GA4 Data Quality Validation

Data quality is foundational to any analytics program, yet many organizations ship measurement implementations without systematic validation. A missing parameter here, a traffic spike there, and six months later your acquisition reports are unreliable, your attribution model trains on corrupted data, and your marketing team makes decisions on faulty signals.

GA4 + BigQuery enable automated auditing. Rather than manually spot-checking reports, define 15+ automated checks that run daily or weekly, alert when thresholds are breached, and give you early warning before data problems compound.

Why systematic validation matters

Without data quality checks, you discover problems reactively:

• A bot surge inflates pageviews on March 15. You notice it March 22 when reporting on March’s baseline.
• An iOS app stops sending purchase events. By the time you notice, two weeks of conversion data is incomplete.
• A third-party integration starts double-counting orders. The corruption spreads through attribution models, audience cohorts, and forecasting.

Automated checks catch these issues within hours of occurrence, when remediation is still possible.

15+ Essential Data Quality Checks

1. Event count anomalies (sudden drops)

Detect when event volume drops below a moving baseline. A sudden 40%+ drop often signals a measurement breakage.

-- Check: Daily event count vs. 7-day rolling average
WITH daily_events AS (
  SELECT
    CURRENT_DATE() - 1 AS check_date,
    COUNT(*) AS event_count
  FROM `project.dataset.events_*`
  WHERE _TABLE_SUFFIX = FORMAT_DATE('%Y%m%d', CURRENT_DATE() - 1)
),
baseline AS (
  SELECT
    AVG(event_count) AS avg_7day,
    STDDEV_POP(event_count) * 2 AS alert_threshold
  FROM (
    SELECT
      COUNT(*) AS event_count
    FROM `project.dataset.events_*`
    WHERE _TABLE_SUFFIX BETWEEN FORMAT_DATE('%Y%m%d', CURRENT_DATE() - 8)
                            AND FORMAT_DATE('%Y%m%d', CURRENT_DATE() - 2)
    GROUP BY _TABLE_SUFFIX
  )
)
SELECT
  d.check_date,
  d.event_count,
  ROUND(b.avg_7day) AS expected_count,
  ROUND((d.event_count / b.avg_7day - 1) * 100, 1) AS pct_change,
  'DROP_ALERT' AS alert_type
FROM daily_events d, baseline b
WHERE d.event_count < (b.avg_7day - b.alert_threshold);

Action: Volume drops > 30% warrant immediate investigation. Check measurement tags, client-side errors, and server-side filtering rules.

2. Event count anomalies (sudden spikes)

Traffic spikes may be legitimate (viral post, PR coverage) or problematic (bot surge, tag duplication). Raise alerts so you can inspect the spike in real time.

-- Check: Daily event count spike detection
WITH daily_events AS (
  SELECT
    CURRENT_DATE() - 1 AS check_date,
    COUNT(*) AS event_count
  FROM `project.dataset.events_*`
  WHERE _TABLE_SUFFIX = FORMAT_DATE('%Y%m%d', CURRENT_DATE() - 1)
),
baseline AS (
  SELECT
    AVG(event_count) AS avg_7day,
    STDDEV_POP(event_count) * 2 AS alert_threshold
  FROM (
    SELECT
      COUNT(*) AS event_count
    FROM `project.dataset.events_*`
    WHERE _TABLE_SUFFIX BETWEEN FORMAT_DATE('%Y%m%d', CURRENT_DATE() - 8)
                            AND FORMAT_DATE('%Y%m%d', CURRENT_DATE() - 2)
    GROUP BY _TABLE_SUFFIX
  )
)
SELECT
  d.check_date,
  d.event_count,
  ROUND(b.avg_7day) AS expected_count,
  ROUND((d.event_count / b.avg_7day - 1) * 100, 1) AS pct_change,
  'SPIKE_ALERT' AS alert_type
FROM daily_events d, baseline b
WHERE d.event_count > (b.avg_7day + b.alert_threshold);

Action: Spikes > 50% should trigger immediate review. Cross-reference with marketing activity calendar. If unexplained, check for tag duplication or bot traffic.

3. Missing required parameters

Events should have specific parameters depending on event type. For example, purchase events must include transaction_id and value.

