No one ever counts 5 million of anything with perfect accuracy. A RLA is done to ensure that the count was accurate. They select a number of ballots randomly and see how the results compare with the initial count. The number chosen depends on how accurate you have to prove the count. Statisticians can show, with enough ballots counted, that the original result is within say .5% of the recount, with 90% accuracy.
Choosing 1.5 million random ballots is not difficult, but it is tedious. Counting all 5 million votes was deemed better than a RLA, and easier to accomplish. It showed they were within .01% of the original count, with as close to 100% accuracy as they could get.