A word problem describes a hidden number. "Three more than twice a number is 17" says what was done to the number (doubled, then 3 added) and what came out (17). Writing x for the number turns each phrase into a symbol: "twice a number" is 2x, "three more than" adds 3, and "is" becomes =. The sentence becomes 2x+3=17, and the usual solving steps give x=7.
Real stories need one more decision: exactly what the letter stands for, with units. In the bike problem, w is the number of weeks of saving, not an amount of money. Writing "let w = weeks" first stops you from adding weeks to dollars and tells you how to read the answer. Every condition must appear somewhere. If the shop also charged $12 for a helmet, the target would be 180+12=192, so 60+15w=192 gives w=8.8, which means saving for 9 weeks.
A model that only describes, with no missing value to find, is a formula rather than an equation to solve. "A taxi charges $3 plus $2 per mile" gives C=3+2m, true for every trip. It becomes an equation to solve only when the story adds a target, such as "the fare was $15": then 3+2m=15 and m=6 miles. Look for a stated result that the unknown must produce.
Think of it like this. Modeling is translation between languages. In French, "a red car" becomes "une voiture rouge": the meaning survives but the word order changes. In the same way, "7 less than x" becomes x−7, not 7−x. A translation is checked by reading it back; a model is checked by putting its answer into the story.
Core idea. Modeling with equations turns a word problem into equations you can actually solve for the unknowns.