-- Check: Purchase events without required parameters
SELECT
  event_date,
  COUNT(*) AS missing_count,
  ROUND(COUNT(*) / (
    SELECT COUNT(*)
    FROM `project.dataset.events_*`
    WHERE _TABLE_SUFFIX = FORMAT_DATE('%Y%m%d', event_date)
      AND event_name = 'purchase'
  ) * 100, 2) AS pct_of_purchases
FROM `project.dataset.events_*`
WHERE _TABLE_SUFFIX = FORMAT_DATE('%Y%m%d', CURRENT_DATE() - 1)
  AND event_name = 'purchase'
  AND (
    NOT EXISTS (
      SELECT 1 FROM UNNEST(event_params) AS param
      WHERE param.key = 'transaction_id'
    )
    OR NOT EXISTS (
      SELECT 1 FROM UNNEST(event_params) AS param
      WHERE param.key = 'value'
    )
  )
GROUP BY event_date
HAVING missing_count > 0;

Threshold: Flag if > 5% of purchase events lack transaction_id or value. This corrupts revenue reporting.

4. Duplicate transactions

Duplicate transactions inflate revenue. Check for multiple purchase events with identical transaction_id within short timeframes.

-- Check: Duplicate transaction IDs
WITH transaction_ids AS (
  SELECT
    event_date,
    (SELECT value.string_value FROM UNNEST(event_params)
     WHERE key = 'transaction_id') AS transaction_id,
    COUNT(*) AS occurrence_count
  FROM `project.dataset.events_*`
  WHERE _TABLE_SUFFIX = FORMAT_DATE('%Y%m%d', CURRENT_DATE() - 1)
    AND event_name = 'purchase'
  GROUP BY event_date, transaction_id
)
SELECT
  event_date,
  COUNT(*) AS duplicate_txn_ids,
  SUM(occurrence_count - 1) AS excess_events
FROM transaction_ids
WHERE occurrence_count > 1
GROUP BY event_date;

Threshold: Any duplicate transaction ID is a warning. Investigate tag firing rules or backend retry logic.

5. Internal traffic leaking into production

Exclude internal/office traffic from analytics to avoid skewing conversion rates and polluting audiences.

-- Check: Internal IP traffic reaching GA4
WITH params AS (
  SELECT
    event_date,
    user_id,
    (SELECT value.string_value FROM UNNEST(event_params)
     WHERE key = 'ip_address') AS ip_address,
    COUNT(*) AS event_count
  FROM `project.dataset.events_*`
  WHERE _TABLE_SUFFIX = FORMAT_DATE('%Y%m%d', CURRENT_DATE() - 1)
  GROUP BY event_date, user_id, ip_address
)
SELECT
  event_date,
  COUNT(DISTINCT user_id) AS internal_users,
  SUM(event_count) AS internal_events,
  ARRAY_AGG(DISTINCT ip_address LIMIT 10) AS sample_ips
FROM params
WHERE ip_address LIKE '203.0.113.%'  -- Replace with your office IP range
   OR ip_address LIKE '10.0.%'
GROUP BY event_date;

Action: If internal traffic appears, verify that GA4 filters are active under Admin → Data Filters.

6. Enhanced measurement double-counting

GA4’s enhanced measurement auto-tracks page views, scrolls, and form submissions. If you also manually fire these events, you’ll double-count.

-- Check: Potential enhanced measurement + manual event overlap
SELECT
  event_date,
  event_name,
  COUNT(*) AS event_count,
  COUNT(DISTINCT user_id) AS unique_users,
  ROUND(COUNT(*) / COUNT(DISTINCT user_id), 2) AS avg_events_per_user
FROM `project.dataset.events_*`
WHERE _TABLE_SUFFIX = FORMAT_DATE('%Y%m%d', CURRENT_DATE() - 1)
  AND event_name IN ('page_view', 'scroll', 'form_submit')
GROUP BY event_date, event_name
ORDER BY avg_events_per_user DESC;

Threshold: If average events per user > 1.5 for page_view on a single-page app, investigate. On multi-page sites, ratios > 1.2 warrant review.

7. Referral spam detection

Referral spam in the session_source parameter pollutes attribution. Common patterns: (direct) (none) with referrer, or obvious bot domains.

-- Check: Suspicious referrer spam patterns
SELECT
  event_date,
  (SELECT value.string_value FROM UNNEST(event_params)
   WHERE key = 'session_source') AS source,
  (SELECT value.string_value FROM UNNEST(event_params)
   WHERE key = 'session_medium') AS medium,
  COUNT(DISTINCT user_id) AS user_count,
  COUNT(*) AS event_count
FROM `project.dataset.events_*`
WHERE _TABLE_SUFFIX = FORMAT_DATE('%Y%m%d', CURRENT_DATE() - 1)
  AND (
    (SELECT value.string_value FROM UNNEST(event_params)
     WHERE key = 'session_source') IN (
      'bestwebsiteslinks.com', 'edakma.ru', 'sady.ua', 'buttons-for-website.com'
    )
    OR (
      (SELECT value.string_value FROM UNNEST(event_params)
       WHERE key = 'session_source') LIKE '%bot%'
      OR (SELECT value.string_value FROM UNNEST(event_params)
          WHERE key = 'session_source') LIKE '%spam%'
    )
  )
GROUP BY event_date, source, medium
ORDER BY event_count DESC;

Action: Add detected spam sources to GA4’s referral exclusion list under Admin → Data Streams → Enhanced Measurement.

