Skip to main content
Synced from the repo — do not edit here

Canonical source: docs/claude/gtm-principles.md. This page is generated by docs/scripts/sync-handbook.mjs. Edit the source file in the repo; changes appear here on the next build.

GTM Principles & KPI Decisions (distilled)

2026-07-12, from George's article batch #2 (AI-visibility audit; "27 things to $10k MRR"). Companion to growth-signals-playbook.md. Rule of use: these are PRINCIPLES to bake into planning/prompts and a small number of CONCRETE tools/KPIs — not 27 new projects.

A. AI-visibility audit (GEO measurement — the missing half of #209)​

We shipped the GEO foundations (llms.txt, Content-Signals, Link headers, markdown negotiation, pricing gate). What we never built is the measurement loop:

  1. Ask ChatGPT / Claude / Perplexity / Gemini the buying-intent questions our ICP asks ("best all-in-one platform for a longevity clinic", "Healthie alternatives with AI", "software to run a med-spa with insurance billing") — several times each (answers vary).
  2. Record: are we recommended? who is? which sources do the answers cite (news, Reddit, LinkedIn, directories, review sites)?
  3. Gap map = the cited sources we're absent from, ranked by citation frequency × attainability. That ranked list IS the content/placement backlog.
  4. Re-run monthly → trend our share-of-recommendation vs competitors.

This is a natural market-intel sibling (same web-research machinery, pointed at LLM answers about US instead of competitor sites). Feeds Master Checklist items "SEO/GEO foundation (measure LLM citations)" and "Category leadership in analyst/LLM answers".

B. SaaS growth principles — the ones we ADOPT (mapped to our stack)​

KPIs to instrument (real decisions):

  • Activation rate, not signups — define the aha-path (setup → first booking → first Atlas action), instrument it, put it in the weekly COO brief. (Checklist item exists; this makes it the #1 growth KPI.)
  • PCPL for outbound (from the signals playbook).
  • Churn exit interview — every cancellation triggers a personal-tone email asking why (cheapest research; sometimes a save). Fold into the churn-saves playbook item.

Tactics we adopt (attach to existing checklist items — no new projects):

  • Pricing page is a product — reviews/proof ON the pricing page (not just homepage testimonials); people read reviews at the moment of deciding to pay.
  • Annual upfront — already in the pricing model ($8,495 yr-1); push it in sales motion for cashflow.
  • Affiliate 30% recurring — when the partner program launches, don't cheap out; recurring % is what makes affiliates grind. (Partner item exists.)
  • Integration marketplaces as discovery — one good integration listing (EHR/calendar/Stripe app marketplace) can out-acquire ten native features. (Marketplace-listing item exists.)
  • 60-second demo video, not 6 minutes — direct tie to the launch-video engine (#264): the v1 clip length target is ≤60s.
  • Public changelog weekly — already our Loop A discipline; keep it.
  • Community as retention moat — checklist item exists (clinic-owner network).

Principles for Atlas growth-arm prompts:

  • "Distribution beats features at this stage: default time split 80% distribution / 20% build when at-parity." (For the operator's prioritizer, weigh growth candidates accordingly.)
  • "Speed-to-feedback beats polish; charge from day 1; talk to users before building."

Explicitly NOT adopted: niche-down-tiny (we're already vertical), free trials debate (pricing model is set: no free tier), Product-Hunt-day obsession.

C. Continual learning for agents (Replit) — validation + the two gaps​

The article's frame: agents improve at 3 layers — model / harness / context. We don't own frontier weights, so ALL our learning is harness + context. Validation: our architecture already matches their loop — eval harness (#183) ≈ offline benchmarks; goal-gate (#157); learning digest + intake ≈ their self-improvement loop with human launch-gates; their "humans keep: hypothesis selection, eval curation, launch approval" = exactly our George-gates.

Gap 1 — Telescope-style trace clustering (adopt): we capture signals (atlas_signals, knowledge-gaps, buildfail labels) but cluster only weekly-and-coarsely. Their pattern: summarize failure trajectories → embed → density-cluster (HDBSCAN) → emergent issue groups → the clusters file the work. That's the systematic version of our knowledge-gap digest, and the engine #158's "correlate" needs. Filed as its own issue.

Gap 2 — A/B layer for agent changes (defer): prompt/tool/harness changes A/B'd on live traffic with clean attribution. Right for us LATER (needs traffic volume + metric plumbing); noted on the operator plan, not scheduled.

Principle to keep taped to the wall: "Eval curation shapes the hill the agent climbs — if the eval rewards the wrong behavior, the loop faithfully optimizes the wrong thing."