Skip to main content
EmergentEMERGENT
Emergent SolutionsKnowledge Base

GA4 Data Discrepancies

Understanding and resolving data differences in GA4 reporting.

Common Discrepancy Sources

Why Numbers Don't Match

┌─────────────────────────────────────────────────────────┐
│              Data Discrepancy Factors                    │
├─────────────────────────────────────────────────────────┤
│                                                         │
│  Collection Issues         │  Processing Differences   │
│  ─────────────────────────────────────────────────────  │
│  • Ad blockers (10-30%)    │  • Sampling in reports    │
│  • JavaScript errors       │  • Data thresholds        │
│  • Consent blocking        │  • Attribution models     │
│  • Network failures        │  • Time zone settings     │
│  • Bot filtering           │  • Currency conversion    │
│                                                         │
│  Platform Differences      │  Implementation Errors    │
│  ─────────────────────────────────────────────────────  │
│  • Different measurement   │  • Duplicate events       │
│  • Lookback windows        │  • Missing events         │
│  • Conversion counting     │  • Wrong event values     │
│  • Session definitions     │  • Inconsistent user ID   │
│                                                         │
└─────────────────────────────────────────────────────────┘

GA4 vs. Backend Data

Transaction Mismatches

Typical Range: GA4 tracks 85-95% of backend transactions

| Gap Size | Likely Cause | |----------|--------------| | 5-10% | Normal (ad blockers, tracking failures) | | 10-20% | Implementation issues | | >20% | Significant tracking problem |

Diagnosis Query

-- Compare GA4 to backend transactions
WITH ga4_transactions AS (
  SELECT
    (SELECT value.string_value FROM UNNEST(event_params)
     WHERE key = 'transaction_id') as transaction_id,
    (SELECT value.double_value FROM UNNEST(event_params)
     WHERE key = 'value') as ga4_value
  FROM `project.analytics_XXXXXX.events_*`
  WHERE event_name = 'purchase'
    AND _TABLE_SUFFIX = FORMAT_DATE('%Y%m%d', DATE_SUB(CURRENT_DATE(), INTERVAL 1 DAY))
),
backend_transactions AS (
  SELECT
    order_id as transaction_id,
    total_value as backend_value
  FROM `project.backend.orders`
  WHERE DATE(created_at) = DATE_SUB(CURRENT_DATE(), INTERVAL 1 DAY)
)
SELECT
  b.transaction_id,
  b.backend_value,
  g.ga4_value,
  CASE
    WHEN g.transaction_id IS NULL THEN 'Missing in GA4'
    WHEN ABS(b.backend_value - g.ga4_value) > 1 THEN 'Value Mismatch'
    ELSE 'Match'
  END as status
FROM backend_transactions b
LEFT JOIN ga4_transactions g ON b.transaction_id = g.transaction_id
WHERE g.transaction_id IS NULL
   OR ABS(b.backend_value - g.ga4_value) > 1

Common Fixes

| Issue | Solution | |-------|----------| | Missing transactions | Implement server-side tracking | | Value mismatches | Verify currency, tax inclusion | | Duplicate transactions | Add transaction_id deduplication |

GA4 vs. Google Ads

Conversion Discrepancies

| Factor | GA4 | Google Ads | |--------|-----|------------| | Lookback window | 30-90 days | 30-90 days | | Attribution | Data-driven | Data-driven | | Conversion counting | Once per event | Per-click option | | Session definition | 30 min inactivity | New each click |

Expected Variance

  • Within 10%: Normal differences
  • 10-30%: Check import settings
  • >30%: Configuration issue

Troubleshooting Steps

  1. Verify linking:

    • GA4 Admin → Google Ads Links
    • Check conversion import settings

  2. Compare settings:

    • Attribution model match
    • Lookback window match
    • Conversion counting method

  3. Check data:

    • Filter by Google Ads source in GA4
    • Compare date ranges exactly

GA4 vs. Meta Ads

Attribution Differences

| Factor | GA4 | Meta Ads | |--------|-----|----------| | Default model | Data-driven | 7d click / 1d view | | View-through | Not default | Included | | Cross-device | Limited | Full |

Why Meta Shows More

Meta counts:

  • View-through conversions (saw ad, didn't click)
  • Cross-device conversions
  • Shorter attribution window (inflates recent value)

