This paper presents a PI control configuration for a class of multi-input multi-output (MIMO) plants. The control design procedure consists of a modeling error estimator based on a reduced-order observer structure and a nominal feedback function. The estimated modeling error signals are used in the feedback loop to counteract the effects of uncertain parameters. It is shown that this design procedure leads to a classical multivariable PI control configuration with novel parametrization of the controller gains. An advantage of the proposed control design is that easy tuning procedures can be designed. In addition, several special two-input two-output PI control configurations with decentralized structure can be obtained. As an application example, the separation control in an integrated three-product (Petlyuk) distillation column is studied. Sufficient conditions to achieve decentralized regulation in terms of the steady-state gain matrix are provided. Numerical simulations are used to show the effectiveness of the proposed PI controller under measurement dead times and uncertain parameters.