8. Session count vs. user count ratio anomalies

The session-to-user ratio typically ranges from 1.2 to 1.5. Ratios > 2.0 suggest session-splitting bugs or bots with session management.

-- Check: Session to user ratio anomalies
SELECT
  event_date,
  COUNT(DISTINCT user_id) AS users,
  COUNT(DISTINCT concat(user_id, '_', (
    SELECT value.int_value FROM UNNEST(event_params)
    WHERE key = 'ga_session_id'
  ))) AS sessions,
  ROUND(
    COUNT(DISTINCT concat(user_id, '_', (
      SELECT value.int_value FROM UNNEST(event_params)
      WHERE key = 'ga_session_id'
    ))) / COUNT(DISTINCT user_id), 2
  ) AS session_to_user_ratio
FROM `project.dataset.events_*`
WHERE _TABLE_SUFFIX = FORMAT_DATE('%Y%m%d', CURRENT_DATE() - 1)
GROUP BY event_date
HAVING session_to_user_ratio > 2.0;

Threshold: Ratio > 2.0 warrants investigation. Check for session ID initialization issues or bot traffic with randomized sessions.

9. Conversion rate anomalies

Sudden dips in conversion rate (purchases, key events) signal breakage.

-- Check: Conversion rate vs. baseline
WITH daily_metrics AS (
  SELECT
    event_date,
    COUNT(DISTINCT user_id) AS users,
    SUM(CASE
      WHEN event_name IN ('purchase', 'signup', 'key_event') THEN 1
      ELSE 0
    END) AS conversions
  FROM `project.dataset.events_*`
  WHERE _TABLE_SUFFIX = FORMAT_DATE('%Y%m%d', CURRENT_DATE() - 1)
  GROUP BY event_date
),
baseline AS (
  SELECT
    AVG(conversions / users) AS avg_rate
  FROM (
    SELECT
      event_date,
      SUM(CASE
        WHEN event_name IN ('purchase', 'signup', 'key_event') THEN 1
        ELSE 0
      END) / COUNT(DISTINCT user_id) AS conversions,
      COUNT(DISTINCT user_id) AS users
    FROM `project.dataset.events_*`
    WHERE _TABLE_SUFFIX BETWEEN FORMAT_DATE('%Y%m%d', CURRENT_DATE() - 8)
                            AND FORMAT_DATE('%Y%m%d', CURRENT_DATE() - 2)
    GROUP BY _TABLE_SUFFIX
  )
)
SELECT
  d.event_date,
  ROUND(d.conversions / d.users * 100, 2) AS cr_pct,
  ROUND(b.avg_rate * 100, 2) AS baseline_cr,
  ROUND((d.conversions / d.users / b.avg_rate - 1) * 100, 1) AS pct_change
FROM daily_metrics d, baseline b
WHERE (d.conversions / d.users) < (b.avg_rate * 0.7)
  OR (d.conversions / d.users) > (b.avg_rate * 1.3);

Action: CR drops > 30% suggest broken conversion tracking. Check GTM tag firing, validation rules, and backend integration.

10. Data freshness monitoring

GA4 has a reporting latency of 24–48 hours for most data. Check that yesterday’s data is complete.

-- Check: Data freshness
SELECT
  MAX(_TABLE_SUFFIX) AS latest_date,
  COUNT(*) AS event_count,
  CURRENT_DATE() - 1 AS expected_date,
  IF(
    PARSE_DATE('%Y%m%d', MAX(_TABLE_SUFFIX)) = CURRENT_DATE() - 1,
    'COMPLETE',
    'INCOMPLETE'
  ) AS freshness_status
FROM `project.dataset.events_*`
WHERE _TABLE_SUFFIX >= FORMAT_DATE('%Y%m%d', CURRENT_DATE() - 2);

Action: If freshness_status = ‘INCOMPLETE’ on day 2, check GA4 reporting API and BigQuery export configuration.

11. Cardinality threshold warnings

High-cardinality dimensions (like user IDs in custom parameters) waste storage and slow queries. GA4 warns when a single property or dimension exceeds 50,000 unique values in a day.

