How do you balance latency and operational costs when choosing between different AI models for Airwallex Pty?
