First see the convergence of the paths of White’s queen and rook at h7, where he would mate were it not for the protection supplied to that square by Black’s queen. One way to interpret this is to say Black's queen is frozen on the seventh rank (it can’t afford to go anyplace out of reach of h7); another is to say that White can, by playing RxN, force Black’s queen over to h7. Both perspectives are accurate; the question is which is more useful. To answer this you need to consider your other offensive options (or near-options) and how Black’s queen bears on them. Perhaps the simplest thing is to imagine Black’s queen over on h7, for then it's easier to see the triangle between Black’s king, rook, and f6 pawn—and thus a queen fork for White with Qxf6+. Black’s queen is revealed to be overworked: it protects the mating square h7 (and the knight that sits there), and it also protects the forking square f6. The fork doesn’t come into existence until the queen moves out of the way, so RxN rather than Qxf6 is the right way for White to begin.

There still is a loose end, of course: the rook that is the target of the fork is guarded by the knight on c6. So naturally you start with 1. BxN, b7xB—then 2. RxN+, QxR; 3. Qxf6+ and 4. QxR, winning a piece.

You also might have seen the idea here by simply examining your checks: Qxf6 and RxN. You see that Black’s queen recaptures in both cases—and this causes you to wonder what happens if you play one of the checks and then the other.