Compound Marketing Starter Kit

Your last campaign should make the next one smarter.

Most marketing teams have amnesia. Every campaign starts from a blank page, every messaging test result dies in a Slack thread, every customer call full of the exact words your buyer uses evaporates the moment it ends. You pay full price for the same lesson twice.

Compound Marketing is the fix: a simple loop that captures what you learn in a durable, structured, retrievable place, so each cycle starts ahead of the last one. Borrowed from the way engineers do compound engineering, pointed at GTM.

This repo is the starter kit: three ledgers and four prompts. Clone it, run one campaign through the loop, and your campaigns start talking to each other.

---

The loop

   ┌──────────────────────────────────────────┐
   │                                            │
   ▼                                            │
 PLAN  ──────►  WORK  ──────►  REVIEW  ──────► COMPOUND
(brief it,    (ship the     (what won,      (write the
 pulling      campaign,      what lost,       lesson back
 past         asset, test)   what            into a place
 lessons                     surprised       the next PLAN
 first)                      you)             reads from)

The first three steps you already do. Compound is the one everyone skips, and it's the only one that turns four boxes into a flywheel.

A normal retro fails to compound on three counts. It's unstructured, so you can't find the one relevant lesson when you need it. It's not retrievable at planning time, so even when the lesson exists nobody pulls it. And it's not machine-readable, so you can't put an agent on it. The ledgers in this repo fix all three.

---

What's in here

compound-marketing-starter/
├── README.md                  # you are here
├── ledger/
│   ├── messaging-ledger.md     # value props tested, won/lost, by segment
│   ├── voice-of-customer.md    # verbatim buyer language from calls/reviews
│   └── campaign-ledger.md      # hook + format + channel + result
└── prompts/
    ├── 1-plan.md               # reads the ledger before briefing
    ├── 2-work.md               # drafts the asset in the buyer's words
    ├── 3-review.md             # turns results into a structured review
    └── 4-compound.md           # writes the lesson back into the ledger

The ledgers are the asset. Campaigns are disposable. The ledger is the thing that's worth more every quarter, because it's the only file in your stack that compounds.

---

Quickstart

With Claude Code (recommended)

1. Clone this repo and open it in Claude Code.
2. Plan: paste prompts/1-plan.md, fill in your campaign and segment. Claude reads /ledger and briefs you with "what we already know" up top.
3. Work: paste prompts/2-work.md to draft the asset, pulling real buyer language from voice-of-customer.md.
4. Ship it. Collect results.
5. Review: paste prompts/3-review.md with your numbers to get a structured review (worked / didn't / surprised).
6. Compound: paste prompts/4-compound.md with that review. Claude returns ledger rows ready to paste. Paste them. You're now smarter for next time.

Not a Claude Code user?

The four prompts are plain markdown skills, so they run in any agent. Drop them into Claude Cowork (a desktop app, no terminal, built for non-technical work), OpenAI Codex (as SKILL.md skills with the rule in AGENTS.md), or Google Antigravity (markdown skills). Full runtime-by-runtime setup is in IMPLEMENTATIONS.md.

By hand (no AI needed)

The ledgers are just markdown tables. Fill a row in after every campaign, messaging test, and customer call. Read the relevant rows before you brief the next thing. The discipline is the whole point; the AI just makes the retrieval automatic.

---

The one rule

THE COMPOUND STEP IS NON-NEGOTIABLE.

A campaign that ships without writing its lesson back into a ledger taught you nothing you'll keep. Skip Plan if you're in a hurry. Never skip Compound.

---

Built in public as part of The Traction Point. Steal it, fork it, tell me what you changed.