Sales Coaching

An AI coach that studies everything you do in sales — your emails, your calls, your deals — finds the patterns in what works and what doesn't, and gives you specific, actionable coaching to get better.

This isn't generic sales advice ("always be closing"). It's coaching derived from YOUR data: your top-performing emails, your winning calls, your successful deal patterns, your most common objection fumbles, and your specific product's market.

What makes this different from sequence-performance:

Composite	Focus	Output
sequence-performance	How is this campaign doing?	Campaign metrics + copy diagnosis
sales-coaching	How can THIS person sell better?	Personal skill assessment + coaching plan

The other composite analyzes the work. This one coaches the worker.

When to Auto-Load

Load this composite when:

• User says "how can I improve my sales", "coach me", "what am I doing wrong", "sales coaching", "help me sell better"
• User says "review my selling style", "analyze my calls", "what patterns do you see in my sales"
• User asks "why am I losing deals", "why aren't people responding", "what are my best emails doing differently"
• After a bad quarter/month and the user wants to diagnose personal performance
• As a periodic (monthly/quarterly) self-improvement exercise

---

Step 0: Configuration (One-Time Setup)

User Profile

Question	Purpose	Stored As
What is your role?	Calibrate coaching level	user_role
How long have you been in sales?	Experience-appropriate advice	experience_level
What do you sell? (product/service, 2-3 sentences)	Context for all analysis	product_description
What's your average deal size?	Calibrate what matters	avg_deal_size
What's your typical sales cycle?	Calibrate velocity expectations	sales_cycle_days
What do you think your biggest weakness is?	Starting point for coaching	self_assessed_weakness

Role options:

• "sdr" — Focus on prospecting, outreach, qualification, meeting booking
• "ae" — Focus on discovery, demos, negotiation, closing
• "founder" — Focus on everything (wearing multiple hats)
• "sales_leader" — Focus on team patterns, not just individual

Data Sources

Question	Options	Stored As
Where are your email campaigns?	Smartlead / Instantly / Outreach / CSV	email_tool
Where are your call recordings/transcripts?	Gong / Chorus / Fireflies / Otter / local files / none	call_tool
Where is your pipeline?	Salesforce / HubSpot / Pipedrive / Close / Supabase	crm_tool
Do you have call transcripts available?	Yes (path or tool) / No	transcripts_available
How far back should we analyze?	30 / 60 / 90 / 180 days	analysis_window

Store config in the current working directory or wherever the user prefers.

---

Step 1: Collect Sales Data

Purpose: Pull all available sales data for analysis. The more data types available, the richer the coaching. But the system works with whatever is available.

Input Contract

data_sources: { ... }                 # From config
analysis_window: integer              # Days to look back
user_role: string                     # From config

Data Collection Matrix

Data Type	What to Pull	What It Reveals
Email campaigns	All campaigns in window — copy, metrics, replies	Writing quality, messaging effectiveness, personalization skill
Email replies	Full reply text, classification	Objection handling, how prospects respond to you
Call recordings/transcripts	Full transcripts or recordings	Talk-to-listen ratio, discovery skill, objection handling, closing technique
Pipeline deals	All deals in window — stage, outcome, timeline	Deal management, velocity, where deals stall or die
Meeting notes	Post-call notes (if available)	Follow-through, note-taking discipline
Calendar	Meetings booked, held, no-shows	Time management, meeting quality

Pull from each configured source:

Email Data

email_data: {
  campaigns: [
    {
      name: string
      date_range: { start: string, end: string }
      sequence: [
        {
          touch: integer
          subject: string
          body: string
          sent: integer
          opens: integer
          replies: integer
        }
      ]
      replies: [
        {
          sender: string
          sender_title: string
          reply_text: string
          category: string            # positive, objection, not interested, etc.
          touch_triggered: integer
        }
      ]
    }
  ]
  total_campaigns: integer
  total_emails_sent: integer
  total_replies: integer
  overall_reply_rate: percentage
}

