Integration

Zembula + Google Vertex AI Search for commerce

Google Vertex AI Search for commerce predicts the products to recommend to a person, from catalog and event data. Zembula can read those predictions from a placement and show them in an email.

It's our business to grow yours

Book a demo

What Zembula can read from Google Vertex AI Search for commerce

Vertex AI Search for commerce is a Google Cloud service. It predicts the products that fit a person, from catalog data and visitor events. Zembula can call the predict method for a placement and read the recommended products for one visitor ID. The request carries the visitor ID, the product details and the user info.

The products that come back are items from your catalog, and each one has a product ID. The models learn from user events, which are the views, carts and purchases of a person. Google runs the predict call on the server side. To show recommendations, an email needs a visitor ID and a placement.

Google sets quotas on this service, including 60,000 predictions per minute and 300 product reads per minute. The quotas apply to recommendations only. You need a Google Cloud project, and the service uses OAuth2. A catalog without search enabled can hold up to 40,000,000 products.

DataDetailLooked up byHow fresh
Recommended productsProducts that Google predicts for a person, from a placement.visitor IDopen time
Catalog productsThe catalog items that recommendations return.product IDset at setup
User eventsViews, carts and purchases that train the models.visitor IDset at setup

How the data arrives

  • API: Zembula can call the predict method of a placement and read the recommended products for a visitor ID.

What you need

  • An API key or a data export from Google Vertex AI Search for commerce. Zembula sets up the connection for you.
  • A Google Cloud project with Vertex AI Search for commerce and a catalog.
  • The placement that returns your recommendations.
  • A visitor ID in your email service provider, sent as a merge tag in the image URL.

Limits to plan for

LimitWhat it means
Predictions per minute: 60,000.Open-time predictions above this rate go over quota.
Product reads per minute: 300.Catalog reads above this rate go over quota.

Ideas for email

  • Email a person the predicted products for a viewed item.
  • Add predicted add-on products to a cart reminder.
  • Refresh a recommendation block with the results of a placement.

Last checked: 19 November 2026.

Performance marketing dials, inside email. For the first time.

See how we do it by booking a demo.