Find the friction before proposing the test
Conversion-rate optimization is not a collection of button-colour tricks. Start with analytics, search terms, support conversations, return reasons, session recordings and customer interviews. Look for repeated problems such as unclear shipping, poor filtering, missing specifications, confusing variants or errors at payment. Then choose one problem and one hypothesis at a time. A redesign that changes navigation, photography, copy, pricing presentation and checkout simultaneously may improve results, but it will not reveal which change mattered.
Measure the step you are actually trying to improve
Overall conversion can hide where the problem lives. Track product views to cart, cart to checkout, checkout completion, search usage, filter engagement and differences by device or traffic source. A change to category filters should be judged partly on discovery behaviour, not only final sales. Protect trust while experimenting. Aggressive urgency, hidden fees or manipulative defaults may create a short-term lift while increasing complaints or returns. Document each hypothesis, change and result so failed experiments prevent the same weak idea from being recycled later.