Deploying the model to a Vertex AI endpoint allows you to use Vertex AI's managed infrastructure, minimizing maintenance effort.
Batch prediction is ideal for this use case, as the churn predictions are required weekly, not in real-time. It is cost-efficient and straightforward to schedule batch jobs to generate predictions for the entire customer dataset at regular intervals.
Scheduling the batch prediction weekly aligns perfectly with the customer retention team's plan to act on predictions weekly.

One-hot is a good way to use categorical variables in regressions problems
https://academic.oup.com/rheumatology/article/54/7/1141/1849688
https://cloud.google.com/bigquery-ml/docs/reference/standard-sql/bigqueryml-auto-preprocessing

Feature attribution in Vertex AI (using tools like SHAP or integrated gradients) provides transparency into which features most influence the model's predictions. This helps identify whether demographic features are contributing disproportionately, which is essential for detecting and addressing fairness concerns. It aligns with stakeholder needs for explainability and bias detection.

https://developers.google.com/machine-learning/guides/text-classification/ it is a Word Embedding case

A multilingual embedding model maps semantically equivalent text from different languages into a shared vector space, so a Spanish query retrieves English and Japanese items from one index.
This avoids per-language indexes and the information loss and latency of a translation step in the query path.