After training a text classification model in TensorFlow using AI Platform, you need to perform batch predictions on a large dataset of text data that is stored in BigQuery. Your goal is to minimize computational overhead during this process. What is the most efficient approach?