-- Check: Custom dimension cardinality
SELECT
  event_date,
  (SELECT key FROM UNNEST(event_params) LIMIT 1) AS dimension_name,
  COUNT(DISTINCT (SELECT value.string_value FROM UNNEST(event_params)
                  WHERE key = 'user_id')) AS cardinality
FROM `project.dataset.events_*`
WHERE _TABLE_SUFFIX = FORMAT_DATE('%Y%m%d', CURRENT_DATE() - 1)
  AND EXISTS (
    SELECT 1 FROM UNNEST(event_params) AS param
    WHERE param.key = 'user_id'
  )
GROUP BY event_date, dimension_name
HAVING cardinality > 50000;

Action: High-cardinality dimensions should be sent to BigQuery only, not GA4’s front-end, to preserve reporting capacity.

12. Custom dimension utilization audit

Ensure custom dimensions are actually in use and not wasting your dimension quota (max 25 per property).

-- Check: Which custom dimensions are actually populated
SELECT
  param.key AS dimension_name,
  COUNT(DISTINCT user_id) AS users_with_value,
  COUNT(*) AS total_occurrences,
  ROUND(COUNT(*) / (
    SELECT COUNT(*)
    FROM `project.dataset.events_*`
    WHERE _TABLE_SUFFIX = FORMAT_DATE('%Y%m%d', CURRENT_DATE() - 1)
  ) * 100, 2) AS fill_rate_pct
FROM `project.dataset.events_*`
CROSS JOIN UNNEST(event_params) AS param
WHERE _TABLE_SUFFIX = FORMAT_DATE('%Y%m%d', CURRENT_DATE() - 1)
  AND param.key LIKE 'custom_%'
GROUP BY param.key
ORDER BY total_occurrences DESC;

Action: Dimensions with < 1% fill rate are wasting quota. Disable them and repurpose the slot.

Setting up alerts

Via Cloud Monitoring (recommended for automated response)

Create a Cloud Monitoring alert that triggers when the BigQuery query returns rows:

1. In Cloud Monitoring, click Alerts → Create Policy.
2. Under Select a metric, choose BigQuery → Query Result.
3. Point to your data quality check query.
4. Set the threshold: “Alert when result count > 0”.
5. Add notification channels (email, Slack webhook, PagerDuty).
6. Save the policy with a name like “GA4 Data Quality - Event Count Spike”.

Via Scheduled Queries

If you prefer to consolidate all checks into a single report:

1. In BigQuery, create each check query as a Scheduled Query.
2. Set frequency to Daily at 2 AM UTC (gives GA4 time to export data).
3. Destination table: project.dataset.data_quality_checks
4. Append results each run.
5. Set up a Data Studio dashboard that visualizes all check results.

Note

BigQuery export to a standard table has 24–48 hour latency. If you need real-time alerts, use the GA4 Realtime API (capped at 100 concurrent requests, max 10 second window).

DumbData.co audit tool

For a faster, code-free audit, DumbData.co provides a GA4 audit checklist that includes:

• Event count consistency
• Parameter completeness
• Conversion tracking validation
• Custom dimension coverage
• Audience member counts

Use it as a monthly spot-check before reviewing your automated query results.

Monthly audit schedule

Run these checks on a regular cadence:

Frequency	Checks
Daily	Event count anomalies, spike detection, data freshness
Weekly	Missing parameters, duplicate transactions, session ratio
Monthly	Referral spam, dimension utilization, dimension cardinality
Quarterly	Review conversion rate baseline, audit dimension necessity

Maintain a data quality log documenting each check run, any alerts, and remediation actions taken. This log becomes invaluable when investigating attribution questions months later (“Was our data clean in January?”).

Common mistakes

Running checks too infrequently

Waiting until your monthly review to check data quality means problems compound. Run critical checks (event count anomalies, missing parameters, duplicates) daily. By the time you see the spike in your dashboard, a week of corrupted data is already in BigQuery.

Ignoring small data quality issues

A 2% gap in purchase events seems minor. But multiplied across 30 days and 10 properties, that’s 6 days of missed conversion data. Small issues compound. Set thresholds low and investigate early.

Not alerting stakeholders to data quality changes

When you deploy a filter to remove bot traffic, event counts drop 15% overnight. If you don’t proactively communicate this to your marketing team, they’ll lose confidence in your data. Always notify stakeholders before and after data quality deployments.

Confusing BigQuery tables with GA4 reports

BigQuery export is raw, unfiltered data. GA4 reports apply bot filtering, deduplication, and sampling. A discrepancy between your BigQuery check and GA4’s UI is normal. Always query BigQuery for data quality — never rely on GA4 reports for validation.

Related articles

BigQuery Integration Set up and troubleshoot GA4 to BigQuery export for custom analysis.

Conversions Mark events as conversions and monitor conversion data completeness.

Bot Detection & Filtering Filter bot traffic server-side to prevent corrupted conversion data.

Custom Dimensions Dimension cardinality, utilization, and quota limits.