This hub covers the concepts needed to build and maintain a Meta Ads to BigQuery pipeline: the nested actions array, attribution data that retroactively updates, token expiration that silently breaks pipelines, and measurement methodology changes in 2021 and 2025.
The API
Meta Ads Insights API Structure — How the Marketing API is organized: the five-level object hierarchy, the Insights “edge,” versioning cadence, System User authentication, and rate limits.
The API exposes 70+ fields. Most complexity lives in one: actions. Conversion data returns as a nested JSON array, not flat columns; flattening it is where most transformation effort goes.
Handling the Actions Array
Meta Ads Actions Array in BigQuery — UNNEST patterns in BigQuery, pairing actions with action_values for revenue data, making action types configurable through dbt variables, normalizing for cross-platform reporting, and detecting unmapped action types. Covers both the raw SQL pattern and the dbt-integrated approach.
Attribution Windows
Meta Ads Attribution Windows — How Meta’s attribution windows work, which windows survived the January 2026 deprecation (1d_click, 7d_click, 28d_click, 1d_engaged_view, 1d_view), and what the June 2025 on-Meta/off-Meta attribution split means for warehouse data. On-Meta events (lead forms) now attribute to impression time; off-Meta events (website purchases) attribute to conversion time.
iOS Signal Loss
iOS 14.5 Signal Loss and Meta Measurement — Apple’s ATT rollout in 2021 permanently changed what Meta can measure. IDFA availability collapsed to ~6%. Default attribution windows shortened. Meta increasingly reports modeled (estimated) conversions rather than deterministic ones. Aggregated Event Measurement introduced a 72-hour delay for iOS conversion data.
Understanding this context explains why recent data always changes, why reach can’t be summed across breakdowns, and why implementing the Conversions API recovers 15-25% of attributed conversions compared to Pixel-only measurement.
Keeping It Running
Meta Ads Pipeline Maintenance — Operational patterns for production pipelines: System User token setup (tokens that don’t expire), spend reconciliation against Ads Manager, API version lifecycle management, circuit breaker implementation to prevent rate limit exhaustion after consecutive failures, and raw data storage before transformation.
Related Garden Notes
These notes from the broader advertising data cluster connect to this pipeline directly:
- Ad Platform API Landscape — Cross-platform API comparison; Meta section covers the actions array at a high level
- Ad Pipeline Engineering Challenges — Rate limits, schema changes, attribution window normalization, and privacy regulation across all platforms
- Ad Data Extraction Tools — Decision framework for Fivetran vs. Airbyte vs. dlt vs. custom Python vs. BigQuery DTS
- dbt Ad Reporting Patterns — Cross-platform UNION patterns and the
dbt_facebook_adspackage from Fivetran - Meta CAPI Server-Side Setup: Deduplication and Event Match Quality — Conversions API implementation including event deduplication and user data mapping for Event Match Quality
- Late-Arriving Data and the Lookback Window Pattern — The dbt pattern for handling Meta’s retroactive attribution updates
Source Article
Meta Ads to BigQuery: a practical pipeline guide — end-to-end pipeline walkthrough with extraction tool selection and specific configurations for BigQuery and dbt.