An analytics stack should be a tool, not a tax. The mission is to understand what to build next and who to talk to. Here is a pragmatic architecture you can stand up in two evenings that keeps you honest without drowning you in dashboards.

Layer 1 — Capture the truth at the source

Start with your Astro frontend and API routes. Emit a single track helper that sends JSON payloads to Cloudflare Workers or a Supabase Edge Function. Each event should include:

{
  event: 'Project Created',
  userId: 'user_123',
  plan: 'pro',
  occurredAt: new Date().toISOString(),
  value: 1
}

Store events in a Postgres or Tinybird table keyed by user. The schema can be schemaless JSONB, but add generated columns for event name and week number so queries stay cheap.

Layer 2 — Normalize in a spreadsheet

Export the week’s events into Google Sheets using Airbyte, Coupler.io, or even a daily CSV sync you trigger manually. Create one tab per metric:

  • Activation (project_created within 7 days of signup)
  • Expansion (invoices with amount > plan_price)
  • Retention (logins in the last 28 days)

Use array formulas to bucket cohorts. The goal is fast iteration, not perfect SQL.

Layer 3 — Review with a Friday operating memo

Numbers matter only when they change your calendar. Every Friday, open your sheet and answer four questions in a running doc:

  1. What moved up or down this week?
  2. Why do we think it happened?
  3. What experiment or conversation follows?
  4. What will we review next Friday to know if it worked?

Paste charts sparingly. Focus on narrative and commitments. This memo becomes the spine of your investor updates, team syncs, and personal focus.

Layer 4 — Teach your customers with the data

Share one anonymized insight from your data every month in your blog or newsletter. Teaching with proof builds trust, attracts the right customers, and keeps you accountable to outcomes.

Once this stack feels cramped, you will know exactly which step fails you. Until then, resist the allure of warehouses and BI suites. Depth over breadth: collect only the data that helps you change what you build next week.