Lumiere Luxe — Shopify Store, Analytics Pipeline & BigQuery Dashboard

Most of my earlier projects ended at the front end — a form, a landing page, a lead captured in a CRM. That covers the marketing ops side of what I do, but not the analytics side I'm building toward. I wanted one project that went further downstream: a real storefront with real tracking, a genuine backend integration rather than a mocked one, and an actual SQL and BI layer on top, all tied to the same fictional business end to end. Lumiere Luxe is that project — a fake beauty brand used to build out a Shopify store, wire up GTM and GA4 properly, prove out a real webhook into a database, and then use that data (plus a synthetic dataset for volume) to write SQL and build a dashboard in BigQuery and Looker Studio.

• A Shopify dev store on the Dawn theme with a custom color scheme, a hero and promo banner section, and a 12-product catalog split across three collections (Skincare, Makeup, Tools) with auto-populating collection pages.
• Full GTM/GA4 tracking across the storefront and checkout: storefront events fire through GTM in theme.liquid as usual, but Shopify's checkout runs in a sandboxed environment where GTM itself won't reliably load — so checkout tracking is implemented as a direct gtag.js call inside a Custom Pixel instead, working around the platform limitation rather than leaving checkout untracked.

• A real webhook integration: Shopify's orders/create webhook fires to a small Flask app, which verifies the request's HMAC signature, parses the order payload, and inserts it into a Postgres database — including nested line items stored as JSONB, since an order can hold a variable number of products.
• The Flask app is deployed to Render with gunicorn as the production server, backed by a hosted Postgres instance on Neon. The first pass ran locally behind ngrok, then Cloudflare Tunnel, to expose it to Shopify — both hit reliability issues on their free tiers, so the app was deployed to a real (free-tier) host instead, which is also closer to how this would actually be run in production.

• Since a dev store only generates a handful of real orders and GA4 events, I generated a larger synthetic dataset in Python/pandas — a set of orders and a matching set of GA4-style events, with purchase events carrying a transaction_id that ties back to a specific order — to get enough volume for the funnel and revenue analysis to actually mean something.
• Both datasets were loaded into BigQuery (ll_orders and ll_ga4_events), separate from the real GA4 export tied to the live storefront, and used to write SQL covering revenue trends, order status, repeat-customer analysis, session and device breakdowns, and a full funnel from session start through purchase.
• A join between the orders and events tables (on transaction_id) answers the one question neither table can answer on its own — revenue by traffic source — and is saved as a BigQuery view so Looker Studio can connect to it directly.

Business questions the dashboard is built to answer:

• What's the monthly revenue and order volume trend?
• What's the average order value for paid orders?
• How many orders are paid, refunded, or cancelled?
• What share of customers are repeat buyers versus one-time?
• Which devices and traffic sources are driving the most sessions?
• Where in the funnel — from session start to purchase — are visitors dropping off?
• Which traffic source is generating the most actual revenue, not just sessions?