Brand advertising is a major objective in digital advertising, yet most real-time bidding methods are designed for performance campaigns. We propose a lightweight model predictive control framework for brand auction ads that exploits their stable engagement patterns and fast feedback loops. The method builds monotone bid-to-spend and bid-to-conversion models online using isotonic regression, then chooses bids by inverting these models under budget and cost constraints. The framework is fully online, computationally lightweight, and easy to deploy. Offline simulations and online
A/B tests show improved spend efficiency and cost control over PID and dual-gradient baselines.