BigQuery

​

Overview

Connection Type:

​

Installation Instructions

​

Multi-touch attribution support

Converge preprocesses all events and enriches them with all available profile and session information to make your attribution data warehouse queries as simple as possible.

​

Profile Information

Every event that is forwarded includes all available information Converge has on that profile until that moment, ranging from $ip_address all the way to $email.

Events that typically do not have an $email property associated with them (e.g. $page_load) will still contain this property if we were able to identify this profile before the $page_load happened.

​

Profile Merging

Converge may combine multiple profiles into one if there is an overlap in aliases. This means that the profile_id field in your events table may become out of date. Under the ‘Advanced’ tab of the BigQuery connection configuration, you can configure a table to stream these profile merges to. You can create this table as follows:

• Under your Converge dataset (converge_data), create a new table, e.g. profile_merges.
• In the Schema section, flip the switch to Edit as text and paste the schema below. Click on Create Table after.
  Schema

  JSON

  [
    {
      "name": "timestamp",
      "type": "DATETIME",
      "mode": "REQUIRED",
      "description": "Timestamp when the event occurred"
    },
    {
     "name": "old_profile_id",
     "type": "INTEGER",
     "mode": "REQUIRED",
     "description": "Old Profile ID that was merged"
    },
    {
      "name": "new_profile_id",
      "type": "INTEGER",
      "mode": "REQUIRED",
      "description": "New Profile ID resulting from the merge"
    }
  ]
  

To use this table, it’s important to note that profiles maybe merged multiple times. For example profile A and B could be mapped to C as follows:

• (A, C)
• (B, C) Later on, profile C could be merged into profile D:
• (C, D) In this case, profiles A, B, and C, would all be mapped to D. To deal with this, we recommend creating a (materialized) view that recursively resolves these mappings. Here’s an example query to achieve this:

WITH RECURSIVE profile_merges AS (
  SELECT 
    timestamp, 
    old_profile_id, 
    new_profile_id, 
    ARRAY[old_profile_id] AS visited_profiles,
    1 AS depth,
    FALSE AS is_circular_reference
  FROM 
    `your-project-name.your-dataset-name.profile_merges`
  UNION ALL
  SELECT 
    pmr.timestamp, 
    pm.old_profile_id,
    pmr.new_profile_id, 
    ARRAY_CONCAT(pm.visited_profiles, [pmr.new_profile_id]) AS visited_profiles,
    pm.depth + 1 AS depth,
    IF(pmr.new_profile_id IN UNNEST(pm.visited_profiles), TRUE, pm.is_circular_reference) AS is_circular_reference
  FROM 
    profile_merges pm
  JOIN 
    `your-project-name.your-dataset-name.profile_merges` pmr
  ON 
    pm.new_profile_id = pmr.old_profile_id
  WHERE
    pmr.timestamp > pm.timestamp
    AND NOT pm.is_circular_reference
)
SELECT 
  old_profile_id, 
  MAX_BY(new_profile_id, depth) AS new_profile_id
FROM 
  profile_merges
GROUP BY 
  old_profile_id;

​

Querying attribution data

The events table contains all the data you need to build custom attribution models. You can easily construct the customer’s touchpoints through the session_id and session_properties columns.

A multi-touch model can be broken down into the following steps:

1. Selecting conversion events
2. Selecting touchpoints
3. Joining touchpoints with conversion events according to the attribution window
4. Applying the attribution model
5. Aggregating conversion metrics by touchpoint

This model can be adapted to 100% flexibly build attribution insights. The examples below show how to build basic first-touch and last-touch models and count the number of orders in BigQuery.

