Launch Become Data-First

Data Warehouse & AI Analyst

One home for every number, one mouth that answers in plain English — stood up in a weekend.

The Four-Layer Data Stack

Sources

Every system holding a number you care about. Product DB, Stripe, ad platforms, support, analytics, CRM.

Ingestion (ELT)

Pipes move raw data into the warehouse on a schedule. Extract, Load, then Transform inside SQL.

Airbyte default Fivetran if zero-ops

Warehouse

One database optimised for analytics. Single source of truth, joined against the same customer_id.

BigQuery default

Consumption

BI dashboards for known questions, AI analyst for ad-hoc ones. Three dashboards: daily, weekly, monthly.

Metabase + AI analyst

Warehouse Comparison

Warehouse Best For Pricing Watch Out

BigQuery GCP, ad data, zero ops Free 1TB / mo SELECT * cost spikes

Snowflake Heavier B2B, $10M+ path ~$2-4 / credit Auto-suspend burns

Postgres Sub-10GB, already pay Bundled Fights prod DB for CPU

Five Modelled Tables (Never Query Raw)

dim_customers One row per customer, joined across product + Stripe + support.

fct_subscriptions Subscription changes. Source of truth for MRR, ARR, churn.

fct_events Product events: signup, activation, key feature use.

fct_ad_spend Daily spend per platform per campaign. CAC inputs.

fct_tickets Support interactions. Response time, satisfaction, ticket-to-customer ratios.

AI Analyst

Plain English, 10-Second Answers

BIRD benchmark: ~76% strict accuracy on messy schemas. On well-modelled tables with dbt descriptions, production-reliable. Always show SQL, always verify row counts. Read-only role, no exceptions.

Metabase AI Default BI + analyst

Worked Example

Maya at Bridgewell, $42K MRR

4 sources to one warehouse in a weekend. Monday: types "blended CAC by channel?" into AI analyst, gets $184 Google / $42 SEO / $620 LinkedIn in 8 seconds. Kills LinkedIn spend. Decision that used to take 3 days.

$25 / month total cost