Setting up an experiment starts from an existing offer, not a blank slate — you pick a base offer, then define one or more challenger variants that override specific fields on it (price, copy, layout, product, or placement, depending on what's testable for that offer's context). Traffic is split evenly across variants by default, computed automatically from however many variants you've defined, rather than something you set manually per variant.
Before an experiment can start, you choose a primary metric and supply a baseline rate, a minimum detectable effect, and your approximate daily traffic; the wizard uses those to calculate the sample size the test will need, which is also what determines roughly how long it'll need to run to reach a reliable result rather than an underpowered one. You can also configure an automatic end condition rather than watching the test and ending it by hand.
This is a genuinely different tool from a single offer's Smart Rules: Smart Rules decide whether an offer shows to a given shopper, while an experiment decides which version of an eligible offer that shopper sees, for the purpose of measuring which version performs better — the two layers operate independently and stack on top of each other.