Skip to content

method.md

Latest commit

 

History

History
204 lines (148 loc) · 6.28 KB

method.md

File metadata and controls

204 lines (148 loc) · 6.28 KB

GTM Repository Method

The Core Problem This Solves

Most go-to-market teams suffer from knowledge fracture:

  • Each campaign starts from scratch
  • Insights live in someone's head or scattered Notion docs
  • Call transcripts are searchable but not synthesized
  • "What worked last time?" requires asking around

This repository methodology creates a compounding knowledge system where every campaign makes future campaigns smarter.

The Mental Model

Repo-Level: Durable Foundation

The source/ directory contains your institutional knowledge:

  1. ICP Definition (source/icp/)

    • Who you sell to
    • How you win
    • What you don't do (exclusions)
    • Your positioning

  2. Persona Taxonomy (source/personas/)

    • Champion, blocker, economic buyer, technical buyer, end user
    • Labels tied to real people in your CRM
    • How each persona thinks about your solution

  3. CRM Data (source/crm/)

    • Historical ARR and deal data
    • Account and contact information
    • Exported regularly with timestamps
    • Derived datasets (deals_with_arr, accounts_enriched)

  4. Call Transcripts (source/transcripts/)

    • Tied to deals via join keys (Salesforce ID, etc.)
    • Normalized format for analysis
    • Raw exports preserved for auditing

  5. Enrichment Data (source/enrichment/)

    • Industry, headcount, tech stack
    • From Clay, Clearbit, ZoomInfo, or similar
    • Refreshed periodically to stay current

  6. Founder Context (source/founder_notes/)

    • Audio notes about the market (transcribed)
    • Strategic context that isn't in transcripts
    • The "why" behind product and market decisions

Analysis: Pattern Recognition

The analysis/ directory is where you synthesize insights:

  1. Segmentation (analysis/segmentation/)

    • Segment hypotheses (e.g., "mid-market fintech 200-1k headcount")
    • Rules for bucketing accounts (SMB, mid-market, enterprise)
    • Evidence from past wins

  2. Situational Changes (analysis/situational_changes/)

    • What changed in their business that made them buy NOW
    • Common triggers: funding events, new hire, compliance deadline, tool failure
    • Evidence table with citations back to transcripts

  3. Win/Loss Themes (analysis/win_loss/)

    • Why you win (differentiators)
    • Common objections and how they're handled
    • Where you lose and to whom

Campaign-Level: Execution with Context

Each campaign in campaigns/ is a working set:

  1. Pull Context from source and analysis

    • Define target segment and personas
    • Pull relevant customers who match the profile
    • Reference past wins in this segment

  2. Do Campaign-Specific Research

    • News about target accounts
    • Funding events, leadership changes, tech stack shifts
    • Store in research/snapshots/YYYY-MM-DD/

  3. Generate Messaging Hypotheses

    • What situational change are we responding to?
    • How do we position against alternatives?
    • What's the hook in the first email?

  4. Execute and Track

    • Outreach log (who, when, which sequence)
    • Replies and outcomes (JSONL for easy append)
    • Decisions made during the campaign

  5. Feed Learnings Back

    • What worked? What didn't?
    • New objections discovered?
    • Should this segment be refined?
    • Write results/learnings.md and roll it into analysis/ or source/

Key Design Choices

1. File-Based, Not Database

Everything is CSV, JSON, JSONL, Parquet, or Markdown. Why?

  • Versioned with Git
  • Easy to diff and review
  • No database to maintain
  • Works with any data tool (pandas, duckdb, Excel)

2. Dated Snapshots

Raw exports and research are timestamped:

  • crm_deals_2026-01-10.csv
  • research/snapshots/2026-01-10/

This preserves historical context. You can recreate analysis from any point in time.

3. Normalized + Raw

Keep both:

  • Raw: Exactly what came from the source system
  • Normalized: Cleaned, joined, standardized schema

Raw is for auditing. Normalized is for analysis.

4. Join Keys Everywhere

Document identifiers in source/joins/keys.md:

  • Salesforce Account ID
  • Salesforce Opportunity ID
  • Gong Call ID
  • Email hash (for privacy-preserving joins)

Every dataset should be joinable.

5. Prompts Are Code

Reusable prompts live in prompts/:

  • segmentation.md - How to segment customers
  • situational_change_extraction.md - How to find "what changed"
  • messaging_generation.md - How to draft sequences

Treat prompts like code: version them, review them, improve them.

The Compounding Loop

1. Ingest CRM/transcripts/enrichment → source/
2. Run analysis → analysis/
3. Start campaign with context → campaigns/YYYY-MM-DD_name/
4. Execute campaign → results/
5. Extract learnings → results/learnings.md
6. Roll learnings back → analysis/ and source/
7. Repeat (with more context than last time)

Each loop:

  • Improves your segment definitions
  • Surfaces more situational changes
  • Refines your messaging
  • Updates your ICP understanding

Success Metrics for This Methodology

You know this is working when:

  • New campaigns reference 3+ past campaigns in their brief
  • Segment definitions have evidence citations
  • Messaging hypotheses cite specific transcripts
  • Win/loss themes are updated monthly
  • Founders can query this repo to understand the market

Anti-Patterns to Avoid

❌ One-off analysis in a notebook that never gets saved

  • Always export to analysis/ or outputs/

❌ Campaigns that don't write learnings.md

  • Knowledge dies if not recorded

❌ Stale source data

  • Set calendar reminders to refresh CRM and enrichment

❌ No join keys

  • Document how datasets connect, or you'll lose ability to cross-reference

❌ Prompts scattered in ChatGPT history

  • Copy working prompts into prompts/ immediately

Getting Started (Minimal v0)

If you're starting from scratch:

  1. Export your CRM → source/crm/exports/
  2. Define your ICP → source/icp/icp.md
  3. List your personas → source/personas/personas.yaml
  4. Pick one segment to test → analysis/segmentation/segments.md
  5. Run one campaign → campaigns/2026-01-10_test/
  6. Write what you learned → campaigns/2026-01-10_test/results/learnings.md

Then repeat with more context each time.

The goal: Make it impossible to do one-off work. Every action should make the next action easier.