Toast POS ETL: Extracting Menu Data Into Your Analytics Warehouse

Quartino, a 120-seat Italian restaurant in Chicago's River North, hit the ceiling of Toast Analytics' prebuilt dashboards. The CFO wanted to merge dine-in, takeout, and Caviar delivery data into one BigQuery warehouse with a dbt semantic layer. We dissect the ETL pipeline, real cost math, and the honest answer to "should we move off Toast Analytics."

Airbyte Connector and Toast API Extraction

Quartino runs an Airbyte Cloud Toast POS connector that pulls /orders, /menus, /labor, and /cash-management endpoints nightly at 02:00 EST. Average sync time is 7 minutes 12 seconds with a 24-hour loss-tolerance window.

Toast API's 1,000-requests-per-hour rate limit pushed the batch size to 500. Failed syncs alert PagerDuty — only 2 retries in 90 days, zero data loss.

dbt Models and Semantic Layer

BigQuery has a raw_toast dataset for Airbyte landing; stg_toast_* staging handles type casts and UTC normalization; marts/finance and marts/menu_engineering house business logic. dbt tests cover not_null, unique, relationships, and a custom freshness check.

• fct_daily_sales — 14 dimensions, 28 metrics, $0.04/day BigQuery slot cost.
• dim_menu_item — SCD Type 2 tracking price-change history.
• fct_menu_engineering — Stars/Plowhorses/Puzzles/Dogs matrix refreshed weekly.

POS-Only vs Warehouse Analytics

Toast Analytics' strength: zero IT investment, manager panel up in 5 minutes, real-time intraday. Its ceiling: no cross-channel joins, no custom cohorts, no ML feeding. Quartino couldn't pipe delivery-marketplace data into Toast — that triggered the ETL decision.

Monthly stack cost: Airbyte Cloud $120, BigQuery storage + compute $45, dbt Cloud Developer $100 — total $265/month. The freed-up part-time analyst hours saved $1,800/month. ROI landed in month 7.

FAQ

Should I cancel Toast Analytics entirely? No. Front-of-house managers keep using Toast for intraday; the warehouse handles strategic analysis and CFO reporting.

Fivetran or Stitch instead of Airbyte? Both work. Fivetran is more stable but its Toast connector is ~40% pricier. For 1-2 locations Airbyte wins; 10+ locations tilt to Fivetran.

Snowflake instead of BigQuery? At restaurant volumes the difference is marginal. BigQuery's on-demand pricing wins at low volume; Snowflake's compute separation matters past 5 TB.

Found this helpful? Share it.