Export your analytics metrics into BigQuery

This page describes how to export Vertex AI Search for retail analytics metrics into BigQuery. Exporting analytics metrics lets you retain the metrics and write SQL for your own analysis.

Vertex AI Search for retail analytics metrics are generated automatically for you. For more information about measuring the business impact of Vertex AI Search for retail with analytics metrics, see View analytics.

Exporting metrics data to BigQuery has the following limitations:

• The data that is exported to BigQuery tables in your Vertex AI Search for retail project are authorized views, not materialized views.

• You cannot change or update the data in these tables. If you need to change or update your data, export or copy the view to another destination.

• You can't export metrics for user events that are older than 180 days or metrics that were generated more than 180 days ago. Any metrics generated before the 180-day window are not retrievable.

Before you begin

Before you can export your metrics into BigQuery, you must have completed the procedures in Before you begin. This includes importing your user events and, optionally, your catalog information.

For more information about importing, see:

• Import catalog information
• Record real-time user events
• Import historical user events

Create a dataset in BigQuery

You need to create a dataset in BigQuery to hold your analytics metrics data.

You must create the datasets in the same project where you implemented Vertex AI Search for retail.

1. If you haven't used BigQuery in your project before, enable the BigQuery API and make sure that you have the IAM role that lets you create datasets and tables.

   See Before you begin and Access control with IAM

   in the BigQuery documentation.

2. Create a dataset in BigQuery in the US (multiple regions in the United States) multi-region. For example, name it retail_data. If you create your dataset in a different region, exporting your data into BigQuery fails. CMEK for the dataset is not supported; enabling it causes the export to fail.

   For information about creating a dataset, see Creating datasets in the BigQuery documentation.

   This dataset will be used to hold the data table(s) that you export. The following procedures describe how to export.

Export your analytics metrics to a BigQuery table

Use the exportAnalyticsMetrics method to export your retail catalog into a BigQuery table.

Before using any of the request data, make the following replacements:

• PROJECT_ID: The ID of the Vertex AI Search for retail API project where you created the BigQuery dataset.
• DATASET_ID: The name of the dataset that you created in Create a dataset in BigQuery. For example, use retail_data or retail_metrics_data.
• TABLE_ID_PREFIX: A prefix for the table ID. This prefix can't be an empty string. A suffix of retail_analytics_metrics is added to complete the table name. For example, if the prefix is test, the table is named test_retail_analytics_metrics.

Request JSON body:

{
  "catalog": "projects/PROJECT_ID/locations/global/catalogs/default_catalog",
  "outputConfig":
  {
    "bigqueryDestination":
    {
      "datasetId": "DATASET_ID",
        "tableIdPrefix": "TABLE_ID_PREFIX",
        "tableType": "view"
    }
  }
}

To send your request, expand one of these options:

cat > request.json << 'EOF'
{
  "catalog": "projects/PROJECT_ID/locations/global/catalogs/default_catalog",
  "outputConfig":
  {
    "bigqueryDestination":
    {
      "datasetId": "DATASET_ID",
        "tableIdPrefix": "TABLE_ID_PREFIX",
        "tableType": "view"
    }
  }
}
EOF

curl -X POST \
     -H "Authorization: Bearer $(gcloud auth application-default print-access-token)" \
     -H "Content-Type: application/json; charset=utf-8" \
     -d @request.json \
     "https://retail.googleapis.com/v2/projects/PROJECT_ID/locations/global/catalogs/default_catalog:exportAnalyticsMetrics"

@'
{
  "catalog": "projects/PROJECT_ID/locations/global/catalogs/default_catalog",
  "outputConfig":
  {
    "bigqueryDestination":
    {
      "datasetId": "DATASET_ID",
        "tableIdPrefix": "TABLE_ID_PREFIX",
        "tableType": "view"
    }
  }
}
'@  | Out-File -FilePath request.json -Encoding utf8

$cred = gcloud auth application-default print-access-token
$headers = @{ "Authorization" = "Bearer $cred" }

Invoke-WebRequest `
    -Method POST `
    -Headers $headers `
    -ContentType: "application/json; charset=utf-8" `
    -InFile request.json `
    -Uri "https://retail.googleapis.com/v2/projects/PROJECT_ID/locations/global/catalogs/default_catalog:exportAnalyticsMetrics" | Select-Object -Expand Content

