call-recording-analyzer

name

call-recording-analyzer

description

Gong/Chorus-style call transcript analysis using Clari data. Scores calls against MEDDIC framework, flags missed discovery questions, surfaces competitor mentions, extracts action items, and identifies coaching moments. Feeds deal-momentum-analyzer and morning-brief with call quality signals. Use when: 'analyze call', 'call review', 'score my call', 'what did I miss', 'call coaching', 'review transcript', 'Clari analysis', 'call scorecard', 'how did my call go', 'review my last call with [company]'.

Replace manual call review with automated Clari transcript analysis. Every call Tim makes gets scored against MEDDIC dimensions, flagged for missed discovery questions, and fed into downstream skills (deal-momentum-analyzer, morning-brief). Targets: identify 3+ coaching moments per call, surface competitor intel automatically, and reduce post-call admin from 10 minutes to under 60 seconds.

<quick_start> After a call: "analyze my last call with Baylor" → pulls Clari transcript → scores against MEDDIC → outputs scorecard

Batch review: "review today's calls" → analyzes all calls from today → summary report

Coaching mode: "call coaching for my University of Michigan demo" → deep analysis with talk ratio, question quality, objection handling

Trigger phrases:

• "analyze call [company]"
• "call review [company]"
• "score my call"
• "what did I miss on [company] call"
• "call coaching [company]"
• "review my last call"
• "review today's calls"
• "how did my call go"
• "Clari scorecard" </quick_start>

<success_criteria>

• MEDDIC scorecard generated for every analyzed call
• Missed discovery questions identified (what SHOULD have been asked)
• Competitor mentions extracted and mapped to battlecard responses
• Talk-to-listen ratio calculated (target: prospect talks 60%+)
• Action items extracted with owner assignment
• Coaching moments flagged (interruptions, missed objection follow-ups, premature pitching)
• Results feed into deal-momentum-analyzer Signal 4 (call momentum scoring)
• Analysis completes in under 60 seconds per call </success_criteria>

Architecture

TRIGGER                    DATA PULL                 ANALYSIS                OUTPUT
──────────────────────────────────────────────────────────────────────────────────
"analyze call X"  →  Clari: transcript +     →  MEDDIC scoring      →  Call Scorecard
                     summary + action items   →  Discovery gaps       →  Coaching Report
                  →  HubSpot: deal context    →  Competitor extract   →  Action Items
                  →  Prior call history       →  Talk ratio analysis  →  Deal Update
                                              →  Objection mapping    →  Morning Brief Feed

Stage 1: Data Collection

Run these MCP calls in parallel:

1a. Clari Call Data

Tool	Purpose
clari_search_calls	Find the call by company name, attendee email, or date range
clari_get_call	Full call metadata: duration, participants, recording URL
clari_get_call_summary	AI summary, key moments, action items, sentiment

Search strategy:

1. If company name provided: clari_search_calls(query=company_name, daysBack=7)
2. If "last call" or "today's calls": clari_search_calls(daysBack=1)
3. If attendee specified: clari_search_calls(attendeeEmail=email, daysBack=14)

1b. HubSpot Deal Context

Tool	Purpose
hubspot_search_companies	Company record for the prospect
hubspot_search_deals	Active deals associated with this company
hubspot_search_contacts	Contact records for call participants

Pre-Flight Golden Rules Check:

• If contact.lifecyclestage = 'customer' → Skip. Route to CSM.
• If hubspot_owner_id IN ('82625923', '423155215', '190030668') AND Tim is NOT collaborator → Skip.

1c. Prior Call History

Tool	Purpose
clari_search_calls	Find previous calls with same contacts (last 90 days)
clari_get_call_summary	Summaries of prior calls for progression context

Stage 2: MEDDIC Call Scoring

Score the call transcript against each MEDDIC dimension (0-100 total).

M — Metrics (0-20 points)

Evidence	Points
Prospect stated specific success metrics	20
Prospect described general goals	12
Tim asked but prospect deflected	8
Not discussed	0

Missed question: "How will you measure success? What does ideal look like in numbers?"

E — Economic Buyer (0-20 points)

Evidence	Points
EB identified by name and confirmed as budget holder	20
EB referenced indirectly	12
Tim asked, answer unclear	8
No discussion of budget authority	0

ATL/BTL Validation (MANDATORY — per CLAUDE.md § ATL/BTL Classification v1.0): Classify ALL call attendees. If NO ATL attendee → Flag ⚠️ COACHING MOMENT: "Get the economic buyer on the next call."

