01
Write one test question
Choose a specific uncertainty: does showing the product in use answer buyer questions better than showing the item alone? Prepare two truthful creatives for the same approved product. Keep the destination, product option and amount consistent so you are not comparing several offers at once.
Record the hypothesis before you begin. If version A uses a different audience and a lower price while version B changes the video, the result cannot isolate the video’s effect. You can still learn from the campaigns, but describe the comparison honestly rather than calling it a controlled creative test.
02
Set a budget and stop condition first
Choose an amount you can afford to lose and define when you will review the test. Record spending separately for each version. Do not decide a universal winning threshold from someone else’s campaign: product commission, delivery outcomes and your costs differ.
If one version runs out of stock or its destination fails, pause the comparison and record the interruption. Do not quietly include the affected interval as though both versions had the same opportunity. A limited test can finish without enough evidence; that is an acceptable result, not a reason to keep spending indefinitely.
03
Use a small reporting sheet
Keep one row per version with: product reference, creative label, start and end, actual spend, visits where measured, submitted orders, delivered orders, returns and recorded commission. Use matching time windows and retain a separate count of orders still in progress.
If your available reporting cannot reliably separate orders by creative, do not assign them from memory. Keep the campaign result combined or run clearly separated tests with their limitations stated. A traffic label and an official affiliate attribution record are not interchangeable.
04
Wait for meaningful order outcomes
Clicks and messages indicate attention, not earned commission. New, confirmed and shipped orders remain different stages from delivery. Compare versions only after checking which orders have reached recorded outcomes, and flag any later returns when reconciling the test.
For financial review, use recorded delivered commission minus the promotion and content expenses you actually paid. Do not substitute gross customer order value for seller income. A creative that attracts fewer clicks may still produce more useful orders; neither pattern can be established from clicks alone.
05
Choose the next action, not a permanent winner
Describe what the evidence supports: keep a version for another bounded test, revise a confusing explanation, or stop spending. Small counts can be affected by timing and unresolved orders. Do not turn one short campaign into a claim that an image format always wins for Moroccan buyers.
Change the next variable only after recording the current result. Keep previous product and creative notes so you can compare repeat tests. The aim is a repeatable learning process, not guaranteed profitability or a statistical certainty your data cannot justify.