Since I can't share any data and setup from my current project, I use my personal ad accounts and budget to launch UA campaigns. The idea is to showcase how I conduct A/B test and results.
When you launch a brand new adset and you define a target audience, you actually reach a portion of this whole audience. Some marketers call those portions pools. Those pools can be segmented based on purchase behaviours, engagement (such as shares, likes and comments), interests, etc. The longer the testing campaign run, the pool is getting more and more relevant because of Facebook's Machine Learning algorithm。
Of course every marketer's campaign goal is to get FB to hit the highest quality pools / the most suitable pools for our product/service. so the right way to tap into your relevant pools is by providing the highest total value.
Ads within an adset compete for the budget - DON'T RUN DIFFERENT ADS in the same adset! That is why you see I structure campaign that eash ads contain only 1 ad
If you will duplicate a brand new campaign/adset 10 times on audience of 2+mill (to avoid overlap) you will see that you're getting different results on each campaign/adset - it happens because FB hits on different pools even if the targeting is the same.
The main reason that structuring campaigns properly is so important is because it will allow you to split test properly and analyse data better by getting more reliable and accurate data.
There are 2 main ways I often to structure the testing campaign.
1 Campaign - 1 - 3 Adsets / ad 2 - 4 ads (same ad)
1 Campaign - 2 - 6 Adset / ad - 1 ad
For each unique ad, I will run a campaign with 2-6 duplicates of the same adset (same targeting & budget), inside each adset there is 1 ad. This is the favorite way to go with website conversion objective campaign.
Note: Sample creatives are NOT from the employer's campaign assets. (NDA restriction) The following samples are the demonstration of my ad angle skills. Creatives are tested with my own budget.
Use 3rd party tracking redirection & rotation. I distribute traffic among new landing pages. By precisely controlling the traffic volume of each testing object, I can filter out the best performing lander quickly.