D — Decision Criteria (0-15 points)

Evidence	Points
Prospect listed specific evaluation criteria	15
Some factors mentioned	10
Tim asked, prospect uncertain	5
Not discussed	0

D — Decision Process (0-15 points)

Evidence	Points
Clear timeline + stakeholders + process	15
Partial timeline	10
Tim asked, vague	5
Not discussed	0

I — Identify Pain (0-20 points)

Evidence	Points
Specific quantified pain	20
Qualitative pain	12
Pain implied	6
No pain discussion (pitch without discovery)	0

C — Champion (0-10 points)

Evidence	Points
Contact volunteered to champion internally	10
High engagement (detailed questions, requested next steps)	7
Passive engagement	3
No champion signal	0

MEDDIC Classification

Score	Rating	Meaning
80-100	STRONG	Well-qualified deal
60-79	ADEQUATE	1-2 gaps to fill next call
40-59	WEAK	Incomplete discovery
0-39	MISSED	Call was a pitch, not discovery

Stage 3: Coaching Analysis

Talk-to-Listen Ratio

• Target: Prospect talks 60%+, Tim talks 40% or less
• Red flag: Tim talked 60%+ = premature pitching

Question Quality

Type	Quality
Open-ended discovery	HIGH
Calibrated (NSTTD style)	HIGH
Closed confirmation	MEDIUM
Leading/assumptive	LOW
Product feature dump	COACHING MOMENT

Competitive Intelligence Extraction

Scan for: Extron, Blackmagic, Crestron, vMix, Matrox, Teradek, Kaltura, Panopto, YuJa, Echo360

Stage 4: Output Format

╔══════════════════════════════════════════════════════════════╗
║  CALL SCORECARD: [Company Name]                              ║
║  [Date] [Duration] | [Call Type: Discovery/Demo/Follow-up]   ║
╠══════════════════════════════════════════════════════════════╣

MEDDIC SCORE: [XX/100] — [STRONG/ADEQUATE/WEAK/MISSED]

DISCOVERY GAPS — ASK THESE NEXT TIME:
1. [Specific question for biggest MEDDIC gap]
2. [Specific question for second gap]

COACHING MOMENTS:
⚠️ [Context] — [What happened] → [What to do differently]

COMPETITIVE INTEL: [Competitor mentioned + battlecard response]

TALK RATIO: Tim [X%] / Prospect [X%]

ACTION ITEMS:
☐ [Action] — Owner: [Tim/Prospect] — Due: [date]

╚══════════════════════════════════════════════════════════════╝

<downstream_integration>

Feed to Sibling Skills

deal-momentum-analyzer (Signal 4: Call Momentum)

Push: call_recency, call_sentiment, meddic_score, unresolved_actions, competitor_detected

morning-brief

Include: yesterday's call scores, unresolved action items, deals with MEDDIC gaps needing follow-up

meddic-call-prep-auto

Feed prior call gaps into next call prep </downstream_integration>

## Required MCP Tools - **Epiphan Clari MCP:** clari_search_calls, clari_get_call, clari_get_call_summary - **Epiphan CRM MCP:** hubspot_search_companies, hubspot_search_contacts, hubspot_search_deals, ask_agent - **Gmail MCP:** gmail_create_draft

Sibling Skills Referenced

• deal-momentum-analyzer — Consumes call quality signals for Signal 4 scoring
• meddic-call-prep-auto — Receives call gaps to inform next call prep
• morning-brief — Displays yesterday's call summaries

Emit Outcome Sidecar

As the final step, write to ~/.claude/skill-analytics/last-outcome-call-recording-analyzer.json:

{"ts":"[UTC ISO8601]","skill":"call-recording-analyzer","version":"1.0.0","variant":"default",
 "status":"[success|partial|error]","runtime_ms":[estimated ms from start],
 "metrics":{"calls_analyzed":[n],"avg_meddic_score":[n],"coaching_moments_found":[n],"competitors_detected":[n]},
 "error":null,"session_id":"[YYYY-MM-DD]"}

Use status "partial" if some stages failed but results were produced. Use "error" only if no output was generated.