Call Data

call_data: {
  calls: [
    {
      date: string
      prospect_name: string
      prospect_company: string
      call_type: string               # discovery, demo, follow-up, negotiation
      duration_minutes: integer
      transcript: string | null       # Full transcript if available
      recording_url: string | null    # Recording link if available
      outcome: string                 # "next step agreed", "no next step", "closed won", "closed lost"
      notes: string | null            # Post-call notes
    }
  ]
  total_calls: integer
  avg_duration: float
  outcome_distribution: { ... }
}

Pipeline Data

pipeline_data: {
  deals: [
    {
      name: string
      company: string
      stage: string
      created_date: string
      close_date: string | null
      outcome: "open" | "won" | "lost"
      loss_reason: string | null
      amount: number | null
      days_in_pipeline: integer
      source: string | null           # How the deal originated
      touches_before_close: integer | null
    }
  ]
  total_deals: integer
  win_rate: percentage
  avg_cycle_days: float
  avg_deal_size: number | null
}

Output Contract

collected_data: {
  email_data: { ... } | null
  call_data: { ... } | null
  pipeline_data: { ... } | null
  data_richness: "full" | "partial" | "minimal"
  data_summary: string                # "Analyzing X campaigns, Y calls, Z deals over N days"
}

Human Checkpoint

## Data Collected

| Source | Available | Volume |
|--------|-----------|--------|
| Email campaigns | Yes | X campaigns, Y emails sent, Z replies |
| Call transcripts | Yes/No | X calls, Y hours |
| Pipeline deals | Yes | X deals (W won, L lost, O open) |

Analysis window: [start] to [end]
Data richness: [full/partial/minimal]

Proceed with analysis? (Y/n)

---

Step 2: Analyze Email Performance Patterns

Purpose: Go beyond campaign-level metrics. Find patterns in what makes YOUR best emails work and YOUR worst emails fail. Pure LLM reasoning.

Process

A) Identify Top-Performing Emails

Across all campaigns, find:

• Highest reply-rate subject lines — What do they have in common?
• Highest reply-rate email bodies — What patterns exist?
• Emails that generated positive replies — What specifically triggered interest?
• Emails that generated meetings — The gold standard. What did these say?

For each top performer, extract:

Pattern Element	What to Look For
Subject line structure	Signal-reference? Question? Peer-framing? Length?
Opening line	Lead with them or with you? Signal reference? Question? Statement?
Body structure	How many paragraphs? Proof point placement? Length?
Personalization depth	Tier 1/2/3? What was personalized?
CTA type	Specific ask? Open-ended? Time-bound?
Tone	Casual? Professional? Provocative? Empathetic?
Framework used	PAS? BAB? Signal-Proof-Ask? AIDA?
Proof point type	Customer name? Metric? Case study?

B) Identify Worst-Performing Emails

Same analysis on the bottom performers:

• What do low-reply emails have in common?
• Are there anti-patterns? (Long emails, generic openers, weak CTAs, no proof)
• Is there a consistent flaw across campaigns?

C) Compare Winners vs. Losers

Dimension	Top 20% Emails	Bottom 20% Emails	Gap
Avg word count	[X]	[Y]	[shorter/longer]
Subject line length	[X chars]	[Y chars]	[delta]
Opens with "I" or "We"	[X%]	[Y%]	[self-focused vs. prospect-focused]
Contains proof point	[X%]	[Y%]	[proof usage gap]
Personalization tier	[avg tier]	[avg tier]	[personalization gap]
CTA clarity	[assessment]	[assessment]	[delta]
Has signal reference	[X%]	[Y%]	[signal usage gap]

D) Objection Pattern Analysis

Across all replies classified as objections:

Analysis	What It Reveals
Most common objection	What you're running into most often
Objection by campaign/audience	Is targeting driving objections?
How you handle each objection (from follow-up emails)	Are your handles effective?
Objections that lead to meetings vs. dead ends	Which objections are actually handleable?
Objection language patterns	Exact words prospects use (copy these into your messaging)

E) Reply Sentiment Analysis

Category	Count	%	Trend
Positive interest	X	Y%	+/- vs. earlier campaigns
Warm / curious	X	Y%
Objection (handleable)	X	Y%
Objection (terminal)	X	Y%
Not interested	X	Y%
Auto-reply	X	Y%

Key question: Is the ratio of positive-to-negative replies improving over time? Are you getting better at writing emails that generate interest, or are you stagnating?

