An experiment compares a treatment with a suitable control. Start with a question the result can answer, define the primary outcome before reading results and keep unrelated settings stable. Changing bids, landing content and ad copy together makes it difficult to attribute a difference to any one change.
Choose an outcome close to the business decision. Clicks can be suitable for a traffic test, while a lead acquisition test needs useful conversions and a value test needs a consistent value definition. The browser creates a fixed fictional treatment difference rather than implementing statistical inference. It doesn't calculate significance, confidence intervals or a minimum sample size. The readout questions test whether the experiment supports the claim being made.