How an experiment works
An experiment begins with a draft — a copy of a live campaign carrying whatever change you want to test. Turning that draft into an experiment launches it alongside the original, and Google divides the eligible auctions between the two according to a split you choose. Both versions then run at the same hours, in the same auctions, against the same competitors, and the interface reports them side by side.
The split is applied consistently rather than reshuffled on every impression, so a person who comes back tends to stay on the same side of the test instead of seeing both versions. Google also marks each metric with an indication of how confident it is that the gap between the two arms is real rather than noise, which is the part most people skip and most need.
Experiments are a family rather than one feature. A campaign experiment tests a variant of a whole campaign, a custom experiment lets you decide exactly what to change on Search or Display, and a video experiment compares creative on YouTube. All of them share the same shape: one deliberate change, an untouched control, a fixed run, a decision at the end.
Why experiments matter
Comparing this month with last month is the standard way marketing decisions get made, and it is unreliable. Demand moves with the season, competitors change their bids, a festival lands in one period and not the other. An experiment removes the calendar as an explanation, because both versions live through exactly the same weeks.
It is also the safe way to try something that could go badly. A bid strategy change applied to a whole campaign exposes all of your budget to the outcome; the same change in an experiment exposes only a share of it, and can be stopped in a click. For a client or a manager, it turns an argument about opinions into a result anyone can read.
Common mistakes with experiments
Changing more than one thing is the classic. If the variant has new ad copy, a different bid strategy and tighter targeting, a win tells you nothing about which of the three earned it, and you cannot apply the lesson anywhere else.
The rest cluster around impatience. Stopping the moment the variant pulls ahead usually means stopping on noise. Running a test on a campaign with very little traffic, which is common for local businesses and specialist services in Nepal, produces a difference that will not survive another week. And forgetting an experiment has an end date means the change quietly disappears when it lapses, along with any benefit it was producing.
How to act on it
Before you start, write down three things: the single change, the metric that will decide it, and the minimum time it will run. The deciding metric should be a business one — cost per lead, revenue, qualified enquiries — not click-through rate, which moves easily and pays nobody.
Let it run in whole weeks so weekday and weekend patterns cancel out on both sides, and leave it alone in between. When it ends, apply the winner to the original campaign, keep the control, or promote the variant to a standalone campaign — but decide deliberately, and record what you learned. A tested change that nobody wrote down gets re-argued six months later.