Output Contract

email_patterns: {
  top_performers: {
    common_patterns: string[]
    best_subject_lines: [ { subject: string, reply_rate: percentage, pattern: string } ]
    best_openers: [ { opener: string, campaign: string, why_it_works: string } ]
    winning_proof_points: string[]
    winning_ctas: string[]
    winning_tone: string
    winning_framework: string
  }

  bottom_performers: {
    common_anti_patterns: string[]
    worst_subject_lines: [ { subject: string, reply_rate: percentage, issue: string } ]
    common_mistakes: string[]
  }

  winner_vs_loser: {
    dimensions: [ { dimension: string, winners: string, losers: string, gap: string } ]
    biggest_differentiator: string
  }

  objection_patterns: {
    most_common: { objection: string, count: integer, handle_effectiveness: string }
    handleable_objections: [ { objection: string, best_handle: string, conversion_rate: string } ]
    terminal_objections: [ { objection: string, implication: string } ]
  }

  sentiment_trend: {
    direction: "improving" | "stable" | "declining"
    evidence: string
  }

  email_skill_grade: "A" | "B" | "C" | "D" | "F"
  email_skill_summary: string
}

---

Step 3: Analyze Call Performance Patterns

Purpose: If call transcripts or recordings are available, analyze how the user performs on calls. If no call data exists, skip this step.

Process

A) Structural Analysis (From Transcripts)

For each call transcript, measure:

Metric	How to Calculate	What It Reveals
Talk-to-listen ratio	Word count (you) / word count (prospect)	>60% talking = talking too much. Best reps are 40-50% talk.
Longest monologue	Longest uninterrupted stretch by the seller	>90 seconds = you're lecturing, not selling. Keep it under 60s.
Question count	Number of questions asked	Discovery calls should have 10-15 questions. <5 = not discovering.
Question depth	Surface questions ("What do you do?") vs. deep ("What happens when that process breaks?")	Deep questions = strong discovery. Surface = going through motions.
Filler word frequency	Count of "um", "uh", "like", "you know", "basically", "honestly"	High frequency = lack of confidence or preparation.
Call duration	Total time	Too short (<15 min for discovery) = not going deep enough. Too long (>45 min) = not controlling the call.

B) Discovery Quality Analysis

From call transcripts, evaluate:

Dimension	What to Check	Good vs. Bad
Pain discovery	Did you uncover a real pain point?	Good: Prospect describes pain in their own words. Bad: You told them what their pain should be.
Impact quantification	Did you help them quantify the cost of the problem?	Good: "So that's costing you roughly $X/month." Bad: Never quantified.
Decision process	Did you ask who else is involved, timeline, budget?	Good: Clear understanding of BANT. Bad: Left the call not knowing.
Current state	Did you understand how they solve this today?	Good: Know their current tool/process. Bad: No idea what they do now.
Compelling event	Did you identify why now?	Good: Know the trigger (new leader, funding, deadline). Bad: No urgency established.
Next step	Did the call end with a clear, mutually agreed next step?	Good: Specific date/time/action. Bad: "Let me think about it" or "I'll follow up."

C) Objection Handling on Calls

For each objection raised during calls:

Analysis	What to Evaluate
Was the objection acknowledged?	Good: "I hear you, that's a valid concern." Bad: Immediately countering or ignoring.
Was it explored?	Good: "Tell me more about that." Bad: Jumped straight to handle.
Was the handle relevant?	Good: Addressed their specific concern. Bad: Generic response.
Was proof provided?	Good: "Company X had the same concern, here's what happened." Bad: "Trust me."
Did the conversation move forward?	Good: Objection resolved, back on track. Bad: Call stalled or ended.

