Cut your loop at some convenient point. You may need to add components so the cut ends still see the same source and terminating impedance. Apply a frequency sweep to the input side of the cut. Measure the response at the output side of the cut. The "transfer function" is the Output divided by the Input, usually volts, although other units such as current and power are possible. Plot the transfer function against the input variable, usually voltage at changing frequency. Any place the transfer function blows up to infinity (or very large) is a pole. Any place it falls to zero, (usually at zero frequency and infinite frequency), we have a zero. By stability, most engineers mean unconditional stability: stable hot, cold, new, old, high line, low line, parts anywhere in their specified range.... You can tinker one or two units into stability by playing pole-zero games. It may not be unconditional, and forget this for production. Take the lowest rolloff at the high frequency end and lower it with more C until the loop gain drops below 1.000 before the second lowest rolloff is significant. Some margin may be necessary. I go for the first stage which I always design for the highest gain. I add a capacitor from stage output to stage inverting input. (E.G. Plate to grid, Collector to base, or drain to gate.) This is a "Miller Effect" capacitor. The foregoing applies to linear amplifiers. Servo systems are much more complex, but the same definition applies to poles and zeroes.