Although virtually all industrially relevant processes exhibit nonlinear behavior, a significant number are still operated using linear controllers. Such controllers, especially those of the classical single-loop PI(D) type, remain popular because they are relatively easy to design they require very modest modeling efforts, and straightforward tuning rules abound-and they are convenient to implement using standard hardware. In this paper, we present a technique for constructing nonlinear PI-type controllers using low-order empirical nonlinear models. These controllers, by design, retain the simplicity and structure of the familiar PI controllers but are more effective over wider operating regimes and are thus capable of better performance on nonlinear processes. As part of the technique, we present a procedure for process characterization from input/ouput data, as well as explicit extensions for implementation on multi-input multi-output processes. We demonstrate the controller design and implementation procedure on simulations of industrially relevant SISO and MIMO processes.