When evaluating your data, there are several important factors to keep in mind:
Do you have enough volume to generate accurate data? For example, with fewer than 200,000 impressions per placement, per day, it’s difficult to generate a consistent CPM since small changes in CTR or post-click conversions may have big effects. Audience Network uses a predictive model to stabilize performance, and it’s difficult to generate this model without a large enough sample.
Has the volume of your inventory grown or contracted significantly during your test period?
Have you recently completed a marketing push for new users, or has your user mix stayed consistent during the testing period?
Have your user demographics, specifically geographic location, differed significantly in the time since you started the test?
Many campaigns compete for limited inventory at the end of the month, which can have the effect of driving CPMs higher. This happens most often on the final day of the month.
CPMs tend to be highest in Q4 and lowest in Q1. If your tests span several months, you may notice a gradual increase over time, so we recommend you conclude all tests within a short window of 1-2 months.