This paper investigates the potential of commercial buildings to offer ancillary services to the grid. We consider a tracking procurement inspired by secondary frequency control, where the resource needs to commit to a nominal consumption profile ahead of time and then receives a power consumption deviation around this profile. We propose a model-based predictive scheme which formally assesses the amount of power flexibility that can be offered by the resource and of attaining the best trade-off between comfort and financial gain during real-time operation by adjusting the nominal power consumption with a delay, in accordance to the regulations of the Swiss energy market. Extensive simulation results highlight how the possibility to perform adjustments on the scheduled energy profile is of paramount importance for both maintaining comfortable operational conditions and offering larger tracking flexibility. Finally, the effectiveness and robustness of the proposed methodology are demonstrated by means of full-day experiments in an occupied office space.