Question

Your team deployed a Vertex AI model that approves or denies personal loan applications. A new regulation requires that any applicant who is denied must receive a clear, individualized explanation of the specific factors that most influenced their denial, so they can understand how to improve future applications. Compliance and legal teams need this capability without changing the model itself. Which approach best satisfies this responsible-AI requirement?

Accepted Answer

Enable Vertex Explainable AI with feature-based attributions (e.g., sampled Shapley or integrated gradients) and surface per-instance attributions for each denied application.

Answer

Configure Vertex AI Model Monitoring for feature-attribution drift and send the aggregated drift report to denied applicants.

Answer

Log all prediction requests and responses to BigQuery and provide applicants the raw probability score returned by the model.

Answer

Retrain the model using only interpretable, human-readable feature names and publish global feature importance in the model card.