small-signal model
A transistor's real behavior is curvy and nonlinear — push its gate harder and the current it passes bends and saturates rather than tracking in a tidy straight line. That sounds like a nightmare to do math on, until you notice a trick: if you sit the device at a fixed DC operating point (a steady bias of voltage and current) and then only ever wiggle it by a tiny amount, that little patch of the curve looks essentially straight. The small-signal model is exactly that — you zoom way in on the curve right where you're biased and replace the messy nonlinear device with the simple straight-line slope at that one spot. Like approximating a small stretch of a hill by its local tangent: globally curved, locally flat enough to treat with plain linear math.
What the slope gives you are two clean handles. Transconductance gm is how strongly a small input voltage on the gate steers the output current — the slope of current-versus-input — so for a saturated MOSFET it works out to gm = 2*Id/Vov, where Vov is the gate overdrive. Output resistance ro captures the gentler truth that the current isn't perfectly fixed by the gate; it creeps up as you raise the output voltage, and ro is how stiffly the device resists that creep. Together they let you swap the transistor for a controlled current source (gm times the input wiggle) sitting in parallel with a resistor ro, and suddenly the whole amplifier is just a linear circuit you can solve by hand.
Out of those two numbers falls the single most important figure of merit for one device: its intrinsic gain, gm*ro. This is the most voltage gain a single transistor can ever hand you — the gm converts your input wiggle into a current, and that current develops a voltage across the largest load the device itself offers, its own ro. A textbook common-source stage loaded only by ro has a small-signal gain of -gm*ro (the minus sign just means it inverts). Want more gain than one device can give? You stop fighting the math and start stacking or cascading devices, because gm*ro is the ceiling for a single stage.
For a saturated MOSFET, transconductance comes from the bias current and overdrive; multiply by the device's own output resistance and you get the most gain one transistor can deliver.
The model only holds while the signal stays genuinely small (the wiggle much less than the overdrive) and the transistor stays in saturation — drive it hard or into another region and the straight-line approximation quietly stops telling the truth.