“The application of mathematics to natural phenomena is the aim of all science, because the expression of the laws of phenomena should always be mathematical” (Bernard, 1865). The need for mathematical tools has long been acknowledged by physiologists, in order to integrate data from past experiments with the testing of possible hypotheses about underlying mechanisms, and thus to guide future experimental design. A synergy between the wet-lab and modelling approaches enabled Hodgkin-Huxley, Noble and DiFrancesco to unravel the interaction of ion currents in many different cell types. The computational revolution led to a separation between “mathematical modellers” and “physiologists”, reducing communication accordingly. A major criticism of current mathematical modelling in cardiac electrophysiology is that most of the results only replicate what is already known experimentally, even though (or because) the mathematical models have become evermore complex. We propose a set of tools to ensure validated and predictive models (applied here to ion-channel models and voltage clamp experiments). These tools investigate the relation between experimental data, its information content, and model structure. Currently, multiple voltage clamp step experiments are done to determine the kinetics of ion channels (activation, deactivation, inactivation and reactivation); most often these are undertaken in different cells, and the results have to be normalised for analysis. The full set of protocols produces redundant data and we show a much shorter protocol (length <15 s) that suffices to evaluate models’ parameters. We compare the information content of the optimized step protocol to action-potential and ramp clamp protocols. The optimization of experiments is a necessary step for providing predictive models, as there is a lack of knowledge of how much, and which kind of, experimental data suffices for a unique fit of a model’s parameters. As a result many models are fitted using all the available data, which can be inconsistent and leaves no data for model validation. A model simply matching the experimental data does not imply a “good fit” if the model is overparameterised. We analyse simulated experimental protocols and models to establish whether there is sufficient information to fit all of the models’ parameters uniquely. This provides a separation between the ‘training’ data used to fit the model parameters and ‘validation’ data used to evaluate the predictive power of the model. This study of each protocol’s parameter-information content allows us to design optimised patch-clamps for voltage gated ion channels. Fitting model parameters to such protocols enables us to evaluate and then ensure the model structures’ ability to predict the current during any other clamp.