The Search for commerce console provides metrics to help you determine how incorporating Vertex AI Search for commerce is affecting your business.
View metrics
You can view summary metrics for your project on the Analytics page. This page displays site-wide metrics about revenue and orders. See Summary metrics for metrics definitions.
For metrics specific to your serving configs, go to the Serving Configs page. Click a serving config name and select the Analyticstab to see its metrics. See Configuration-specific metrics for metrics definitions.
Metrics definitions
This table provides definitions of the summary metrics that Vertex AI Search for commerce displays on the Analyticspage.
Configuration-specific metrics
You can see metrics for a specific serving config on the Serving Configs page. For metric graphs, click the configuration name and select the Analyticstab.
This table provides definitions for configuration-specific metrics.
predict
or search
queries for this configuration.predict
queries on the shopping cart page.To track clicks from a serving config, Vertex AI Search for commerce
aligns the recommendations in predict
responses and search results in search
responses with ingested user events. If a clicked item appears in the predict
or search
responses for the same visitor ID within a one hour time window, the
click and purchase is treated as a result of the recommendation or search result.
When you configure your recommendations or search requests for the first time, you should confirm that:
- Visitor IDs in the request are the same as the visitor IDs you used in event ingestions.
- The timestamp in the response roughly matches the timestamp for that event.
When Vertex AI Search for commerce metrics are compared to the ideal expected result, or ground truth, the values might be lower, but the trends align.
Unlike for recommendations, the process of attributing click to search requests is not fully automatic. So for metrics to work for Vertex AI Search for commerce, an attribution token is required. However, attribution tokens are not needed for recommendations user events.