Reconciliation

-- GA4 analysis for Meta traffic
SELECT
  DATE(TIMESTAMP_MICROS(event_timestamp)) as date,
  COUNT(*) as ga4_conversions,
  SUM(value) as ga4_revenue
FROM (
  SELECT
    event_timestamp,
    (SELECT value.double_value FROM UNNEST(event_params)
     WHERE key = 'value') as value
  FROM `project.analytics_XXXXXX.events_*`
  WHERE event_name = 'purchase'
    AND (LOWER(traffic_source.source) LIKE '%facebook%'
         OR LOWER(traffic_source.source) LIKE '%instagram%'
         OR LOWER(traffic_source.source) LIKE '%fb%')
)
GROUP BY date
ORDER BY date DESC

Session Discrepancies

GA4 Session Definition

  • 30 minutes of inactivity = new session
  • Midnight in property timezone = new session
  • New campaign parameters = new session

Why Session Counts Differ

| Scenario | Result | |----------|--------| | User opens 10 tabs | 1 session (shared) | | User returns after 35 min | 2 sessions | | User clicks different ad | New session | | Cross-midnight activity | 2 sessions |

Debugging Sessions

// Check session in console
// Look for ga_session_id in requests
// Should be consistent within 30 min

// GA4 stores in cookie: _ga_MEASUREMENTID
// Format: GS1.1.SESSION_ID.SESSION_COUNT...

User Count Differences

New vs. Returning

GA4 defines:

  • New user: First-ever event from device/browser
  • Returning user: Has prior visit (cookie present)

Why Counts Seem Wrong

| Issue | Cause | |-------|-------| | Too many new users | Cookie deleted, incognito, ITP | | User count > session count | Cross-device users | | Sudden increase in new | Tracking code reinstalled |

Sampling Issues

When Sampling Occurs

GA4 samples when:

  • Date range > 7 days
  • Complex explorations
  • High cardinality dimensions
  • Limited processing capacity

Detecting Sampling

Look for:

  • Yellow shield icon in reports
  • "Based on X% of data" message
  • Inconsistent numbers between date ranges

Reducing Sampling

  1. Shorten date range
  2. Use standard reports (less sampling)
  3. Export to BigQuery (no sampling)
  4. Remove unnecessary dimensions

Data Thresholds

What Gets Hidden

GA4 hides data when:

  • Row has fewer than 10 users (privacy)
  • Google signals enabled (demographic data)
  • Small segments

Signs of Thresholds

  • Missing rows in tables
  • (other) row with large values
  • Totals don't match sum of rows

Solutions

  1. Disable Google signals (if not needed)
  2. Increase date range for more data
  3. Use BigQuery export (no thresholds)
  4. Reduce dimension granularity

Reporting Delays

Processing Times

| Data Type | Typical Delay | |-----------|---------------| | Real-time | Minutes | | Standard reports | 24-48 hours | | BigQuery daily | ~24 hours | | BigQuery streaming | Minutes |

Why Data Changes

  • Late-arriving events
  • Attribution recalculation
  • Spam/bot filtering
  • Data-driven model updates

Best Practice

Wait 48-72 hours before final reporting on any date.

Debugging Checklist

Quick Diagnosis

## Data Discrepancy Checklist

### Collection
- [ ] GA4 tag firing on all pages?
- [ ] Purchase event has transaction_id?
- [ ] Values include/exclude tax consistently?
- [ ] Currency code correct?

### Processing
- [ ] Same date range compared?
- [ ] Same timezone settings?
- [ ] Sampling indicator present?
- [ ] Data thresholds applied?

### Platform Comparison
- [ ] Same attribution model?
- [ ] Same lookback window?
- [ ] Same conversion definition?
- [ ] Different deduplication rules?

Validation Query

-- Data quality check
SELECT
  event_date,
  COUNT(*) as events,
  COUNT(DISTINCT user_pseudo_id) as users,
  COUNTIF(event_name = 'purchase') as purchases,
  COUNTIF(event_name = 'purchase'
    AND (SELECT value.string_value FROM UNNEST(event_params)
         WHERE key = 'transaction_id') IS NULL) as purchases_no_txn_id,
  COUNTIF(event_name = 'purchase'
    AND (SELECT value.double_value FROM UNNEST(event_params)
         WHERE key = 'value') IS NULL) as purchases_no_value
FROM `project.analytics_XXXXXX.events_*`
WHERE _TABLE_SUFFIX >= FORMAT_DATE('%Y%m%d', DATE_SUB(CURRENT_DATE(), INTERVAL 7 DAY))
GROUP BY event_date
ORDER BY event_date DESC

Previous: GTM Troubleshooting Next: Cross-Domain Tracking