Data Quality Checklist Template

Monthly for critical datasets (those that feed dashboards viewed by executives or models that drive automated decisions). Quarterly for secondary datasets. After any major pipeline change, schema migration, or source system update, run an ad-hoc audit regardless of schedule. The [Product Analytics Handbook](/analytics-guide) covers building data quality into your analytics practice.

It depends on the field and its use. Required identifiers (user_id, event_date) should have 0% nulls. Optional fields that not all users populate (nps_score, company_name) may have high null rates by design. Fields used as ML features should have null rates below the imputation tolerance of your model. Document the acceptable threshold per column in this checklist.

The team that produces the data owns its quality. The data engineering team owns pipeline reliability. The product team owns event instrumentation correctness. The analytics team owns derived table accuracy. If nobody owns quality for a dataset, it will degrade. Assign a specific individual, not a team, as the data steward for each critical dataset.

Rank by downstream impact. A failing check on a field used by an executive dashboard or a production ML model is critical. A failing check on a field used by one analyst for ad-hoc queries is low priority. Use this formula: Impact = (number of consumers) x (severity of wrong decisions if this data is incorrect). The [AI Readiness Assessment](/tools/ai-readiness-assessment) evaluates data quality as a dimension of ML readiness.

Yes, after your first manual audit. The manual audit identifies which checks matter and what the thresholds should be. Then automate those checks using tools like Great Expectations, dbt tests, Monte Carlo, or custom SQL assertions. Automated checks run on every pipeline execution. Manual audits run monthly to catch issues that automated checks miss (accuracy spot-checks, cross-system reconciliation). ---