name, a pasted context dump, or an
appended ARGUMENTS: line — counts as answers already given. Use it against the question budget;
don't re-ask.
Arriving empty-handed? That works too. The skill opens with at most 3 questions (whose voice, what decision, theme or open sweep) and proceeds on labeled assumptions if they go unanswered.
Example invocation: Mine voice-of-customer for [Competitor A] and [Competitor B], focus on onboarding — informs whether our Q1 bet is a migration tool.
Key Concepts
- Governing protocol: honors the
autonomous-investigationcontract — question budget of 3, search-plan gate, Fact/Inference/Assumption labels, Just Enough Mode, stable schema, 4-option Final Step. Discipline: OSINT's review-and-community layer (seeintelligence-collection-disciplines). - Theme by need, not by feature. "Exports are broken" is a feature complaint; "I can't get my data where my team works" is the underlying need. Theming by need is the same solution-free discipline as JTBD and painstorming — and it's what makes themes portable into discovery.
- Verbatims are the product. Short, real, quoted customer language with URLs. Verbatims teach persona language: the exact words customers use become interview probes and positioning copy. Never fabricate quotes, ratings, review counts, or reviewer roles.
- Every source has a known skew. Reviewers skew negative; vendor communities skew loyal; app stores over-represent update anger. Note the bias per source — public voice is evidence with a known skew, not ground truth.
- Honest frequency. Recurring across sources ≠ concentrated in one thread ≠ isolated but vivid. Say which; one articulate ranter is not a theme.
- When NOT to use: no meaningful public footprint (early-stage, niche enterprise) → run
discovery-interview-prepinstead; you need your users' voice on a private area → mine your own tickets and research; statistical confidence required → this is qualitative theming.
Application
- Credit inline context, then ask only the unanswered questions (max 3):
- Whose customer voice — yours, a competitor's, or a set?
- What decision should this inform?
- Any specific theme to focus on, or open sweep?
- Show the 3-bullet search plan — which voice sources you'll sweep, how you'll select representative verbatims, how observation will be separated from interpretation. Continue unless revised.
- Sweep mixed voice sources — review sites (G2, Capterra, TrustRadius), app stores, Reddit and practitioner forums, community boards, social threads — capturing short real quotes with URLs and noting each source's bias.
- Emit the schema below exactly.
Output schema (do not reorder)
# Voice-of-Customer Snapshot
## 1. Scope
**Products mined:** | **Decision supported:** | **Sources swept:** | **As-of date:**
## 2. Need Themes
For each of the top 3-5 themes:
### Theme: [Underlying need, solution-free, 4 to 8 words]
- **Frequency:** [recurring across sources / concentrated / isolated]
- **Verbatim:** "[short real quote]" — [source, URL]
- **Verbatim:** "[short real quote]" — [source, URL]
- **Who says it:** [role/segment, if evident — labeled]
- **Reading:** [Inference — what this suggests]
## 3. Competitor Weak Points
- **[Competitor]:** [weakness in customers' words; frequency; URL]
- [Max 5, strongest evidence only]
## 4. Switching Triggers
- [What pushes customers off a product; what pulls them; labeled, cited]
## 5. So What?
- **3** opportunity hypotheses (phrased as problems, not features)
- **2** battle-card-ready weaknesses (with evidence quality noted)
- **3** assumptions to validate in real interviews
Each bullet: label, confidence, URL where relevant.
A copy/paste fill-in version of this schema, with quality checks, lives in template.md.
Final Step (offer exactly 4 options)
- Generate discovery interview questions from the top theme (
discovery-interview-prep) - Feed the weaknesses into a competitive battle card (
battle-card-builder) - Build an opportunity solution tree from the top hypothesis (
opportunity-solution-tree) - Re-run scoped to one theme in Verbose Mode
Accept 1, 2, 3, 4, 1 and 2, Verbose Mode, or a custom path.
