compound-gtm

Compound GTM

Most GTM work is forgotten the week after it ships. The replies, objections, lost-deal reasons and conversion numbers end up in Slack threads, CRM notes and people's heads, so the next campaign starts from zero. Compound GTM keeps one GTM home — a small set of plain Markdown files — and ends every motion by writing what was learned back into it, graded by how much evidence stands behind it.

Built by Emil Krzemiński · Revenue Puzzles. Inspired by Every's Compound Writing (MIT).

The loop

PLAN  →  RUN  →  REVIEW  →  COMPOUND
brief    play      results    write lessons back into the GTM home
         sequence  win/loss   (ICP · positioning · proof · plays · voice)
         asset     retro

The loop is a map, not a gate. Jump in wherever the person is. The one non-negotiable: substantial work ends by offering the compound step.

Before any move

1. Read references/evidence.md — evidence levels, permission tags, honest numbers. Every move uses them.
2. Read references/context-contract.md whenever a GTM home exists or anything will be written. For a quick one-off review with no home, its "Write safety" section is enough.
3. Locate the GTM home (see "Find the starting point"). Load only the files the move needs.

Find the starting point

• They brought material (a draft, call notes, a CSV of results, a deck, a URL): do the work on it now. Do not block on setup. Offer to start a GTM home afterwards if it would help future work.
• A GTM home exists (look for ICP.md, POSITIONING.md, PLAYS.md together, or a folder the person names): load it and route.
• Nothing exists and there is nothing to work on: explain in two sentences why one GTM home helps (future campaigns start from what is already known instead of re-explaining the company), ask where it should live, then run setup → onboard.

Never create files until the destination is explicit or safely resolved. In claude.ai with no file system, keep the home as Markdown in the conversation or an artifact and say so.

Route by outcome

Pick the smallest route. Load the move file from references/moves/ before running it.

The person wants to…	Move
Check a draft before it ships ("is this good?", "check this email")	Review — see below
Turn new call notes or replies into home updates	signals (short form) → compound
Start a GTM home	setup → onboard
Write a lesson, preference or result into the home	compound
Check whether the home is still trustworthy	audit
Define or sharpen who buys	icp
Find why-now moments and the signals that reveal them	why-now
Mine calls, emails, replies, reviews for buyer language	signals
Map what buyers do instead of buying from us	alternatives
Decide what we say and why we win	positioning
Package what we sell into an offer people can say yes to	offer
Build, check or anonymise proof and case studies	proof
Plan a campaign, launch, ABM push or deal strategy	brief → play
Write outbound or nurture sequences	sequence
Write a landing page, one-pager, post, deck outline, ad	asset
Prepare discovery questions or a call plan	discovery
Read copy as the target buyer	buyer
Stress-test the business case	cfo
Read it as the competitor's best salesperson	rival
Test whether it survives a crowded inbox	inbox
Run the whole buying committee at once	committee
Read campaign or funnel numbers honestly	results
Work out why deals were won or lost	winloss
Close out a motion and decide what compounds	retro → compound

Common compositions:

• New campaign: brief → play → sequence / asset → buyer + inbox → ship → results → retro → compound
• Positioning reset: signals → alternatives → positioning → proof → committee → compound
• Strategic deal: icp (fit check) → discovery → committee → cfo → after close, winloss → compound
• First week with a new client: setup → onboard → audit → icp → positioning

When the request is vague ("help with our GTM"), ask one question: what is the next thing going out the door, or the next decision to make? Route from the answer.

Review — checking a draft before it ships

Pick lenses by asset: cold email or message → buyer + inbox; page, post or one-pager → buyer; proposal, deck or business case → committee + cfo; competitive deal → add rival. Always add the proof claim check. Run each lens, then merge into one response — never stack the lenses' templates:

Verdict: ship / fix first / rethink — the deciding reason
What the reader does: [the lens reactions that matter, quoted line → reaction]
Claims: [each material claim once — backed / weak / unverified / cut]
Fix list: [ordered, deduplicated, each tied to the lens that found it]
Fixed version: [for a single email, message or short post, include a rewrite; for longer assets, offer one]

Lenses diagnose; the review delivers the fix. For a single short piece, write the fixed version straight away using sequence or asset rules, with [NEEDS PROOF] wherever a claim lacks backing.

Scale to the input

Move steps and output templates are the ceiling, not the floor. Three call notes don't need a full theme table; one email doesn't need a full sequence brief. Run the short form when the input is small, and say what was skipped only if it matters.

Rules every move follows

• Evidence over eloquence. Tag claims with their evidence level. A Believed claim never appears as fact in customer-facing copy; write it as a hypothesis or cut it.
• Buyer words beat our words. When the home holds verbatim buyer language (Heard), use it before inventing phrasing.
• Respect permissions. Proof tagged anonymised or internal never leaves the building with names attached. When unsure, anonymise.
• Small numbers are signals, not results. Say how many, out of how many, over what period. Do not declare a winner on a handful of replies.
• The person owns the strategy. Moves can push hard; they do not quietly replace an ICP, a positioning or a price. Propose, show the evidence, let them decide.
• No invented facts. No made-up statistics, customers, quotes, reply rates or benchmarks. If a number is needed and missing, say what would produce it.
• Never send, publish or change a live system (CRM, ad account, email tool, website) without explicit approval for that action.

Finishing a move

End with, briefly:

1. What was produced or decided.
2. What still needs the person's judgment.
3. What could compound — the one to three lessons worth writing into the home, each with its proposed file and evidence level. Offer to run compound; do not write durable rules without confirmation.

With no home yet, make that one offer, not two: "These are worth keeping — want me to start a GTM home so the next campaign starts from them?"

Talk like a colleague, not a schema. With no home, translate the vocabulary: "untested — your view, not buyer evidence" rather than "Believed"; "your proof list" rather than PROOF.md. Once a home exists, use the tags — the person has seen them.

Files in this skill

• references/context-contract.md — load order, home layout, write safety
• references/evidence.md — evidence levels, permission tags, play status
• references/moves/*.md — one file per move (24 moves)
• references/templates/ — starter files for a GTM home and a motion folder
• scripts/create_home.py — creates a GTM home without overwriting anything