This is a measurement guide, not a customer success story. The brand narrative and conversion uplift previously published here have been withdrawn because their supporting evidence could not be verified. No customer result is claimed.
Choose one question
For example: does adding an on-model scale image to the existing jewelry gallery change purchase behaviour? Specify exactly which image changes. Keep the product description, price, promotion, stock availability and delivery offer consistent between groups.
Approve product accuracy first
Compare the output with the real item. Check chain thickness, clasp, stone count, setting, engraving and metal colour. An attractive image that misrepresents the jewelry should not enter a sales experiment. Keep the source photograph and the approved variant.
Define the comparison before launch
Assign eligible visitors to the original or new gallery concurrently and keep their assignment consistent. Define the conversion event and denominator in advance. If using sessions that resulted in a purchase divided by eligible sessions, use that definition in both groups. Plan the sample and observation window before looking for a winner; a change between two different weeks is not the same as a controlled comparison.
Record the complete outcome
For each group, save eligible sessions, purchasing sessions, revenue, device mix and image loading behaviour. Include returns and complaints when enough time has passed. Do not remove inconvenient results or stop at the first favourable fluctuation.
Interpret the result carefully
Report both conversion rates, their percentage-point difference, sample sizes and uncertainty. A relative change needs a nonzero baseline. A small or inconclusive sample does not justify a growth claim. Decide whether to keep, revise or reject the gallery change; production cost and image fidelity remain separate measures.