Transcript Analysis Skill — Discovery Pipeline

How the Transcript Analysis skill transforms meeting transcripts into structured discovery reports. Four steps from raw input to actionable recommendations: extract, analyze, map, report. Every finding is traceable to an explicit or implicit client signal with a confidence level.

flowchart TD
    INPUT["Input: Meeting transcript
Email chains, chat logs,
audio transcription, process descriptions"]

    INPUT --> S1["Step 1: Data Extraction
Statements, sentiment markers,
metrics, gaps, direct quotes"]

    S1 --> S2["Step 2: Deep Analysis"]

    S2 --> EXPL["Explicit Signals
Direct pain points, feature requests,
clear objections, measurable metrics"]
    S2 --> IMPL["Implicit Signals
Hesitations, conspicuous absences,
tone shifts, assumptions,
comparative language"]

    EXPL --> S3["Step 3: Map Insights to Solutions"]
    IMPL --> S3

    S3 --> SIG["Identify the Signal
Quote or observation
Tagged: explicit / implicit"]
    SIG --> NEED["Interpret the Need
Underlying pain point
Why it matters"]
    NEED --> CAT["Categorize the Impact
Short-term friction
Long-term strategic
UX issue
Business outcome"]
    CAT --> REC["Recommend Solutions
Quick wins (days/weeks)
Strategic (months/quarters)
Each with confidence level"]

    REC --> S4["Step 4: Produce Report"]

    S4 --> EXEC["Executive Summary
2-3 sentence overview"]
    S4 --> PAIN["Explicit Pain Points
Quote + Impact + Recommendation"]
    S4 --> INSIGHT["Implicit Insights
Signal + Interpretation + Action"]
    S4 --> QUICK["Quick Wins
Timeline: days/weeks"]
    S4 --> STRAT["Strategic Recommendations
Timeline: months/quarters"]
    S4 --> OPEN["Open Questions
Gaps + clarifications needed"]
    S4 --> NEXT["Next Steps
Prioritized action list"]

    S4 --> TQ{"Quality gate"}
    TQ -->|"Pass"| DONE["Report delivered"]
    TQ -->|"Flag"| ERR["Error handling paths"]

    ERR --> AMB["Ambiguous data:
Flag + request clarification"]
    ERR --> CONTRA["Contradictory signals:
Surface both, let stakeholders decide"]
    ERR --> MISS["Missing information:
Note gaps, suggest how to fill"]
    ERR --> LOW["Low-confidence inference:
Mark clearly, separate from high"]

    TECH["Advanced Techniques
Chain of Thought: document reasoning
Chain of Density: iterative distillation
Few-Shot Learning: adapt to new domains"] -.-> S2

    class S1,S2,S3,S4 step
    class EXPL,IMPL,SIG,NEED,CAT,REC signal
    class EXEC,PAIN,INSIGHT,QUICK,STRAT,OPEN,NEXT,DONE output
    class TQ choice
    class TECH technique
    class ERR,AMB,CONTRA,MISS,LOW error
    class INPUT quality

Workflow step

Signal detection

Report output

Quality gate

Advanced technique

Error path

Pipeline Steps

Each step builds on the last — from raw data to a structured, confidence-tagged report with traceable recommendations.

Step 1 Data Extraction

Thorough review of all transcript text and supporting materials.

Extract direct statements, sentiment markers, and metrics
Capture gaps, hesitations, or omissions that signal underlying issues
Pull direct quotes that anchor each finding

Step 2 Deep Analysis

Dual-lens signal detection — what's said and what's meant.

Explicit: Pain points, feature requests, objections, metrics
Implicit: Hesitations, absences, tone shifts, assumptions, comparative language
Chain of Thought, Chain of Density, Few-Shot techniques available for complex cases

Step 3 Map Insights to Solutions

Four-part mapping for every finding:

Identify the Signal — quote or observation, tagged explicit/implicit
Interpret the Need — underlying pain point, why it matters
Categorize Impact — short-term friction, strategic, UX, or business outcome
Recommend — quick wins vs. strategic, each with confidence level

Step 4 Produce Report

Structured output with seven sections:

Executive Summary — 2-3 sentence overview
Explicit Pain Points — quote + impact + recommendation
Implicit Insights — signal + interpretation + action
Quick Wins + Strategic Recommendations + Open Questions + Next Steps

Signal Detection

The core differentiator: two parallel signal detectors that catch what's stated and what's left unsaid.

Explicit Signals

Direct statements of pain points

Specific feature requests

Clear objections or concerns

Measurable metrics mentioned

Implicit Signals

Hesitations, pauses, or deflections

What's conspicuously NOT mentioned

Tone shifts or emotional markers

Underlying assumptions revealed by phrasing

Comparative language ("unlike X...")

Future-focused concerns signaling strategic anxiety

Output Structure

Seven-section report format. Every recommendation is traceable to a signal, tagged with a confidence level, and paired with a timeline.

Report Sections

1. Executive Summary 2-3 sentence overview of key findings — the "so what?"

2. Explicit Pain Points Direct quotes → impact assessed → recommendation with confidence level

3. Implicit Insights Observed signal → interpretation → recommendation with confidence level

4. Quick Wins Actions achievable in days/weeks — prioritized, with timelines

5. Strategic Recommendations Multi-month initiatives — phased, with milestones

6. Open Questions What needs clarification — gaps documented, assumptions stated

Quality Standards & Error Handling

Every output is gated. Ambiguous, contradictory, or low-confidence findings are flagged rather than silently included.

Quality Standards

Traceable: Every recommendation anchored to a client signal
Confidence-tagged: High / Medium / Low on every finding
Ambiguity-flagged: Contradictory data surfaced to stakeholders
Gap-documented: Missing information noted with plan to obtain it

Error Handling

Ambiguous data: Flag it, request clarification, document assumptions
Contradictory signals: Surface both interpretations, let stakeholders decide
Missing information: Note gaps explicitly, suggest how to fill them
Low-confidence: Mark clearly, separate from high-confidence findings

Communication Style

Evidence-based: Every claim backed by transcript data or clear inference. Action-oriented: Focus on "so what?" and "what next?" Transparent: Show your work, explain your reasoning. Structured: Bullet points, headings, consistent formatting. Professional but accessible: Technical rigor without jargon overload.