D) Demo / Presentation Analysis

If demo calls exist:

Dimension	What to Check
Feature dumping	Did you show every feature, or only what's relevant to their pain?
"So what?" test	After showing a feature, did you connect it to their specific need?
Prospect engagement	Did the prospect speak during the demo, or was it a monologue?
Customization	Was the demo tailored to their use case, or generic?
Time on product vs. slides	Were you showing the actual product, or presenting slides?

E) Winning vs. Losing Call Patterns

Compare calls that led to progression (next step, closed won) vs. calls that stalled or lost:

Dimension	Winning Calls	Losing Calls	Gap
Talk-to-listen	[ratio]	[ratio]
Questions asked	[count]	[count]
Pain discovered	[yes/no %]	[yes/no %]
Next step agreed	[%]	[%]
Objections surfaced	[count]	[count]
Call duration	[avg min]	[avg min]

Output Contract

call_patterns: {
  structural: {
    avg_talk_ratio: percentage
    avg_longest_monologue_seconds: integer
    avg_questions_per_call: float
    filler_word_frequency: "low" | "moderate" | "high"
    common_fillers: string[]
    avg_call_duration: float
  }

  discovery_quality: {
    pain_discovery_rate: percentage    # % of calls where real pain was uncovered
    impact_quantification_rate: percentage
    decision_process_mapped_rate: percentage
    next_step_agreed_rate: percentage
    common_gaps: string[]             # What's consistently missed in discovery
  }

  objection_handling: {
    objections_per_call: float
    acknowledgment_rate: percentage   # % of times objection was properly acknowledged
    exploration_rate: percentage      # % of times the objection was explored before handling
    resolution_rate: percentage       # % of times the objection was successfully handled
    weakest_objection_type: string    # Which type of objection you handle worst
    best_objection_type: string       # Which type you handle best
  }

  demo_quality: {
    feature_dumping_detected: boolean
    prospect_engagement_level: "high" | "moderate" | "low"
    customization_level: "tailored" | "semi-generic" | "generic"
    common_demo_mistakes: string[]
  } | null

  winning_vs_losing: {
    key_differences: [ { dimension: string, winners: string, losers: string } ]
    biggest_predictor: string         # Single biggest differentiator between winning and losing calls
  }

  call_skill_grade: "A" | "B" | "C" | "D" | "F"
  call_skill_summary: string
} | null

---

Step 4: Analyze Deal Patterns

Purpose: Look at pipeline data to find patterns in deals you win vs. deals you lose. Pure computation + LLM reasoning.

Process

A) Win/Loss Pattern Analysis

Dimension	What to Compare (Won vs. Lost)
Source channel	Which channels produce deals that close?
Lead persona/title	Which titles convert to closed deals?
Industry	Which verticals are you winning in?
Company size	What size companies do you close best?
Deal size	Do larger deals win more or less often?
Sales cycle length	Do faster deals win more?
Number of touches	How many interactions before close?
Stakeholders involved	Single-threaded vs. multi-threaded
Competitive situation	Do you win more in competitive or non-competitive deals?

B) Velocity Analysis

Pattern	What It Reveals
Stage where deals stall longest	Your bottleneck — need to improve skill at this stage
Stage where most deals die	Your kill zone — what's going wrong here?
Deals that closed fastest	Your "sweet spot" — what do easy wins have in common?
Deals that dragged longest before winning	What made these hard? Can you avoid or accelerate?

C) Activity-to-Outcome Correlation

Activity	Correlation with Winning
More emails before first call	Positive, negative, or no correlation?
Faster first response time	Does speed matter?
Multi-channel (email + LinkedIn + call)	Do multi-channel deals win more?
Number of stakeholders contacted	Does multi-threading help?
Follow-up speed after calls	Does fast follow-up predict wins?

Output Contract

deal_patterns: {
  win_profile: {
    best_source: string
    best_persona: string
    best_industry: string
    best_company_size: string
    avg_winning_cycle_days: float
    avg_touches_to_close: float
    common_traits: string[]
  }

  loss_profile: {
    top_loss_reason: string
    loss_stage: string                # Where deals die most
    common_loss_traits: string[]
    recoverable_losses: string        # "X% of lost deals were timing — could be re-engaged"
  }

  velocity: {
    bottleneck_stage: string
    kill_zone_stage: string
    sweet_spot: string                # Description of your easiest wins
  }

  activity_correlations: [
    { activity: string, correlation: "positive" | "negative" | "none", insight: string }
  ]

  deal_skill_grade: "A" | "B" | "C" | "D" | "F"
  deal_skill_summary: string
}

---

Step 5: Build Skill Assessment & Coaching Plan

Purpose: Synthesize email, call, and deal analysis into a personal sales skill assessment and a specific coaching plan. Pure LLM reasoning.

