An important practical consideration in system identification is the judicious use of information for input signal design. Typically, only limited process knowledge is available a priori; hence the input design parameters are not always optimally selected. The quality of the data substantially improves if input design parameters can be refined during experimental execution. The purpose of this paper is to present an enhanced identification test monitoring procedure for multivariable systems that incorporates these ideas to achieve experiments with the shortest possible duration and that are adequately informative for identification purposes. This is enabled by 'on-the-go' manipulation of amplitude, duration, and/or frequency content of the input signals. The decision to continue, modify, or halt the experiment is achieved by a stopping criterion based on a robust control metric that is developed in the paper. These computations are performed using an orthogonal-in-frequency spectral input design, and relying on a computational method that estimates transfer functions (and associated uncertainties) taking into account the system noise and transient behavior. Results are evaluated through a simulation study using a chemical process system under diverse realistic noise structures.