Why is there a discrepancy in the number of users in each AB testing group?

You might expect that the numbers of users in each A/B testing group would be equal, however this is often not the case due something know as the Gamblers fallacy.

This fallacy is the false assumption that streaks must eventually even out in order to be representative of the overall probability. For example, that if you were to toss a coin 100 times, you would have 50 heads and 50 tails.

In reality, because the probability of each event (the event being a coin toss, or a user being assigned a variant in the AB test) is entirely separate from any proceeding or following events, it is totally possible to have a streak of events all landing the same way such that the overall distribution ends up being slightly skewed