Input Contract

email_patterns: { ... }              # From Step 2
call_patterns: { ... } | null        # From Step 3 (if calls available)
deal_patterns: { ... }               # From Step 4
user_role: string                    # From config
experience_level: string             # From config
product_description: string          # From config
self_assessed_weakness: string       # From config

Skill Assessment Framework

Assess the user across sales skill dimensions relevant to their role:

SDR Skill Dimensions

Dimension	Data Source	What to Evaluate
Prospecting quality	Email targeting → reply relevance	Are you reaching the right people?
Email copywriting	Email patterns analysis	Are your emails compelling?
Personalization	Email patterns analysis	Do you go beyond merge fields?
Signal recognition	Email campaign angles	Are you using relevant signals?
Objection handling (written)	Reply analysis, follow-up emails	Do you handle objections well in email?
Qualification	Pipeline data — qualified vs. unqualified meetings	Are you booking quality meetings?
Follow-up discipline	Campaign structure, timing, sequence length	Do you follow up effectively?
Volume & consistency	Activity metrics over time	Are you doing enough, consistently?

AE Skill Dimensions

Dimension	Data Source	What to Evaluate
Discovery	Call transcripts — questions asked, pain uncovered	Do you uncover real pain?
Active listening	Call transcripts — talk ratio, prospect engagement	Do you listen more than talk?
Demo effectiveness	Call transcripts — feature relevance, engagement	Do you demo to their pain?
Objection handling (verbal)	Call transcripts — objection resolution rate	Do you handle pushback well?
Negotiation	Deal data — discount patterns, close rates	Do you protect value?
Deal management	Pipeline data — velocity, stage progression	Do you move deals forward?
Multi-threading	Deal data — stakeholders contacted	Do you go wide in accounts?
Closing	Deal data — win rate, next step agreement	Do you close confidently?

Founder Skill Dimensions

All SDR + AE dimensions, plus:

Dimension	Data Source	What to Evaluate
Storytelling	Call transcripts, email copy	Do you tell a compelling product story?
Market positioning	Email angles, competitive mentions	Do you position well against competition?
Adaptability	Variation across campaigns and calls	Do you adapt your approach to different prospects?

Scoring

For each dimension, assign a grade:

Grade	Criteria
A	Top quartile performance. Clear strength. Data supports excellence.
B	Above average. Competent. Minor improvements possible.
C	Average. Functional but not differentiating. Clear room to grow.
D	Below average. Consistent issues visible in data. Needs focused work.
F	Significant weakness. Data shows this is hurting results. Priority fix.

Coaching Plan

For each dimension graded C or below, produce:

Skill: [dimension name]
Current grade: [grade]
Evidence: [specific data points that drive the grade]
Root cause: [why this is happening — not just what, but why]

Coaching recommendation:
1. [Specific, actionable thing to do differently]
2. [Exercise or practice to build this skill]
3. [Example from their own data of when they did this well vs. poorly]

Model to follow:
[Pull from their own top-performing emails/calls as examples.
"Your email in Campaign X did this perfectly — replicate that approach."]

Measurable goal:
[Specific metric to track improvement. E.g., "Increase question count per
discovery call from 6 to 12 over the next 30 days."]

Personalized Playbook

Distill the user's winning patterns into a personal playbook:

## Your Winning Formula

Based on analyzing [X emails, Y calls, Z deals], here's what works for YOU:

### Your Best Email Pattern
Subject line style: [pattern]
Opening approach: [pattern]
Proof point that resonates: [specific proof]
CTA that converts: [specific CTA]
Ideal length: [word count]
Best framework: [framework]