Google Analytics with BigQuery in 2026: Data Analysis Capabilities

Google Analytics + BigQuery data analysis in 2026 — what the export unlocks, schema basics, ten worked questions, dialect quirks, and where AI data agents fit.

First-person desk note: When I first wired a live GA4 BigQuery export, the schema looked simple until every useful metric hid inside event_params. The ten SQL patterns below are the ones we keep rewriting in reviews — not theory.

Direct answer: what the GA4 BigQuery export unlocks

Desk quantitative notes (composite)

These figures come from anonymized desk replays on sample GA4 BigQuery export datasets — useful for planning, not InfiniSynapse product SLAs or customer win rates.

The GA4 BigQuery event table schema

Each GA4 property linked to BigQuery creates a GA4 BigQuery export dataset named analytics_PROPERTY_ID. Inside it, daily tables named events_YYYYMMDD hold one row per event. Streaming export (intraday) lands in events_intraday_YYYYMMDD. Field types match the official GA4 BigQuery export schema documentation.

The UNNEST pattern every GA4 BigQuery export query uses

Nearly every useful query on a GA4 BigQuery export uses this extraction shape. Official event names and parameters are listed in the GA4 events reference.

-- Extract a parameter value from event_params
SELECT
  event_date,
  event_name,
  (SELECT value.string_value FROM UNNEST(event_params)
   WHERE key = 'page_location') AS page_location,
  (SELECT value.int_value FROM UNNEST(event_params)
   WHERE key = 'engagement_time_msec') AS engagement_ms
FROM `project.analytics_PROPERTY_ID.events_*`
WHERE _TABLE_SUFFIX BETWEEN '20260601' AND '20260628'
  AND event_name = 'page_view';

Three things to remember: _TABLE_SUFFIX is how you partition the date range, (SELECT ... FROM UNNEST(...)) is the canonical extraction shape, and choosing the right value.* field type (string_value, int_value, double_value, float_value) matters for each parameter.

Ten worked questions on GA4 BigQuery export

Copy-paste starters for the weekly questions we see most often on a GA4 BigQuery export. Replace project.analytics_PROPERTY_ID and date bounds before running. Dialect notes follow the BigQuery SQL reference.

1. DAU / WAU / MAU

SELECT
  PARSE_DATE('%Y%m%d', event_date) AS day,
  COUNT(DISTINCT user_pseudo_id) AS dau
FROM `project.analytics_PROPERTY_ID.events_*`
WHERE _TABLE_SUFFIX BETWEEN '20260601' AND '20260628'
GROUP BY day
ORDER BY day;

2. Retention curve by signup cohort

WITH first_seen AS (
  SELECT user_pseudo_id, MIN(PARSE_DATE('%Y%m%d', event_date)) AS cohort_day
  FROM `project.analytics_PROPERTY_ID.events_*`
  WHERE _TABLE_SUFFIX BETWEEN '20260401' AND '20260628'
  GROUP BY 1
),
activity AS (
  SELECT DISTINCT user_pseudo_id, PARSE_DATE('%Y%m%d', event_date) AS act_day
  FROM `project.analytics_PROPERTY_ID.events_*`
  WHERE _TABLE_SUFFIX BETWEEN '20260401' AND '20260628'
)
SELECT
  f.cohort_day,
  DATE_DIFF(a.act_day, f.cohort_day, WEEK) AS week_n,
  COUNT(DISTINCT a.user_pseudo_id) AS users
FROM first_seen f
JOIN activity a USING (user_pseudo_id)
GROUP BY 1, 2
ORDER BY 1, 2;

3. Funnel analysis (view → add_to_cart → purchase)

WITH steps AS (
  SELECT
    user_pseudo_id,
    event_timestamp,
    event_name,
    ROW_NUMBER() OVER (
      PARTITION BY user_pseudo_id ORDER BY event_timestamp
    ) AS step_seq
  FROM `project.analytics_PROPERTY_ID.events_*`
  WHERE _TABLE_SUFFIX BETWEEN '20260601' AND '20260628'
    AND event_name IN ('page_view', 'add_to_cart', 'purchase')
),
ordered AS (
  SELECT
    user_pseudo_id,
    MAX(IF(event_name = 'page_view', event_timestamp, NULL)) AS t_view,
    MAX(IF(event_name = 'add_to_cart', event_timestamp, NULL)) AS t_cart,
    MAX(IF(event_name = 'purchase', event_timestamp, NULL)) AS t_purchase
  FROM steps
  GROUP BY 1
)
SELECT
  COUNTIF(t_view IS NOT NULL) AS step1_users,
  COUNTIF(t_cart IS NOT NULL AND t_cart >= t_view) AS step2_users,
  COUNTIF(t_purchase IS NOT NULL AND t_purchase >= t_cart) AS step3_users
FROM ordered;

This is the HowTo shape for funnel work on a GA4 BigQuery export: project → sequence → step join → optionally materialize.

