What is a sufficient # to give a reliable indication for future performance? Is 5000 impressions enough? 10,000? 30,000? 100,000??

The # of impressions is actually not the best criteria to properly split test creatives. For example, let’s say that the magic # was 30,000 impressions. Well, depending on the size of the demographic you’re targeting, it may take 30 minutes or it may take 30 days to hit your goal.

A better way is to run a full week’s worth of traffic. Account for things like weekday vs. weekend traffic, daytime vs. afternoon vs. evening vs. late night traffic (or simplified “at work” vs. “after work” traffic).

Also, when collecting & comparing data, never compare today’s performance with yesterday or the day before. Instead compare data week by week, meaning Tuesday (today) vs. last Tuesday vs. the Tuesday before that.

So how much testing is enough? A week’s worth.