Efficiency is a term that gets used in a few different guises. In each case what is being quantified is the amount of energy coming out of a system or device compared to the amount coming into it. For example:

Conversion efficiency is used to refer to the conversion of primary energy (fuel or natural resource) into a power plant vs. the electricity that the plant produces. So a coal plant with a typical 40% efficiency only converts 40% of the energy in the coal that goes into it, into electricity out. The remaining energy content of the coal is wasted as heat (unless this heat is used somehow, for example for district heating). A solar PV module of 10% efficiency only converts 10% of the solar energy arriving on its surface into electricity (again the remainder is largely lost as heat, which explains why solar panels heat up often substantially beyond ambient temperature in strong sunshine).

An efficiency measure is sometimes applied to grid transmission and distribution (T&D), by measuring the difference between electricity leaving power plants and that used by end-users; in other words losses between the two (you can hear power lines buzzing for example, indicating energy loss). In a modern power system these losses should as a benchmark be only 6-7% of the electricity that went into the network (i.e. an end-to-end efficiency of 93-94%). In practice, since the amount used by end-users is the amount that is metered, much lower grid efficiency is likely to indicate unmetered usage (electricity theft) rather than physical losses in the T&D infrastructure.

At the end-user premises, we think of efficiency in terms of products and their electricity consumption, or the “leakiness” of heat-containers such as walls and windows. The efficiency of electrical devices is harder to define in terms of a simple energy-in vs energy-out percentage, because in practice the energy output of different devices manifests in so many different ways: sound, heat, light, radiation and so on. So efficiency ratings such as “A”, “B” and so on are based on the energy consumption of a device compared to a benchmark alternative, to give a sense of whether one product will use less or more energy than another to achieve the same result (such as heating up your dinner).