  WITH _events AS (SELECT * FROM `your-project-name.your-dataset-name.your-table-name`),
  _1_conversions AS (
    SELECT
        id,
        profile_id,
        timestamp
    FROM _events
    WHERE
        timestamp > CURRENT_DATETIME() - INTERVAL 7 day
        AND event_name = 'Placed Order'
  ), 
  _2_touchpoints AS (
    SELECT
        profile_id,
        session_id,
        MIN_BY(session_properties, timestamp) AS session_properties,
        MIN(timestamp) AS timestamp
    FROM _events
    WHERE
        timestamp > CURRENT_DATETIME() - INTERVAL 7 day - INTERVAL 7 day
        AND event_name = '$page_load'
    GROUP BY profile_id, session_id
  ),
  _3_conversions_touchpoints AS (
    SELECT
      c.id AS id,
      t.timestamp AS touchpoint_timestamp,
      t.session_properties AS touchpoint_properties
    FROM _1_conversions AS c
    LEFT JOIN _2_touchpoints AS t
    ON c.profile_id = t.profile_id
        AND t.timestamp BETWEEN c.timestamp - INTERVAL 28 day AND c.timestamp
  ),
  _4_attributed_conversions AS (
    SELECT
        id,
        MIN_BY(touchpoint_properties, touchpoint_timestamp) AS touchpoint_properties,
        MIN(touchpoint_timestamp) AS touchpoint_timestamp
    FROM _3_conversions_touchpoints
    GROUP BY id
  )
  SELECT
      JSON_VALUE(touchpoint_properties, '$.utm_source') AS first_touch_utm_source,
      JSON_VALUE(touchpoint_properties, '$.utm_medium') AS first_touch_utm_medium,
      COUNT(*) AS conversions
  FROM _4_attributed_conversions
  GROUP BY first_touch_utm_source, first_touch_utm_medium

  WITH _events AS (SELECT * FROM `your-project-name.your-dataset-name.your-table-name`),
  _1_conversions AS (
    SELECT
        id,
        profile_id,
        timestamp
    FROM _events
    WHERE
        timestamp > CURRENT_DATETIME() - INTERVAL 7 day
        AND event_name = 'Placed Order'
  ), 
  _2_touchpoints AS (
    SELECT
        profile_id,
        session_id,
        MIN_BY(session_properties, timestamp) AS session_properties,
        MIN(timestamp) AS timestamp
    FROM _events
    WHERE
        timestamp > CURRENT_DATETIME() - INTERVAL 7 day - INTERVAL 7 day
        AND event_name = '$page_load'
    GROUP BY profile_id, session_id
  ),
  _3_conversions_touchpoints AS (
    SELECT
      c.id AS id,
      t.timestamp AS touchpoint_timestamp,
      t.session_properties AS touchpoint_properties
    FROM _1_conversions AS c
    LEFT JOIN _2_touchpoints AS t
    ON c.profile_id = t.profile_id
        AND t.timestamp BETWEEN c.timestamp - INTERVAL 28 day AND c.timestamp
    ),
  _4_attributed_conversions AS (
    SELECT
        id,
        MAX_BY(touchpoint_properties, touchpoint_timestamp) AS touchpoint_properties,
        MAX(touchpoint_timestamp) AS touchpoint_timestamp
    FROM _3_conversions_touchpoints
    GROUP BY id
  )
  SELECT
      JSON_VALUE(touchpoint_properties, '$.utm_source') AS last_touch_utm_source,
      JSON_VALUE(touchpoint_properties, '$.utm_medium') AS last_touch_utm_medium,
      COUNT(*) AS conversions
  FROM _4_attributed_conversions
  GROUP BY last_touch_utm_source, last_touch_utm_medium

The session columns have the following structure:

​

Deprecation of pre-processed attribution columns

Converge previously supported the following columns:

• first_touch_properties and first_touch_session_id
• last_touch_properties and last_touch_session_id
• first_touch_paid_properties and first_touch_paid_session_id
• last_touch_paid_properties and last_touch_paid_session_id

These columns are deprecated and are no longer being filled. The recommended way to build multi-touch attribution models is to use the session_id and session_properties columns in the Converge data. These allow full flexibility in building custom attribution models while maintaining correctness.

​

An example query

As an example, the following query on top of the Converge Table gives you the number of times an SKU was ordered:

SELECT
  string(item.sku) sku, count(*) order_count
FROM `your-project-name.your-dataset-name.your-table-name`
LEFT JOIN unnest(json_extract_array(event_properties.items,'$')) item
WHERE event_name = 'Placed Order'
GROUP BY sku
ORDER BY order_count DESC

​

Converge functionality

This integration supports the following Converge connection functionality.

​

Event Mapping

Converge automatically sends all implemented events from the Converge Spec and any custom events to BigQuery.