One image change. 10% more orders. $6,686 in extra sales during the test window
By Daniela Anavitarte Bolzmann ⚡ · August 3, 2026 · Curated by George's Blog
One image change. 10% more orders. $6,686 in extra sales during the test window alone.
Here's how a main image test like this actually runs:
1. Ideate wide with AI
→ We generate lots of different variants first
→ Volume at this stage costs almost nothing
2. Narrow to a few strong concepts
→ Most variants die here, on purpose
→ Only the concepts with a real strategy behind them move on
3. Pre-test the concepts in PickFu
→ Real shoppers vote before anything touches the listing
→ The weak ideas die for a few dollars, not lost sales
4. Iterate on what the votes tell us
→ The feedback shapes the next round of concepts
→ We keep refining until one option clearly leads
5. Validate with an Amazon experiment
→ The leading option goes head to head with the original on live traffic
→ This one: 72% probability of outperforming, 10% more orders, $6,686 more in sales during the experiment period
Where things go wrong:
1. Most brands might never be presented with any options to choose from.
2. Most probably don't have any data to help them choose, so they tend to pick the option "they think" looks the best, which might not be the image that actually performs the best.
That's why pre-testing data matters so much: you want it in hand before you decide what goes live on Amazon, or what goes into a Manage Your Experiments test.
This is the proof.
SAVE & Share with your team. Teams should be running tests like this EVERY MONTH, not once a year.