--- name: transcript-analysis description: Analyze support conversations including phone calls, chat transcripts, and email threads to identify training opportunities, quality issues, compliance violations, coaching moments, and best practices. Use when reviewing call quality, analyzing chat conversations, identifying training gaps, building example libraries, conducting quality calibration, or improving agent communication skills. Triggers on phrases like "transcript analysis", "call quality review", "conversation analysis", "quality audit", "training examples", "coaching moments", "speech-to-text", "chat transcript review". --- # Call and Chat Transcript Analysis Automatically analyze support conversations across all channels to identify training opportunities, quality issues, compliance gaps, and best practices — turning every customer interaction into a learning opportunity. ## Workflow ### 1. Automatic Transcription and Analysis Pipeline 1. **Transcription and data collection**: - **Phone calls**: Automatic speech-to-text via Twilio/RingCentral integration (98% accuracy) - **Live chat**: Real-time text capture from chat platform - **Email threads**: Full conversation history parsed and analyzed - **Video calls**: Transcription via Zoom/Teams API integration - **Multilingual**: Auto-detect language, transcribe/translate to agent's language - **Processing time**: <60 seconds post-call, real-time for chat 2. **Analysis dimensions**: ``` CONVERSATION ANALYSIS FRAMEWORK =============================== OPENING ASSESSMENT: ✓ Professional greeting used ✓ Agent introduced themselves by name ✓ Customer verified/identified ✓ Tone: Warm, professional, empathetic Timing: Greeting within first 10 seconds ACTIVE LISTENING INDICATORS: ✓ Acknowledged customer's issue ✓ Asked clarifying questions ✓ Paraphrased customer's problem to confirm understanding ✓ Didn't interrupt customer Quality: 3+ indicators present = good listening EMPATHY STATEMENTS: Detected: "I understand this is frustrating" Detected: "I apologize for the inconvenience" Detected: "Let me make this right for you" Score: 3 statements (above average: 2.1) ✓ PROBLEM-SOLVING APPROACH: ✓ Systematic troubleshooting (step-by-step) ✓ Used knowledge base resources ✓ Offered multiple solutions when applicable ✓ Confirmed resolution with customer Efficiency: Resolution in 8.2 minutes (category avg: 12.4 min) ✓ CLOSING AND CONFIRMATION: ✓ Summarized actions taken ✓ Confirmed customer satisfaction ✓ Offered additional assistance ✓ Provided follow-up information ✓ Professional sign-off with name CUSTOMER SENTIMENT FLOW: Opening: Neutral (0.1) Middle: Frustrated (-0.4) — issue explained After empathy: Calming (-0.1) Resolution: Positive (0.6) Closing: Very positive (0.8) Overall trajectory: IMPROVING ✓ (+0.7 net change) ``` 3. **Flagging system for human review**: - **Red flags** (immediate review required): - Customer sentiment dropped below -0.7 - Compliance keywords detected (incorrect policy stated) - Agent used unprofessional language - Resolution not confirmed before closing - **Yellow flags** (review within 24 hours): - Missing empathy statements - Resolution time 2x above category average - Customer asked follow-up questions (may not be fully resolved) - Multiple re-explanations of same concept - **Green flags** (positive examples for team sharing): - Customer sentiment improved significantly - Resolution achieved faster than average - Customer explicitly praised agent - Complex issue resolved with first contact ### 2. Quality Scoring and Calibration 1. **Automated quality scorecard**: ``` CONVERSATION QUALITY SCORECARD ============================== Ticket ID: TKT-88234 | Agent: Sarah M. | Duration: 12 min | Date: Jan 12, 2026 Category | Score (0-20) | Weight | Weighted Score ------------------------|--------------|--------|--------------- Opening & Greeting | 19/20 | 10% | 1.9 Active Listening | 18/20 | 15% | 2.7 Empathy & Tone | 17/20 | 15% | 2.55 Problem-Solving | 19/20 | 25% | 4.75 Knowledge Accuracy | 20/20 | 15% | 3.0 Closing & Confirmation | 18/20 | 10% | 1.8 Compliance | 20/20 | 10% | 2.0 ------------------------|--------------|--------|--------------- TOTAL | | 100% | 18.75/20 = 93.8% Grade: A (90-100%) ✓ Team Average: 85.2% → Agent is ABOVE AVERAGE by 8.6% SPECIFIC FEEDBACK: + Excellent: Knowledge accuracy — provided correct solution immediately + Excellent: Problem-solving — systematic, efficient approach + Strong: Empathy — used 3 empathetic statements appropriately + Good: Active listening — paraphrased customer issue correctly ~ Opportunity: Opening — could include more personalization (customer name earlier) ~ Opportunity: Closing — could offer proactive follow-up COMPLIANCE CHECK: ✓ No unauthorized discount promises ✓ Correct refund policy stated ✓ Customer data handled appropriately ✓ Required disclosures included COACHING TAGS: #empathy-excellence #fast-resolver #knowledge-expert ``` 2. **Team quality calibration**: - Weekly calibration sessions: Team leads review 5 conversations together - Automated scoring vs human scoring comparison (measure AI accuracy) - Adjust scoring weights based on calibration feedback - Consistent standards across all evaluators 3. **Trend analysis**: - Quality score trends per agent (improving, stable, declining) - Team-wide quality trends (month-over-month) - Category-specific quality (which issue types have lowest quality?) - Sentiment trajectory analysis (are customers consistently more positive at end?) ### 3. Example Library and Training Material Generation 1. **Best practice example collection**: - Top 5% conversations saved as "gold standard" examples - Tagged by skill demonstrated (empathy, troubleshooting, de-escalation) - Available in LMS for training and self-study - New hire onboarding: Review 20 examples before handling live tickets 2. **Training opportunity identification**: - Common failure patterns across team → group training topic - Individual agent gaps → personalized training assignment - New product feature: First 50 conversations → training improvement - Seasonal issues: Pre-emptive training before peak volume ## Templates & Frameworks ### Transcript Analysis Dashboard ``` TRANSCRIPT ANALYTICS DASHBOARD — January 2026 =============================================== ANALYSIS VOLUME: Conversations analyzed this month: 8,920 Phone calls transcribed: 3,420 (98.2% accuracy) Chat transcripts analyzed: 4,230 Email threads analyzed: 1,270 Video call transcripts: 1,340 Analysis coverage: 96.8% of all customer interactions Languages analyzed: English (7,890), Spanish (623), German (289), French (118) Average analysis time: 42 seconds per conversation Processing backlog: 0 (real-time) QUALITY SCORE SUMMARY: Average quality score: 85.2/100 (↑ 2.1 from December) Score distribution: A (90-100%): 1,240 conversations (13.9%) B (80-89%): 3,890 conversations (43.6%) C (70-79%): 2,560 conversations (28.7%) D (<70%): 1,230 conversations (13.8%) ⚠ Trend: ↑ 2.1 points improvement from December ✓ FLAGGED CONVERSATIONS: Red flags (immediate review): 89 Completed review: 87 (97.8%) Pending review: 2 (2.2%) Average resolution time: 3.4 hours Yellow flags (24-hour review): 342 Completed review: 298 (87.1%) Pending review: 44 (12.9%) Green flags (positive examples): 567 Added to training library: 234 (41.3%) Shared with team: 89 (15.7%) Agent recognition: 67 (11.8%) SENTIMENT ANALYSIS: Average customer sentiment at conversation start: -0.1 (slightly negative — expected) Average customer sentiment at conversation end: +0.4 (positive — good) Average sentiment improvement: +0.5 ✓ (target: >+0.3) Sentiment by channel: Phone: +0.6 (highest improvement) Live chat: +0.4 (average improvement) Email: +0.3 (lowest improvement — expected for async) Video call: +0.5 (high improvement) Conversations where sentiment worsened: 423 (4.7%) Root causes: Issue not resolved: 189 (44.7%) Agent tone perceived as dismissive: 89 (21.0%) Long wait times before agent connection: 78 (18.4%) Incorrect solution provided: 67 (15.8%) TOP TRAINING OPPORTUNITIES IDENTIFIED: Priority | Issue Pattern | Agents Affected | Training Action ---------|-----------------------------------|----------------|----------------- HIGH | Missing empathy in billing tickets| 14 agents | "Empathy in Billing" workshop (Jan 20) HIGH | Incorrect SSO troubleshooting steps| 8 agents | "SSO Troubleshooting" retraining (Jan 18) MEDIUM | Rushing closing without confirmation| 12 agents | "Proper Closing" micro-training (Jan 22) MEDIUM | Not paraphrasing customer issue | 18 agents | "Active Listening" module (Jan 25) LOW | Greeting inconsistencies | 6 agents | Quick reference card (distributed Jan 15) COMPLIANCE MONITORING: Compliance violations detected: 23 Types: Incorrect refund policy stated: 12 (52.2%) Unauthorized discount promise: 5 (21.7%) Missing required disclosure: 4 (17.4%) Customer data handling issue: 2 (8.7%) Compliance violation rate: 0.26% (target: <0.5%) ✓ Trend: ↓ 35% from last month ✓ Agents with compliance violations (repeat offenders): James T.: 3 violations (mandatory retraining assigned) Kevin W.: 2 violations (coaching scheduled) EXAMPLE LIBRARY: Total examples in library: 1,234 New examples added this month: 234 Examples used in training sessions: 89 Agent views of example library: 3,420 (this month) Top example categories: 1. De-escalation (156 examples) 2. First-contact resolution (134 examples) 3. Empathy demonstration (123 examples) 4. Technical troubleshooting (112 examples) 5. Upsell/Expansion (89 examples) AGENT RECOGNITION: "Conversation of the Week" winners: 4 Top agents by quality score: 1. Sarah M. — 93.8% (42 conversations reviewed) 2. Tom R. — 92.1% (38 conversations reviewed) 3. Lisa K. — 91.5% (35 conversations reviewed) Improvement leaders: 1. Maria L. — +8.2 points (from 78.4% to 86.6%) 2. David K. — +6.7 points (new hire, rapid improvement) ``` ### Quality Calibration Framework ``` QUALITY CALIBRATION SESSION TEMPLATE ====================================== Date: January 15, 2026 | Participants: 3 Team Leads, 2 QA Specialists CONVERSATIONS REVIEWED: 5 (randomly selected, anonymized) CALIBRATION RESULTS: Conversation | AI Score | Lead A | Lead B | Lead C | Consensus | Action -------------|----------|--------|--------|--------|-----------|-------- #1 | 88% | 90% | 88% | 87% | 88% | ✓ Aligned #2 | 92% | 91% | 93% | 90% | 91% | ✓ Aligned #3 | 76% | 72% | 78% | 74% | 75% | ⚠ AI over-scores empathy #4 | 85% | 83% | 86% | 84% | 84% | ✓ Aligned #5 | 94% | 96% | 93% | 95% | 95% | ✓ Aligned AVERAGE VARIANCE: 2.4 points (target: <3 points) ✓ AI vs HUMAN CORRELATION: 0.94 (strong alignment) CALIBRATION ADJUSTMENTS: 1. Reduce empathy weight slightly (AI over-scores by ~2 points) 2. Add "solution completeness" sub-criteria to problem-solving 3. Clarify compliance scoring rubric (ambiguity on disclosure timing) NEXT CALIBRATION SESSION: January 22, 2026 ``` ## Integration Points - **Phone systems** (Twilio, RingCentral): Call recording, speech-to-text transcription - **Chat platforms** (Intercom, Zendesk Chat, Drift): Real-time chat transcript capture - **Email systems** (Gmail, Outlook, custom SMTP): Email thread parsing and analysis - **Video platforms** (Zoom, Teams): Meeting transcription, screen sharing context - **NLP services** (AWS Comprehend, Google NLP, OpenAI): Sentiment analysis, topic extraction, language detection - **LMS** (Docebo, Lessonly): Training assignment, example library integration - **Quality management** (MaestroQA, QualityHub): Score synchronization, evaluation workflows - **Communication** (Slack, Teams): Quality alerts, recognition notifications, calibration scheduling ## Edge Cases - **Accented or non-standard speech patterns**: International customers with strong accents reduce transcription accuracy: - Accuracy impact: Drops from 98% to ~85% for heavy accents - Mitigation: Accent-aware models trained on diverse speech data - Human review: Low-confidence transcriptions flagged for human review - Agent context: "Transcription confidence: 78% — review recommended" - Continuous improvement: Opt-in accent data collection for model training - **Sensitive conversations requiring discretion**: Customer discussing medical or financial issues during call: - Sensitivity detection: Keywords trigger "sensitive conversation" flag - Access restriction: Only direct manager and QA specialist can access transcript - Anonymization: Customer name redacted in training examples - Compliance: GDPR/HIPAA considerations for health/financial data - Agent training: Special protocols for handling sensitive topics - **Multilingual conversations mixing languages**: Customer switches between English and Spanish mid-conversation: - Language detection: Per-sentence language identification - Hybrid transcription: Mix of languages preserved accurately - Translation: Full conversation translated to English for analysis - Analysis: Sentiment and quality assessed in both languages - Scoring: Language-switching doesn't penalize quality score - **Long conversations exceeding analysis limits**: 45-minute call with multiple topic shifts: - Segmentation: Conversation split into topic-based segments - Per-segment analysis: Each segment scored independently - Holistic view: Overall score weighted average of segments - Summary: AI generates conversation summary highlighting key moments - Performance: Processing time scales linearly (~60 seconds per 10 minutes) - **Agent disputes quality score**: Agent disagrees with automated quality assessment: - Appeal process: Agent can request human review within 48 hours - Transparency: Full scoring breakdown visible to agent - Calibration: Human reviewer uses same rubric as AI - Feedback loop: Discrepancies fed back to improve AI model - Resolution: Score adjusted if appeal justified, original agent notified ## Output ### Monthly Transcript Analysis Report ``` TRANSCRIPT ANALYSIS MONTHLY REPORT — January 2026 =================================================== ANALYSIS COVERAGE: Total conversations analyzed: 8,920 (96.8% of all interactions) ✓ Phone calls: 3,420 | Chat: 4,230 | Email: 1,270 | Video: 1,340 Languages: English, Spanish, German, French QUALITY IMPROVEMENT: Average score: 85.2/100 (↑ 2.1 from December) ✓ Conversations scoring A-grade: 13.9% (↑ 2.3%) Conversations scoring D-grade: 13.8% (↓ 3.1%) ✓ QUALITY TREND: Consistent improvement for 4 consecutive months COMPLIANCE: Violations detected: 23 (0.26% — well below 0.5% target) ✓ Trend: ↓ 35% from last month ✓ Repeat offenders addressed: 2 agents (mandatory retraining) TRAINING IMPACT: Training sessions driven by transcript analysis: 12 Agents improved after targeted training: Average +5.4 points Example library growth: 234 new examples (total: 1,234) Agent engagement with library: 3,420 views (↑ 18%) RECOMMENDATIONS: 1. Launch "Empathy in Billing" workshop (14 agents affected, high priority) 2. Address SSO troubleshooting knowledge gap (8 agents — retraining planned) 3. Improve closing confirmation practice (12 agents — micro-training) 4. Expand example library for de-escalation scenarios (highest-viewed category) 5. Add real-time quality coaching during live calls (pilot program Q1) ```