Examples
A theme done right (fictional product, illustrative verbatims):
Theme: getting historical data out at contract end
- Frequency: recurring — 9 reviews across two sites plus a forum thread, past 6 months
- Verbatim: "export took three support tickets and still dropped custom fields" — [G2-style review, URL]
- Verbatim: "we stayed a year longer than we wanted because leaving meant losing our audit trail" — [forum thread, URL]
- Who says it: ops managers at 50-200-person firms — Inference (reviewer titles where shown)
- Reading: exit friction is functioning as involuntary retention — Inference; a rival with effortless migration turns this from their moat into their churn event.
Notice the theme name contains no feature ("export tool") — it names the need, so discovery can explore solutions the reviews never imagined.
See examples/sample.md for a complete worked mining run (fictional
FSM-software market) where frequency honesty caps a vivid theme at low confidence and each
source's bias becomes a reading instruction. examples/sample-industrial.md
shows the thin-voice case — what honest mining looks like when the market barely posts reviews.
Common Pitfalls
- Feature-name theming. Clustering by the feature customers blame instead of the need underneath hands your roadmap to the loudest UI complaint.
- Verbatim laundering. Paraphrasing a review and quoting it. If it has quote marks, it must be a real excerpt at a real URL — this domain's do-not-invent list exists because fabricated customer quotes are both tempting and toxic.
- Rant amplification. One vivid one-star review presented as a theme. Frequency honesty is the discipline: recurring, concentrated, or isolated — say which.
- Skew blindness. Reading review sites as a census. The angry and the vocal are over-sampled; the satisfied-and-silent majority never posts. Bias notes per source are mandatory.
- Skipping the validation handoff. Shipping themes straight into the roadmap. The output's "assumptions to validate in real interviews" section is the bridge to discovery — use it.
References
autonomous-investigation(Workflow) — the governing protocolintelligence-collection-disciplines(Component) — OSINT review-mining sources and bias tradecraftjobs-to-be-done(Component) — the solution-free framing themes should land indiscovery-interview-prep(Interactive) — where the validation happensopportunity-solution-tree(Interactive) — structures the opportunity hypothesesbattle-card-builder(Workflow) — consumes the weak points- Adapted from
market-intelligence/voice-of-customer-miner-prompt.mdin thehttps://github.com/deanpeters/product-manager-promptsrepo.
GitHub Repository
Frequently asked questions
What is the voice-of-customer-miner skill?
voice-of-customer-miner is a Claude Skill by deanpeters. Skills package instructions and resources that Claude loads on demand, so Claude can perform voice-of-customer-miner-related tasks without extra prompting.
How do I install voice-of-customer-miner?
Use the install commands on this page: add voice-of-customer-miner to Claude Code as a plugin, or clone its repository into your skills directory, then restart Claude so it picks up the skill.
What category does voice-of-customer-miner belong to?
voice-of-customer-miner is in the Other category, tagged ai.
Is voice-of-customer-miner free to use?
Yes. voice-of-customer-miner is listed on AIMCP and free to install.
Related Skills
LlamaGuard is Meta's 7-8B parameter model for moderating LLM inputs and outputs across six safety categories like violence and hate speech. It offers 94-95% accuracy and can be deployed using vLLM, Hugging Face, or Amazon SageMaker. Use this skill to easily integrate content filtering and safety guardrails into your AI applications.
This Claude Skill helps developers optimize cloud costs through resource rightsizing, tagging strategies, and spending analysis. It provides a framework for reducing cloud expenses and implementing cost governance across AWS, Azure, and GCP. Use it when you need to analyze infrastructure costs, right-size resources, or meet budget constraints.
This Claude Skill analyzes sports betting markets including spreads, over/unders, and prop bets by examining historical trends and situational statistics to identify value bets. It provides structured markdown output with actionable recommendations for educational purposes. Developers should use this for sports betting analysis tools while noting it's designed for entertainment/education only.
This skill quantizes LLMs to 8-bit or 4-bit precision using bitsandbytes, achieving 50-75% memory reduction with minimal accuracy loss. It's ideal for running larger models on limited GPU memory or accelerating inference, supporting formats like INT8, NF4, and FP4. The skill integrates with HuggingFace Transformers and enables QLoRA training and 8-bit optimizers.