You should receive a JSON response similar to the following:

{
  "name": "projects/PROJECT_NUMBER/locations/global/catalogs/default_catalog/operations/12310973052792775280",
  "done": true,
  "response": {
    "@type": "type.googleapis.com/google.cloud.retail.v2.ExportAnalyticsMetricsResponse",
    "outputResult": {
      "bigqueryResult": [
        {
          "datasetId": "DATASET_ID",
          "tableId": "TABLE_ID_PREFIX_retail_analytics_metrics"
        }
      ]
    }
  }
}

Metrics definitions

The exported metrics are the count metrics. There can be slight drift for some count metrics because bot filtering might exclude some events from metrics computation. Bot filtering can help deliver more accurate performance measurements. This does not affect A/B testing, which depends on the dimension external_experiment_ids, whereas bot filtering is applied to all traffic.

Metrics dimensions

Count metrics are the raw counts directly aggregated from relevant user events with the same dimensions. To interpret metrics, you need to know which dimensions the metrics are based upon. For the purpose of analytics metrics, search and browse events are referred to as impression events, and detail_page_view, add_to_cart and purchase events are referred to as conversion events. The following table defines metric dimensions.

Impression metrics

Impression metrics are computed according to prerequisites. Conversion events (detail_page_view, add_to_cart, and purchase events), can be attributed back to an impression event only if they match the following criteria.

• Impression event and conversion event have the same visitor.

• The product ID of the conversion event must appear in the impression event's results.

• Conversion event is within the required time window after the occurrence of the impression event. Different conversion time windows apply for different types of conversion events.

  Conversion path	Conversion time window
  Impression -> Detail Page View	60 seconds
  Impression -> Add To Cart	24 hours
  Impression -> Purchase	48 hours

Search impression metrics

Browse impression metrics

Session metrics

A session is defined as a continuous period of interaction between a user and the system. Any user event marks the start of a session, and the end of a session is marked when no other event occurs for the next 30 minutes. The session is extended another 30 minutes whenever a user event occurs within a started session. After a session ends, a new session begins when the user engages with the system again.

The dimensions that characterize the session metrics are jointly defined by all user events within the session. Refer to metrics dimensions as how dimensions values are set for the session.

Search session metrics

A search session is a session contains at least one search event. Note that a search session can also be a browse session as defined by browse session metrics.

Browse session metrics

A browse session is a session contains at least one browse event. Note that a browse session can also be a search session as defined by search session metrics.

Visitor metrics

A visitor is uniquely identified by the visitor ID. Visitor metrics aggregate all the user event information for the visitor.

The dimensions that characterize the visitor metrics are jointly defined by all user events for the visitor. Refer to metrics dimensions as how dimensions values are set for the visitor.

Search visitor metrics

A search visitor is a visitor performing at least one search event. Note that a search visitor can also be a browse visitor as defined by browse visitor metrics.

Browse visitor metrics

A browse visitor is a visitor performing at least one browse event. Note that a browse visitor can also be a search visitor as defined by search visitor metrics.

Example usage

If you copy or export your metrics data to another destination, you can use SQL to get high-level or specific information about the exported metrics. Some common ways to explore your exported metrics using SQL are provided in the following sections.

Compute click-through rate (CTR)

Compute CTR for search per date per device type per traffic destination.

SELECT
  `date`,
  device_type,
  traffic_destination,
  SUM(detail_page_view_from_search) / SUM(search_event) as search_CTR
FROM copied_table
GROUP BY `date`, device_type, traffic_destination

Get browse-driven detail page view count

Get detail page view events that come from browse events for Google traffic with external_experiment_ids containing test_arm, and display all dimension information. external_experiment_ids is repeated string.

SELECT
  `date`,
  external_experiment_ids,
  device_type,
  serving_config,
  traffic_destination,
  entity,
  detail_page_view_from_browse
FROM copied_table
WHERE
  traffic_destination = 'Google' AND 'test_arm' in UNNEST(external_experiment_ids)

Get most-used serving configs

Get the most-used serving configs between '2023-10-01' and '2023-11-01' for Google traffic. Each search event should have an associated serving config.

SELECT
  serving_config,
  SUM(search_event) as total_search_count
FROM copied_table
WHERE `date` between '2023-10-01' and '2023-11-01' AND traffic_destination = 'Google'
GROUP BY serving_config
ORDER BY total_search_count