4. Attribution audit (first touch vs converting session)

WITH sessions AS (
  SELECT
    user_pseudo_id,
    event_timestamp,
    traffic_source.source AS source,
    traffic_source.medium AS medium,
    event_name
  FROM `project.analytics_PROPERTY_ID.events_*`
  WHERE _TABLE_SUFFIX BETWEEN '20260601' AND '20260628'
),
first_touch AS (
  SELECT AS VALUE ARRAY_AGG(s ORDER BY event_timestamp LIMIT 1)[OFFSET(0)]
  FROM sessions s
  GROUP BY user_pseudo_id
),
converters AS (
  SELECT AS VALUE ARRAY_AGG(s ORDER BY event_timestamp DESC LIMIT 1)[OFFSET(0)]
  FROM sessions s
  WHERE event_name = 'purchase'
  GROUP BY user_pseudo_id
)
SELECT
  f.source AS first_source,
  c.source AS convert_source,
  COUNT(*) AS users
FROM first_touch f
JOIN converters c USING (user_pseudo_id)
GROUP BY 1, 2
ORDER BY users DESC;

5. Channel performance

SELECT
  traffic_source.medium AS medium,
  COUNTIF(event_name = 'purchase') AS purchases,
  SUM(IF(event_name = 'purchase', ecommerce.purchase_revenue, 0)) AS revenue
FROM `project.analytics_PROPERTY_ID.events_*`
WHERE _TABLE_SUFFIX BETWEEN '20260601' AND '20260628'
GROUP BY medium
ORDER BY revenue DESC;

6. Custom event rates

SELECT
  COUNTIF(event_name = 'generate_lead') AS custom_events,
  COUNTIF(event_name = 'page_view') AS page_views,
  SAFE_DIVIDE(
    COUNTIF(event_name = 'generate_lead'),
    COUNTIF(event_name = 'page_view')
  ) AS custom_per_pageview
FROM `project.analytics_PROPERTY_ID.events_*`
WHERE _TABLE_SUFFIX BETWEEN '20260601' AND '20260628';

7. Page-level engagement

SELECT
  (SELECT value.string_value FROM UNNEST(event_params) WHERE key = 'page_location') AS page,
  AVG((SELECT value.int_value FROM UNNEST(event_params) WHERE key = 'engagement_time_msec')) AS avg_engagement_ms
FROM `project.analytics_PROPERTY_ID.events_*`
WHERE _TABLE_SUFFIX BETWEEN '20260601' AND '20260628'
  AND event_name = 'page_view'
GROUP BY page
ORDER BY avg_engagement_ms DESC
LIMIT 50;

8. Scroll depth analysis

SELECT
  (SELECT value.string_value FROM UNNEST(event_params) WHERE key = 'page_location') AS page,
  (SELECT value.int_value FROM UNNEST(event_params) WHERE key = 'percent_scrolled') AS pct,
  COUNT(*) AS events
FROM `project.analytics_PROPERTY_ID.events_*`
WHERE _TABLE_SUFFIX BETWEEN '20260601' AND '20260628'
  AND event_name = 'scroll'
GROUP BY page, pct
ORDER BY page, pct;

9. Search query analysis

SELECT
  (SELECT value.string_value FROM UNNEST(event_params) WHERE key = 'search_term') AS query,
  COUNT(*) AS searches
FROM `project.analytics_PROPERTY_ID.events_*`
WHERE _TABLE_SUFFIX BETWEEN '20260601' AND '20260628'
  AND event_name = 'view_search_results'
GROUP BY query
ORDER BY searches DESC
LIMIT 100;

10. Revenue by source

SELECT
  traffic_source.source AS source,
  SUM(ecommerce.purchase_revenue) AS revenue,
  COUNT(*) AS purchase_events
FROM `project.analytics_PROPERTY_ID.events_*`
WHERE _TABLE_SUFFIX BETWEEN '20260601' AND '20260628'
  AND event_name = 'purchase'
GROUP BY source
ORDER BY revenue DESC;

These ten cover the majority of weekly analyst work on a GA4 BigQuery export dataset. The companion marketing data analysis playbook explains how these feed business-level KPIs.

BigQuery cost notes for GA4 BigQuery export work

Cost accidents on a GA4 BigQuery export almost always come from scanning every daily shard. Treat partition filters as mandatory.

• Partition scanning. Always use _TABLE_SUFFIX BETWEEN in your WHERE; without it, BigQuery scans every daily table from day 1.
• SELECT * costs. The GA4 events table is wide and nested. Project only the columns you need.
• Materialized aggregates. If you re-run the same daily aggregation many times, write the result to a separate table and query that.
• BI tool defaults. Tableau and Looker can issue large scans if connectors are not configured for incremental refresh.

The BigQuery best practices documentation spells out the cost model.

Where AI data agents fit on top of GA4 BigQuery export

Three patterns where an AI data analyst earns the seat on a GA4 BigQuery export dataset:

• UNNEST boilerplate. Every GA4 query needs the UNNEST pattern. An agent drafts the SQL with the correct extraction shape from a plain-English question.
• Cross-property analysis. A team with three GA4 properties — main site, marketing site, app — runs the same question across all three datasets. The agent handles the UNION and renaming.
• Ad-hoc anomaly investigation. "Why did engagement drop on iOS Wednesday?" — the agent UNNESTs the right params, segments by device, and returns a chart with the SQL.

See the AI database query pillar guide for the connection pattern and database + knowledge base binding for how to seed GA4 event definitions as bound context.

FAQ

Sources and references

1. [Vendor] Google. GA4 BigQuery export schema documentation. support.google.com/analytics/answer/7029846.
2. [Vendor] Google. BigQuery SQL reference. cloud.google.com/bigquery/docs/reference/standard-sql.
3. [Vendor] Google. BigQuery cost best practices. cloud.google.com/bigquery/best-practices.
4. [Vendor] Google. GA4 events reference. developers.google.com/analytics/ga4/events.
5. [Independent] Yao et al. ReAct paper. arxiv.org/abs/2210.03629.
6. [Vendor] Anthropic. Building Effective Agents. anthropic.com/research/building-effective-agents.
7. [Standard] NIST. AI Risk Management Framework. nist.gov/itl/ai-risk-management-framework.
8. [Independent] BIRD-SQL benchmark. bird-bench.github.io.
9. [Policy / About] InfiniSynapse — Editorial standards & author credentials. infinisynapse.com/en/editorial-standards. Company: About InfiniSynapse.

Related guides