Skip to main content

Google launches Recommendations AI ecommerce tool in public beta

Google Recommendations AI
Image Credit: Google

During its Cloud Next 2019 conference, Google unveiled Recommendations AI, a fully managed service designed to help retail-oriented businesses deliver personalized product recommendations to their customers. Beginning today, following a lengthy preview period with early adopters that include Sephora, Boozt, and Digitec Galaxus, Recommendations AI is available in beta to eligible Google Cloud customers.

Google says Recommendations AI was informed by work across properties like Google Ads, Google Search, and YouTube. Using machine learning to dynamically adapt to customer behavior and changes in variables like assortment, pricing, and special offers, it ostensibly boosts click-through rates and conversions on web, mobile, and email while increasing the revenue driven by recommendations and total revenue per visit.

According to product manager Pallav Mehta, Recommendations AI excels at handling recommendations in scenarios with long-tail products and cold-start users and items. Thanks to “context hungry” deep learning models developed in partnership with Google Brain and Google Research, it’s able to draw insights across tens of millions of items and constantly iterate on those insights in a real-time way. Recommendations AI is also capable of correcting for bias with popular or on-sale items and can better handle seasonality (or items with sparse data), while its infrastructure allows for daily model retraining.

From a graphical interface, businesses using Recommendations AI can integrate, configure, monitor, and launch recommendations while connecting data by using existing integrations with Merchant Center, Google Tag Manager, Google Analytics 360, Cloud Storage, and BigQuery. Recommendations AI can incorporate unstructured metadata like product name, description, category, images, product longevity, and more, and it can customize recommendations to deliver desired outcomes, such as engagement, revenue, or conversions.

VB Event

The AI Impact Tour – Atlanta

Continuing our tour, we’re headed to Atlanta for the AI Impact Tour stop on April 10th. This exclusive, invite-only event, in partnership with Microsoft, will feature discussions on how generative AI is transforming the security workforce. Space is limited, so request an invite today.
Request an invite

Google Recommendations AI

Recommendations AI also lets Google Cloud customers apply rules to fine-tune what shoppers see and diversify which products are shown, filtering by product availability and custom tags. It supports international product catalogs in multiple geographies and serves recommendations anywhere in a customer’s journey, whether on a homepage, during order confirmation, or in a shopping cart.

Once the initial data import is complete, Recommendations AI users get a choice of model type and optimization objective. Model training and tuning takes two to five days, Google says, and the model’s recommendations can be previewed before they’re served to customers.

To accompany the Recommendations AI public beta, Google is introducing a new pricing structure with three volume-based price tiers for predictions and a separate charge for model training and tuning. All new Recommendations AI customers will receive a $600 credit on top of the general $300 free credit for new Google Cloud customers, which the company says is typically sufficient to train a model and test its performance in production through a two-week A/B test.

Google’s Recommendations AI competes with Amazon Personalize, which similarly taps machine learning to serve up suggestions for websites, SMS, email, and apps. According to Amazon, Personalize addresses problems like creating recommendations for new users or products without historical data via API calls that automate the tasks required to build, train, tune, and deploy a recommendation model.

VB Daily - get the latest in your inbox

Thanks for subscribing. Check out more VB newsletters here.

An error occured.