Services
Image and content testing on Amazon, one visual change at a time
We test the pictures and modules shoppers see first, using Amazon's experiment tools where your account is eligible and a sequenced comparison where it is not.
- Main image tests planned before launch
- Experiment tools used where eligible
- Every test logged with its stop rule
Why visual changes are a good place to test
Pictures are one of the few parts of the page that shoppers judge before reading. A change to the main image or the first module can alter that judgment quickly, which makes visual tests a clear way to learn what the page needs. The catch is that many visual changes tend to happen together in practice, so a test has to change one thing and nothing else.
On Amazon, controlled experiments are available to brands whose accounts meet the platform's eligibility rules, and the supported content changes over time. Where a controlled tool is not available, we use a sequenced comparison and state that the evidence is weaker. The choice of method is part of the plan, not a detail added later.
How a visual test is designed
A test that changes two things teaches nothing about either.
-
Pick the first image
We choose the image the funnel points to as weakest, usually the main image or the first module.
-
Write one variant
The variant changes one thing, such as the angle, background, scale or caption, and the rest of the page stays the same.
-
Set the stop rule
The minimum data, the window and the change needed to call a result are written down before the test begins.
-
Log the outcome
The result, outside events such as stock or advertising changes, and the decision go into a log that outlives the test.
Conditions that spoil a visual test
We check these before we start and while the test runs.
-
Stock problems
A product that runs out during a test makes the comparison meaningless, so we pause or flag the test.
-
Large discounts
A sudden price change draws shoppers for reasons unrelated to the picture, so it is noted in the log.
-
Advertising swings
Campaigns switched on or off change the traffic mix, which can change conversion on its own.
-
Too little traffic
A low-traffic ASIN may never reach a readable sample, so we say so before the test starts.
Experiment tool or sequenced comparison
| Experiment tool | Sequenced comparison | |
|---|---|---|
| Traffic split | Handled by the tool | Whole traffic, in turn |
| Comparison | Built into the tool | Compared by hand |
| Eligibility | Depends on your account | Open to any listing |
| Strength | Stronger where available | Weaker, stated in plan |
Risks, limits and what we cannot promise
Visual tests can mislead. Shoppers arrive for different reasons on different days, and a variant that wins in one week can lose in the next. A clear result on the main image does not mean the listing will convert better overall, because the rest of the page still shapes the decision. A test can also show a real effect that is small, which may not justify the cost of a new shoot.
We cannot promise that a test will win, the size of any lift, or a date by which a result will be clear. We can promise a design you can inspect, a stop rule written before launch and a log of what happened.
Testing questions
Which image should we test first?
Usually the main image, because every shopper who lands on the page sees it. If the main image is already strong and the funnel points elsewhere, the first module may be the better choice. We decide that during the review using the data you share.
Can we test a new shoot against the current images?
Yes, but a new shoot is expensive, so we test the cheapest change first. A reorder or a crop can answer the question before a full shoot is booked. If the test shows a clear need, the brief for the shoot is much sharper.
What if no variant wins?
A flat result tells us the picture is probably not the main barrier on this ASIN. The next test moves to the funnel step the data points to, and the log records why the image idea was set aside for now.
Do you run tests without our data?
We can design a test from the public listing, but it will be harder to read without sales and traffic figures. We say in the plan how much weight the result can carry if you choose not to share your numbers.
Tell us which image you think is weak
Use the request form or message us on WhatsApp with the ASIN and the image you suspect. If you are not sure, say so and we will suggest one from the funnel.
What to send
- ASIN or store link
- The image or module you suspect
- Whether your account has an experiment tool available
- Recent unit sessions, if you share them
Choose the image worth testing first
Request a page review and we will name the first visual change to test, with the reason.