Reliable predictions from systems biology models require knowing whether parameters can be estimated from available data, and with what certainty. Identifiability analysis reveals whether parameters are learnable in principle (structural identifiability) and in practice (practical identifiability). We introduce the core ideas using linear models, highlighting how experimental design and output sensitivity shape identifiability. In nonlinear models, identifiability can vary with parameter values, motivating global and simulation-based approaches. We summarise computational methods for assessing identifiability noting that weakly identifiable parameters can undermine predictions beyond the calibration dataset. Strategies to improve identifiability include measuring different outputs, refining model structure, and adding prior knowledge. Far from a technical afterthought, identifiability determines the limits of inference and prediction. Recognising and addressing identifiability is essential for building models that are not only well-fitted to data, but also capable of delivering predictions with robust, quantifiable uncertainty.