What’s the minimum viable setup to stop most analytics outages?

Add contracts at ingestion, 5 dbt/GE/Soda checks on your top 10 tables, and a publish gate that fails if SLIs breach SLOs (freshness, completeness, uniqueness). Wire alerts to PagerDuty with a runbook. That gets you 80% of the benefit quickly.

Do we need to buy a data observability tool?

Not necessarily. Start with dbt/GE/Soda + OpenLineage + Grafana/Prometheus. If your estate is massive or multi-tenant, tools like Monte Carlo or Databand can help with automated anomaly detection and ownership workflows.

How do we handle schema changes from upstream teams?

Use versioned data contracts. Bet on additive changes only; require a review for breaking changes. Implement a CI check that blocks merges when contracts change without version bumps. Run both v1 and v2 during a migration window and backfill.

We have AI-generated ETL code that keeps drifting. Can this help?

Yes. Enforce contracts in CI, add schema compatibility checks, and run quality suites on every PR. GitPlumbers does AI code refactoring and vibe code cleanup to make AI-generated code safe before it hits production pipelines.

How do we keep costs in check for large datasets?

Use partition pruning, metadata-based checks (row counts from `_PARTITIONS_SUMMARY`/`INFORMATION_SCHEMA`), approximate functions, and canary sampling. Only run full scans on nightly windows or after anomalies.

Data-engineering · Dec 7, 2025 · 10 minute read

The Night the CFO’s Dashboard Went Dark: Building Data Quality Gates That Actually Prevent Analytics Failures

If your “source of truth” is lying by 2%, the business will find out before you do. Here’s how to build data reliability into the pipeline so downstream analytics don’t crater at 2 a.m.

Back to all posts

The Night the CFO’s Dashboard Went Dark: Building Data Quality Gates That Actually Prevent Analytics Failures

Key takeaways

Implementation checklist