Cloud Digital Leader Practice Question: Google Cloud products, services, and solutions

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  BigQuery ML — build a collaborative filtering model using SQL.

  Why it's wrong here

  BigQuery ML supports building machine learning models using SQL, including matrix factorization for collaborative filtering, and it works well for batch offline experimentation. However, you still own the ML workflow: choosing the algorithm, tuning hyperparameters, validating offline metrics, and then exporting the model to a separate serving layer (e.g., Vertex AI or Cloud Run) to get the sub-100ms inference latency that live e-commerce demands. It lacks the retail-specific data schema, feature engineering, and automated model lifecycle management that Recommendations AI provides.

• ✓

  Recommendations AI (Vertex AI Search for Retail)

  Why this is correct

  Recommendations AI, now part of Vertex AI Search for Retail, is purpose-built for e-commerce personalization: it starts with deep neural network models pre-trained on large-scale retail patterns and fine-tunes them on your product catalog and user event data (views, clicks, purchases). The service automatically handles feature engineering, model training, and low-latency serving, returning personalized recommendations through a simple predict API. It also addresses cold-start users/items and supports optimizing for business objectives like revenue lift, without the need for ML engineering.

• ✗

  Cloud SQL — query purchase history to find commonly bought-together products.

  Why it's wrong here

  Cloud SQL is a relational database for transactional workloads, and while standard SQL aggregation on purchase history can surface simple association rules like frequently co-purchased products, that is a static, rule-based approach. It cannot capture a user's current session context, personalize for individual preferences in real time, or leverage deep learning over behavioral data as Recommendations AI does. Moreover, complex affinity-query logic in SQL doesn't scale to low-latency personalized recommendations for millions of users.

• ✗

  Cloud Dataflow — stream user clickstream data to build recommendations in real time.

  Why it's wrong here

  Cloud Dataflow is a unified stream and batch data-processing service, ideal for ingesting and transforming clickstream events into a feature store, but it does not train or host recommendation models. Building recommendations on Dataflow would require writing custom ML training and inference pipelines, integrating with external frameworks, and managing model versioning and serving infrastructure — far more effort than using a purpose-built, managed recommendation engine.

Go deeper

Related to this question