Radioactive decay is random. It is impossible to predict when a particular nucleus will decay. Two identical nuclei sitting side by side may decay seconds apart or millions of years apart, and nothing you do, such as heating the sample, will change that.
A sample contains a very large number of nuclei, though. Each nucleus has the same chance of decaying in a given time, so across a large number of nuclei the decay follows a pattern. This is why half-life works: in each half-life, half of the undecayed nuclei decay. If a sample has 4000 undecayed nuclei and a half-life of 5 minutes, about 2000 undecayed nuclei will be left after 5 minutes.
So half-life lets us predict the activity of a large sample, but it cannot tell us which nucleus will decay next or when. With only a few nuclei the pattern is unreliable, and a count rate measured over a short time will fluctuate slightly from one reading to the next.