Should the experiment goal be purchases or add-to-cart clicks?
Hello everyone. I am a CRO specialist at an agency and I keep having the same debate with clients. They want "purchase" as the primary goal on every test, including tests on category pages that sit five steps away from checkout. My position is that the goal should sit closer to the element being tested, but I would like to ground the advice in how Croct actually works.
Specifically: how does the choice of primary goal interact with the thresholds an experiment must clear before it resolves? And is there a way to keep an eye on the macro metric without making it the primary goal
2 answers
The thresholds themselves make the argument for putting the goal closer to the tested element.
An experiment stays in progress until each variant reaches 1000 visitors and 25 conversions of the primary goal, over at least a week. The conversion threshold is the one a distant macro goal punishes. If your category page test uses purchase as the goal and only 1.5% of category visitors ever buy, you need enormous traffic per variant before each one collects 25 purchases. Worse, the signal is diluted: four other funnel steps sit between the tested element and the goal, each adding noise that has nothing to do with your change.
A micro goal close to the element, add-to-cart from that page, resolves much faster because the conversion rate is higher and the causal chain is short. The honest caveat to give clients is proxy risk: you can win the micro metric while doing nothing for revenue.
The workable pattern is both. Set the micro conversion as the primary goal so the experiment resolves in reasonable time, then use the results drill-down by custom events to check the macro impact on the same participants. You get a fast decision plus a sanity check that the proxy did not lie.
The 25-conversions-per-variant framing is exactly the concrete argument I needed. Merci, this goes straight into my client onboarding deck.
One practical note on setup: the primary Goal ID has to reference an event that is already being tracked, you cannot point it at a metric that exists only in your head. If the client store runs one of the commerce integrations, events like orderPlaced and checkoutStarted are tracked automatically and the conversion-indicating ones record a Goal Completed, so both the micro and macro candidates are usually already available. Worth checking the event stream before the goal debate, sometimes the event they want is not wired up yet and that settles the discussion for the first test.