When dashboard figures do not reconcile, where should you start? Separate metrics measured by Snowflake DMFs from business-specific reconciliation, then define thresholds, response ownership and acceptance tests.
Dashboard revenue does not match the source business system. A successful load still leaves some of yesterday’s orders missing. Job status alone will not identify these problems. Start by defining which data must be available, by when, and in what condition before the business can use it.
Begin with missing required fields, duplicate business keys, arrival of required data and source reconciliation. Prioritize reports used for decisions or billing instead of monitoring every table at the same frequency. This also makes response ownership and priorities easier to establish.
What Are Data Quality Checks, and How Do They Differ from Job Monitoring?
Job monitoring asks whether a load completed; data quality checks ask whether the loaded data is fit for use. Successful loads can contain missing order identifiers or duplicate orders. Lower row counts are not automatically errors: holidays and closing dates must be considered.
The unit of monitoring matters. Missing records from one store and duplicates from another can cancel each other out in a company-wide count. Define checks at the company, location or business-date level where an owner can assess the issue.
Four Checks to Define First
| Area | What to check | Condition to define |
|---|---|---|
| Missing values | Whether required fields are present | Treatment of empty strings, NULLs and placeholder codes |
| Duplicates | Whether a business record has been loaded twice | Whether the key includes an order line identifier |
| Freshness | Whether required business dates are available by the deadline | Business date, load time, holidays and corrections |
| Validation | Whether counts and amounts reconcile with the source | Definitions for cancellations, returns, tax, rounding and scope |
A rule that flags repeated order identifiers may work for one row per order but not one row per product line. Define what a row represents and check the relevant combination of columns for composite keys. Distinguish missing amounts from actual zero values before replacing NULLs with zero.
What Snowflake DMFs Measure and What the Business Must Define
Snowflake’s documentation describes a DMF (Data Metric Function) as a building block for measuring data properties. Measurement and acceptance criteria are separate: a missing-value count does not by itself determine when business use should stop.
Snowflake: Introduction to data quality checks
Examples include NULL_COUNT for NULL values, DUPLICATE_COUNT for duplicates and FRESHNESS for freshness. A duplicate check on one column does not establish composite-key correctness, and a recent load timestamp does not prove that every location has delivered its records. Design business-specific reconciliation separately.
Snowflake: System data metric functions
The documentation states that Data Quality Monitoring requires Enterprise Edition and scheduled DMFs incur serverless compute costs. Check your edition, required privileges, target tables and evaluation intervals. Start with important aggregates to assess detection value and operating costs before expanding coverage.
A Hypothetical Sales Report: From Detection to Response
This is a hypothetical design example, not a reported outcome. Suppose every location’s sales for the previous business day are needed at 9 a.m. A check at 8:30 a.m. could compare expected and received locations and notify an owner of missing data. These times are illustrative and must be adapted to the actual load and reporting schedule.
A recent update timestamp is insufficient. Compare expected locations with those actually received, and align the business-day definition, time zone, corrections and returns. Consider withholding a report from normal distribution if required fields are missing or amounts still fail source reconciliation, even when the load has completed.
Include the affected dataset, business date, failed condition, investigation location and response deadline in alerts. Link to an access-controlled investigation view instead of copying personal data into chat. The business owner should decide in advance whether to pause distribution or allow clearly marked provisional values.
Checklist Before Commissioning Implementation
- Identify the report, users, decisions it supports and when it is needed.
- Prepare correct examples and cases with missing data, duplicates, delays and reconciliation gaps. Use anonymized test data.
- Record each measurement method, acceptance condition, exception, notification destination and response owner.
- Test unnecessary alerts and missed issues, including holidays and days with no expected arrivals.
- Recheck after reloads or corrections and verify report redistribution. Assign ownership for monitoring costs and rule changes.
Acceptance testing should not stop at delivery of an alert. Use test data with intentional omissions and repeated loads, then exercise detection, owner decisions, correction and verification. Document how monitoring rules will change when reporting definitions change.
The following guide covers monitoring the update process itself. Combine job monitoring with checks on the resulting data.
Snowflake task monitoring and delayed data updates
Frequently Asked Questions
Must Every Column Have Zero Missing Values?
Separate required and optional fields. Treating every missing optional value as an error can overwhelm the response process. Start with fields that affect business use and distinguish missing, pending and not-applicable values.
Does a Matching Row Count Prove the Data Is Correct?
No. Substituted records, offsetting duplicates and omissions, or incorrect amount and date conversions can remain. Check keys, reporting periods, important amounts and correspondence to source records together.
Should Every Check Be Automated from the Start?
One approach is to pilot a few important reports, validate checks and response procedures, and confirm that alerts are actionable before expanding. Test both the ability to measure a rule and the ability of staff to respond.
When consulting For f, the target report, source data and examples of discrepancies help define the scope. We can help clarify the business requirements before the implementation scope is finalized.
For f AI and data analysis services
Discuss data quality and operational design
Official Sources and Verification Date
Product details checked on November 3, 2026. Product facts are based on the sources below; the checklist and operational example are editorial proposals. Availability and interfaces may change, so verify current documentation and your environment before implementation.
