Ai Voc Review Insights

AI-powered Voice of Customer (VoC) review intelligence agent using DeepSeek-style analysis. Deep semantic analysis of customer reviews to extract pain points...

Security Scan

The skill is an instruction-only VoC/review analysis agent and its declared requirements and instructions are coherent with that purpose; it does not request credentials, install code, or perform unexpected actions.

Purpose & Capability

Name, description, and runtime instructions all describe semantic VoC analysis of pasted reviews. There are no declared env vars, binaries, or installs that would be unrelated to text analysis, so the requested capabilities match the stated purpose.

Instruction Scope

SKILL.md asks the user to paste 20–200 reviews and provides detailed analysis frameworks. It does not instruct the agent to read system files, access credentials, or call external endpoints. One minor oddity: the frontmatter lists allowed-tools: Bash, but the document contains no shell commands or install steps; this likely is harmless but could be clarified (it does not by itself cause network exfiltration).

Install Mechanism

No install spec and no code files are present (instruction-only). Nothing will be written to disk or downloaded as part of installation.

Credentials

The skill declares no required environment variables, credentials, or config paths. That is proportionate for a text-analysis instruction-only skill.

Persistence & Privilege

always is false and the skill is user-invocable; it does not request elevated or persistent privileges beyond normal skill invocation.

Assessment

This is an instruction-only VoC analysis skill and appears internally consistent. Before using it: (1) avoid pasting sensitive personal data or credentials into the reviews you provide; (2) confirm how your agent/model provider logs or stores inputs (your reviews will be sent to the model for analysis); (3) if you want assurance about behavior, ask the publisher to explain the 'allowed-tools: Bash' frontmatter and to confirm the skill will not run shell commands or call external endpoints; (4) review the GitHub homepage/license if you need provenance. Overall the skill is coherent and low-risk, but treat any pasted customer data as potentially visible to the model/service.

AI VoC Review Intelligence

Deep AI-powered Voice of Customer analysis — go beyond basic sentiment to extract purchase motivations, hidden pain points, unmet needs, and product-market fit signals from customer reviews across any platform.

Commands

voc analyze <reviews>             # full VoC analysis of review set
voc pain-points <reviews>         # extract and rank customer pain points
voc motivations <reviews>         # identify purchase motivations
voc unmet-needs <reviews>         # find unserved customer needs
voc personas <reviews>            # build customer persona from reviews
voc jobs-to-be-done <reviews>     # JTBD analysis from review language
voc compare <reviews1> <reviews2> # compare VoC between two products
voc opportunity <reviews>         # identify product development opportunities
voc marketing <reviews>           # extract marketing messages from reviews
voc report <product>              # full VoC intelligence report

What Data to Provide

• Reviews — paste 20-200 customer reviews (more = better analysis)
• Star distribution — 1-5 star count breakdown
• Product category — context for benchmarking
• Competitor reviews — for comparative VoC analysis
• Your marketing copy — to align with customer language

VoC Analysis Framework

Level 1: Surface Analysis (Standard Review Analysis)

What customers say explicitly:

"The product is great quality"
"Arrived quickly"
"Easy to assemble"
"A bit expensive but worth it"

Basic sentiment: positive/negative/neutral classification

Level 2: Semantic Analysis (What They Really Mean)

Reading between the lines:

Review: "Exactly what I needed" → Unmet need was real, product solves it
Review: "Better than I expected" → Category has history of disappointing products
Review: "I was skeptical but..." → High purchase anxiety in this category
Review: "Bought this as a gift" → Gifting is a significant use case
Review: "Replaced my old [brand]" → Competitor switching signal
Review: "My husband/wife loves it" → Multi-person household use
Review: "Works in my [specific context]" → Niche use case validation

Level 3: Jobs-to-be-Done (JTBD) Analysis

Functional jobs (what they hire the product to do):

• "I need to [task]"
• Extract the core functional use from review language

Emotional jobs (how they want to feel):

• "I feel confident/safe/proud/excited when..."
• Extract emotional outcomes from positive reviews

Social jobs (how they want to be perceived):

• "My [guests/family/colleagues] noticed..."
• Extract social signaling from reviews

JTBD template from reviews:
When I [situation], I want to [motivation], so I can [outcome].

Example from reviews of a standing desk converter:
When I work from home all day, I want to avoid back pain,
so I can stay productive without discomfort.

→ Marketing message: "Work pain-free all day. Designed for the modern home office."

