v1.3.03 loop iterations

Voice of Customer Miner

Reading a hundred reviews one at a time tells you almost nothing — patterns hide in the aggregate. The Voice of Customer Miner pulls themes, frequency, intensity, and the exact phrases customers actually use, so you can act on signal instead of reacting to anecdotes.

What this skill does

Customer feedback gets collected by most businesses and analysed by almost none. Reviews are read one at a time, tickets are resolved individually, NPS scores show up in a dashboard and nothing changes. The Voice of Customer Miner converts that backlog into a structured analysis — themes ranked by frequency and intensity, exact customer phrasing extracted for marketing, a priority matrix that tells product what to fix first.

The MINE framework runs underneath: Map the themes (group by underlying intent, not surface wording), Intensity analysis (frequency, strength, trend, correlation with churn or retention), Nuggets (exact quotes that capture the pattern), Extract actions (specific recommendations per theme, owned by product, marketing, support, or leadership). The threshold for a theme is three or more mentions — one customer is an anecdote, three is signal, and the skill refuses to act on anecdotes.

Customer language gets preserved verbatim. The phrases customers use to describe your value are better marketing copy than anything your team will write, because they're how the customer explains the product to their friend at the pub. "It's like having a CFO without the £150K salary" beats "Comprehensive financial oversight for growing businesses" every time, and the skill flags exact phrases that appear repeatedly across different respondents as headline candidates. Paraphrasing loses the authenticity that makes the phrase work.

Themes get plotted on a frequency-by-intensity matrix that maps directly to action. High frequency plus high intensity is the "fix now" quadrant. High frequency plus low intensity gets addressed systematically. Low frequency plus high intensity gets investigated rather than ignored — three furious customers can be a leading indicator of thirty mildly annoyed ones who never bothered to write in. Low frequency plus low intensity is noted and parked.

What's not said gets surfaced as deliberately as what is. If nobody mentions a feature you've been investing in, that's a finding. If nobody mentions a competitor you've been worried about, that's also a finding. Sentiment is broken down by source — support tickets indicate friction, reviews indicate overall sentiment, social indicates public perception — and weighted accordingly. Small samples (under 20 responses) get flagged as hypothesis-level rather than conclusion-level, because pretending you have signal you don't have is worse than admitting the data isn't there yet.

When this triggers

·You have hundreds of reviews, NPS comments, or tickets and no time to read them individually
·You're rebuilding a sales page and want the headline to match how customers describe you, not how you describe yourself
·Churn is creeping up and you need to find what the leavers are actually telling you
·Product is debating which feature to build next and the customer data has gone unmined
·You want the exact phrases real customers use so the next campaign sounds like them, not the marketing team

Example

Trigger

User pastes 80 NPS responses from a fractional finance product, six months of data.

Output

Voice of Customer Analysis Source: NPS verbatim, in-app survey Period: Oct–Mar | Volume: 80 responses Executive Summary Three findings dominate: (1) the "CFO without the salary" framing appears in 14 separate responses — strongest marketing handle in the data. (2) Onboarding friction clusters tightly around the bank connection step (mentioned in 19 of 26 detractor responses). (3) Pricing perceived as fair in 71% of promoter responses but unclear in 40% of passive responses — the issue is comprehension, not amount. Theme Map | Theme | Frequency | Intensity | Trend | Key quote | |---|---|---|---|---| | "CFO replacement" framing | 14 (18%) | High | Up | "It's like having a CFO without the £150K salary" | | Bank connection friction | 19 (24%) | High | Stable | "Spent 40 minutes trying to connect Monzo, gave up twice" | | Pricing comprehension | 21 (26%) | Medium | Stable | "I still don't fully understand what tier I'm on" | | Speed of insights | 11 (14%) | High (positive) | Up | "First time I've actually understood my margin" | Priority Matrix Fix Now (High freq + High intensity) · Bank connection flow — 19 of 26 detractors mention this Improve Systematically (High freq + Low intensity) · Pricing page clarity — rewrite tiers with concrete examples Investigate (Low freq + High intensity) · Three responses mention a missing forecasting feature — not enough signal to build, enough to interview five users Marketing Goldmine Customer language to use: · "CFO without the £150K salary" (14 uses, near-verbatim) · "Finally see my margin" (8 uses) · "Stops me guessing about cash" (6 uses) Testimonial candidates: 4 promoters with specific quotes, flagged for outreach. Competitor Intelligence [Competitor A] mentioned 9 times — switching reason in 7 of them was the "monthly bookkeeper hand-off problem." Worth a dedicated comparison page.

Get this skill + 7 more

Get the full Customer Support pillar (8 skills) or the complete library.

Get the full stack — $299

What you get

167-line SKILL.md, ready to drop into ~/.claude/skills/
Tested through 3 Karpathy-loop iterations (versions v1.0.0 → v1.3.0)
Triggers automatically when relevant — no command to remember
Lifetime updates as the skill is refined further