We characterize the optimal design of data analytics for algorithmic pricing in a market where firms aim to collude under secret price cuts. Firms obtain information either through predictive analytics before price setting --- i.e., ex ante information provision --- or through diagnostic analytics after price setting --- i.e., ex post information provision. The two modes of information provision generate a trade-off between a price sophistication effect and a profit dissipation effect. Such trade-off hinges upon the heterogeneity of demand, the efficiency of information technology, and the number of firms. Our analysis provides novel insights into the theory of harm for algorithmic pricing