Docs

Tracking Campaigns Triggered From Your Data Warehouse

Braze's zero-copy Canvas triggers now fire campaigns directly from warehouse data — track the source computation or you can't explain why a message sent.

Engagement

Customer engagement platforms are collapsing the pipeline between your data warehouse and a live campaign send. Braze’s Cloud Data Ingestion now connects directly to Snowflake, BigQuery, Databricks and Redshift, and its zero-copy Canvas triggers can fire a campaign the moment a computed field changes in the warehouse — no Hightouch or Census reverse-ETL step in between. That’s a genuine latency win: a churn-risk score computed overnight can trigger a retention message within minutes instead of waiting for the next sync window. But it also means a campaign can now be caused by a dbt model run, a warehouse job, or a batch recompute that nobody on the marketing team saw happen. When someone asks “why did this user get this message,” the answer used to live in a segment definition. Now it might live three hops upstream in a SQL transformation.

The risk is specific to zero-copy activation. Traditional reverse-ETL syncs left an audit trail — a sync job, a destination table, a timestamp you could inspect. A zero-copy trigger reacts to a field changing value, which means the causal chain runs from warehouse compute job, through the field update, through the trigger condition, to the send — and if you’re only logging the send, you’ve lost everything upstream of it. Without that chain, a spike in a specific Canvas’s send volume looks identical whether it’s real user behaviour or a bug in a warehouse transformation that flipped a computed field for the wrong cohort.

Data Points to Track

  • Triggering field and its value at fire time, not just the campaign and user ID, so you can reconstruct which computed attribute crossed the trigger condition
  • Warehouse job or model run ID, captured via your dbt/orchestration metadata and joined back to the trigger event, linking a send to the specific transformation that produced it
  • Trigger latency, the gap between the warehouse field update timestamp and the campaign send timestamp, to catch sync delays or unexpected batching
  • Field change magnitude, where relevant (e.g. a churn score moving from 0.3 to 0.9 versus 0.79 to 0.81), since threshold triggers are sensitive to noise near the boundary
  • Downstream engagement outcome per trigger source, so you can tell whether a specific warehouse model’s triggers convert differently from another’s

Setup Steps

  1. Tag every Canvas trigger with its source field and computation name in Braze before turning on zero-copy activation, so the trigger event itself carries enough metadata to trace backwards.
  2. Log warehouse job completion events (model run ID, row count, affected user count) to your own analytics store, independent of Braze, so you have a warehouse-side record to join against.
  3. Join trigger fire events to job run IDs on a schedule, building a table that answers “which job caused this send” without needing to open both systems manually.
  4. Alert on trigger volume anomalies — a Canvas that normally fires for 200 users a day suddenly firing for 20,000 is very likely a warehouse transformation bug, not a genuine behavioural shift.
  5. Add a dry-run stage for new zero-copy triggers where the computed field updates and gets logged but the Canvas doesn’t send, letting you validate the trigger logic against real data before it reaches users.

Actionable Insights

A trigger volume spike that correlates with a specific warehouse job run ID, rather than with any change in user behaviour, is a strong signal the bug is upstream in the transformation — not in the campaign logic, and not worth debugging in Braze at all. Long trigger latency that grows over time usually means the warehouse sync or the computed field’s refresh schedule has drifted, and it’s worth checking before assuming the Canvas configuration is at fault. And when two different warehouse models feed triggers into the same Canvas with meaningfully different downstream engagement rates, that’s a case for splitting them into separate, independently measurable triggers — folding them together hides which computation is actually producing valuable sends versus noise.

Expert help

Need help tracking this in your app?

Our team sets up analytics pipelines for mobile and web teams every day. Talk to us and get your first events flowing in under an hour.

Talk to an expert