What is a good ad testing strategy for an online store?
The 7 tests, in the order that pays
Most stores test what is easy to change (a headline, a button) and skip what changes the result (the offer, the first three seconds). The order below starts with the tests that most often double or halve a result and ends with the fine-tuning.
| Test | Change | Keep the same | Judge on |
|---|---|---|---|
| 1. Offer | Discount vs bundle vs free shipping | Creative and audience | Matched orders and profit per order |
| 2. Hook | The first 3 seconds or the first line | The rest of the ad | Click rate, then matched orders |
| 3. Format | Video vs image vs carousel | Message and offer | Matched orders per 1,000 spent |
| 4. Audience | Broad vs interests vs lookalike | The ad itself | Matched orders and new customers |
| 5. Landing page | Product page vs collection vs landing page | The ad | Matched orders per click |
| 6. Copy | Benefit vs problem vs proof | Image and offer | Matched orders |
| 7. Call to action | "Shop now" vs "See the range" and similar | Everything else | Click rate |
1. Test the offer first
The same ad with "20% off" and with "free gift over $50" can sell at very different rates, and the profit per order differs too. Judge an offer test on profit, not only on ROAS: a discount that wins on ROAS can still lose money.
2. Then the hook
People decide in the first seconds whether to keep watching. Keep the rest of the video identical and swap only the opening: a problem, a result, a face, a product close-up. Use click rate to drop the clear losers early, then confirm the survivors on orders.
3. Format
Video, still image and carousel reach people in different placements. Run the same message in each format. Compare orders per 1,000 spent, because formats differ in cost per view.
4. Audience
With Meta's broad and Advantage+ audiences doing much of the targeting, audience tests matter less than they did. Still worth one test: broad vs your best lookalike. Look at how many of the orders came from new customers. A retargeting-heavy audience often looks better on ROAS while bringing in fewer new buyers.
5. Landing page
Sending the click to the product page, a collection or a dedicated landing page changes how many visitors buy. This one is judged in your store, not in Ads Manager: orders per click.
6. Copy, and 7. Call to action
Useful, but rarely decisive. Run these once the bigger tests are settled, and expect small differences that need more orders to prove.
How to run a fair test on Meta
Use the A/B test option in Ads Manager when you can. It splits the audience into random groups that do not overlap, so the same person does not see both versions (Meta Business Help Center). Two ad sets launched side by side without that split compete for the same people, and the result partly reflects who Meta showed each version to.
Give each version enough results to settle. Meta says an ad set leaves its learning phase after about 50 optimization events within 7 days (Meta Business Help Center). With fewer purchases than that, call tests only on big differences, or test on add to carts first.
Pick the winner on real orders
By default, Ads Manager credits a sale to an ad when the buyer clicked within 7 days or saw the ad within 1 day. View credit can make the version that was shown more often look like the winner, even if it did not bring more buyers.
| Stores | Platform's sales vs matched sales (median) | Platform reported more in | |
|---|---|---|---|
| Meta | 91 | 1.12x | 68% of stores |
| Google Ads | 65 | 0.81x | 23% of stores |
Stores with at least about $1,000 of spend on the channel in the 4 weeks; stores where one side saw under half of the other are left out as broken tracking. No store is named.
The gap is not the same for every ad. That is the point: if version A gets more of its sales from views and version B from clicks, Ads Manager can rank them the wrong way round. Before you move budget, check both versions against orders matched in your store. In TrueROAS that is the matched sales column per ad; without a tool, compare orders with the ad's UTM tags in Shopify or GA4.
When not to test
- Under about 20 orders a week from ads: small tests will not settle. Test big changes only, like a new offer or a new angle.
- During a sale or a launch: the result says more about the week than about the ad.
- When tracking is broken: if Meta and your store disagree by more than 2x, fix tracking first. A test on broken numbers picks a random winner.
Keep a test log
Write down each test: the date, what changed, the result on matched orders, and what you did next. After a few months the log tells you what works for your store, and stops you testing the same idea twice.

