This paper proposes a machine learning pipeline to predict changes in existing Philippine consumer price index (CPI) measures for various commodity groups using aggregate quantity-based e-commerce information and bidirectional long-short term memory (Bi-LSTM) recurrent neural networks. It explores a solution to the inevitable challenge of forming the appropriate basket of goods from among potentially hundreds of thousands of e-commerce records once full-blown data sharing agreements with technology companies are in place. Specifically, e-commerce items are systematically measured for their predictive contribution through backwards sequential feature selection. Results show e-commerce data is useful for forecasting some commodity groups but not others, which may indicate that the shift from brick-and-mortar to online trade is nearing maturity in some Philippine industries.