How to A/B Test Product Images for Your Salla Store

How do you A/B test product images on a Salla store?

Salla does not ship a built-in image split-testing tool, so merchants test the practical way: run a sequential swap test (change the hero image for a fixed window and compare conversion in your Salla analytics), run ad-level A/B tests on Snapchat, TikTok, or Meta where true randomized splits exist, or watch category-page click-through after an image change. All three give you an honest directional read on which image sells — without any extra software.

Why the hero image is worth testing

A shopper on a category page chooses between products they have never touched, based almost entirely on one thumbnail. Multiple angles, on-model shots, and a clean hero measurably influence whether a visitor clicks and whether they buy — but which image works best for your store is not something any blog post can tell you. It depends on your product, your audience, and your price point. That is exactly the question a test answers, with your own data.

The competitive context makes it worth the effort: 70,000+ Saudi stores run on Salla, many competing inside the same categories. An image advantage compounds — the store that tests learns something every month; the store that guesses stays where it is.

Method 1: The sequential swap test

The workhorse method, and the one the step-by-step below walks through. You are not splitting traffic — you are splitting time.

• Pick one product with steady traffic — a best-seller, not a long-tail item.
• Record a baseline: visits, add-to-carts, and orders for the past 14 days from your Salla analytics.
• Swap the hero image for the challenger. Change nothing else — price, title, description, and the rest of the gallery stay frozen.
• Run the challenger for the same 14 days, then compare conversion rates, not raw orders. Traffic moves week to week; rates are steadier.

Sequential testing has a known weakness: the two windows are different weeks, so a promotion, a payday, or a season can pollute the comparison. Mitigate it by avoiding campaign weeks entirely, and never test across Ramadan, Eid, or White Friday boundaries — demand shifts in those seasons will drown any image signal.

Method 2: Ad-level A/B on Snapchat, TikTok, and Meta

Ad platforms give you what a storefront cannot: a true randomized split. Upload two image variants of the same product as two ads in one ad set, with a shared budget and audience, and the platform serves them head to head.

• Read CTR to learn which image stops the scroll.
• Read cost per purchase (CPA) to learn which image actually sells.
• The winner earns a promotion to your store hero. It is still worth confirming with a sequential swap — an ad impression and a product page are different contexts — but in practice the two reads usually agree.

This is the fastest route to a statistically honest answer, because the platform randomizes for you, and you were likely paying for the ads anyway.

Method 3: Category-page click-through checks

Between formal tests, watch which products earn clicks from your category pages relative to how often they appear. A product that gets scrolled past while its neighbors get opened has a thumbnail problem — that image is your next test candidate. This is not a controlled experiment; treat it as a triage list that tells you what to test next, never as a verdict.

What to test first: the priority order

Test the variables with the biggest visual difference first. Subtle changes need enormous traffic to resolve; format changes do not.

Change one variable per test. If you swap the format and the background at once, a win teaches you nothing about why it won.

The honest statistics note

A small store cannot test like a platform with millions of sessions. If a product gets 300 visits a month, a two-week window may hold only a handful of orders in each arm — enough for a directional read, not a p-value. That is fine. Treat testing as stacked evidence: run the swap, confirm with an ad-level split, keep the winner, and retest your top products each season instead of chasing certainty from one small sample. Weeks, not days; direction, not decimals. And be suspicious of anyone promising precise lift percentages from small-store traffic — including any imagery vendor.

A simple eight-week testing calendar

You do not need a testing department — you need a routine. A realistic cycle for a small team:

One cycle per season is enough for most stores. The goal is a store that steadily accumulates knowledge about its own shoppers — not a laboratory.

Producing variants is no longer the bottleneck

The real blocker for image testing was always production. A second hero used to mean another studio session at SAR 4,000–40,000 ($1,000–$10,000) or a freelancer at SAR 1,100–1,900 per product ($300–$500) with a 1–2 week wait — absurd overhead for an experiment that might lose. Generation removes the constraint: from one flat-lay phone photo, Rokon produces the on-model version, the alternate pose, the different model look, and the lifestyle-background variant in about 30 seconds each, and multi-image reference keeps the actual garment — stitching, logo, fabric, silhouette — identical across every variant, so your test compares presentation, not products. On the Pro plan an image costs roughly SAR 2 (~$0.50), which prices a full five-variant test matrix at about SAR 10. Give every generated variant a quick pre-publish fidelity check — neckline, print, logo — so you are always testing images you would be proud to ship.

You do not need special software to learn which images sell on your Salla store — you need a challenger image, a calendar, and 14 patient days. Rokon's free tier includes 5 free generations (150 credits), no credit card, enough to produce your first challenger hero today and let your own store data pick the winner.