Pain Point Extraction Matrix

Extract all pain points and classify:

Dimension 1: Frequency

• Mentioned in >20% of reviews: Critical issue
• Mentioned in 10-20%: Significant issue
• Mentioned in 5-10%: Notable issue
• Mentioned in <5%: Edge case

Dimension 2: Intensity

• "Terrible", "awful", "destroyed", "complete waste": Severity 5
• "Disappointed", "frustrated", "annoyed": Severity 4
• "Could be better", "wished it had": Severity 3
• "Minor issue", "small complaint": Severity 2
• Implied, not stated directly: Severity 1

Dimension 3: Resolution Potential

• Product redesign needed: Hard (3-6 months)
• Listing/instruction update: Easy (<1 week)
• Packaging/insert improvement: Medium (2-4 weeks)
• Customer service response: Immediate

Pain Point Matrix:
Pain Point           Freq   Intensity  Resolution  Priority
Instructions unclear 18%    3          Easy        HIGH
Strap breaks easily  12%    5          Hard        HIGH
Bag smaller than shown 9%   4          Listing fix MEDIUM
Color slightly off    6%    2          Listing fix LOW

Customer Persona Building

From review language patterns, identify buyer segments:

Segment 1: Core buyers (most reviews)

Demographics: [infer from review context]
Trigger: [what prompted purchase]
Use case: [primary use]
Success metric: [what makes them happy]
Quote: "[representative review excerpt]"

Segment 2: Edge case buyers (cause most problems)

Demographics: [who writes the negative reviews]
Mismatch: [how product doesn't meet their expectations]
Fix: [listing change to filter them out or meet their needs]

Segment 3: Surprise buyers (unexpected use cases)

Discovery: [how they found your product]
Use case: [unexpected application]
Opportunity: [new marketing angle or product variation]

Purchase Motivation Analysis

Extract why people buy, beyond the obvious:

Rational motivators (stated reasons):

• Quality, price, functionality, specifications

Emotional motivators (unstated reasons):

• Status, identity, relationships, fear/risk reduction
• Safety ("my child will be safe")
• Belonging ("everyone in our community uses this")
• Achievement ("I finally solved this problem")

Trigger events (what caused the purchase NOW):

• "After moving to a new home"
• "Since working from home"
• "After my old one broke"
• "Doctor recommended"
• "Saw on TikTok"

Unmet Needs Identification

Find gaps in the market from review language:

Explicit unmet needs:

• "I wish it came in [X]"
• "Would be perfect if it also [function]"
• "Need something like this but for [use case]"

Implicit unmet needs (inferred from workarounds):

• "I had to [work around]" → product doesn't do X natively
• "It would help if..." → feature request pattern
• Comparisons to competitors: what competitor does better

Competitive Switching Signals

From reviews mentioning competitors:

"Switched from [Brand X]" → X is your direct competitor
"Better than [Brand X]" → X is in buyer's consideration set
"[Brand X] stopped working, got this" → X has quality issues
"Half the price of [Brand X]" → X is premium alternative

Marketing Message Extraction

The best marketing copy comes directly from customer words:

Reviews say:                 → Marketing copy:
"Finally found one that..."  → "The [product] you've been searching for"
"Works exactly as advertised" → "What you see is what you get"
"Gift for my husband, he loves it" → "The gift he'll actually use"
"Solved my [problem]"        → "[Problem]? Problem solved."
"Worth every penny"          → "Invest in quality. Feel the difference."

Sentiment Evolution Analysis

Compare early reviews vs. recent reviews:

Early reviews (product launch): Focus on unboxing, first impressions
Recent reviews (mature product): Focus on durability, long-term value

Declining sentiment pattern:
Early avg: 4.5 stars → Recent avg: 3.9 stars
Signal: Quality or supplier change, investigate manufacturing

Workspace

Creates ~/voc-intelligence/ containing:

• analyses/ — full VoC reports per product
• personas/ — customer persona profiles
• pain-points/ — pain point matrices
• marketing/ — extracted marketing messages
• jtbd/ — jobs-to-be-done frameworks

Output Format

Every VoC analysis outputs:

1. VoC Executive Summary — 5 key findings in plain language
2. Pain Point Matrix — all pain points scored by frequency × intensity
3. JTBD Framework — functional, emotional, and social jobs identified
4. Customer Personas — 2-3 buyer segments with profiles
5. Unmet Needs List — product/feature gaps discovered
6. Marketing Messages — 5 ready-to-use copy lines from customer language
7. Competitor Switching Map — which competitors appear and in what context
8. Product Roadmap Signals — prioritized improvements by business impact