Mathematical models are useful to represent, understand, and predict the response of animals to the nutrient supply. These models often use a compartmental representation, involving a large number of compartments and parameters. However, few (if any) mechanistic models are capable to fully grasp the temporal and spatial aspects of nutrient metabolism, and much of what we do not understand or fail to quantify ends up as maintenance or inefficiencies. Practical application of a model requires the transformation of the model into a user-friendly tool. Decisions have then to be made as to which parameters will be hard-coded in the tool compared to those that can be modified by the user based on knowledge and observations. Proper parameterization of a model allows identifying parameters that have most impact on outcomes (in an independent way) and facilitates parameter estimation. For example, empirical equations such as the Gompertz or logistic function are often used to represent the potential protein or lipid deposition in growing animals. One of the parameters in these models is the protein or lipid mass at maturity. However, little information is available on the mature protein or lipid mass of animals such as pigs and poultry fed ad libitum throughout their life. Reparameterization of the model (e.g., replacing the mature protein mass by the protein deposition during the productive life of the animal) and the use of proxies (using body weight gain as a proxy for protein deposition) are essential for the practical application of the model. Recent developments in sensor technologies including real-time monitoring will undoubtedly revive the interest in nutritional modeling, especially in the context of precision livestock farming. These developments may involve a shift in paradigm in model development from a “concept-driven approach” towards a more “data-driven approach”, where models have to accommodate observed data.