A full swimming pool holds 50,000 litres, and a pump removes 2,000 litres every hour. No formula is given, so you build one. What changes? The volume. By how much? The same amount every hour, so the structure is linear: V(t)=50000−2000t. Now you can predict: the pool is empty when 2000t=50000, after 25 hours.
Modeling runs in a cycle: pick the quantities, choose the structure, fit the numbers, predict, then compare with reality. If a real pool drains more slowly as the water gets shallower, because there is less pressure at the outlet, the linear model will be wrong near the end and you revise it. The main choice is the pattern of change: a fixed amount per step gives a line, a fixed percent per step gives an exponential, and a quantity that rises and then falls may be quadratic.
Not every word problem is modeling. If a problem hands you C=15n+40 and asks for C when n=10, the model is already built and your job is evaluation. Curve fitting sits in between: the form is chosen for you ('fit a line to these points') and you find only its constants. Modeling starts one step earlier, when you must decide what shape the relationship has.
Think of it like this. A model is like a scale model of a bridge in a wind tunnel: built to copy the features that matter for one question, how it bends in wind, and to ignore the rest, such as paint and traffic. It predicts well for that question and can be useless for others, such as how fast the real bridge will rust.
Core idea. Mathematical modeling turns a real situation into a function or equation so you can compute and forecast answers you could not just look up.