Before you scale templated creative, you need a clean answer to one question: does the templated image beat the raw catalog image, with everything else kept equal?
This guide walks you through setting up your first catalog campaign on Meta as a controlled test. The test runs for 14 days.
Before you start
You'll need:
A Meta catalog claimed in Hunch Light. See Claiming an existing Meta catalog in Hunch Light.
At least one template created in Hunch Studio and synced to that catalog. See Dynamic creative templates and Template collections in Hunch Light.
A Meta Pixel and Conversions API (CAPI) sending ViewContent, AddToCart and Purchase events.
Access to Meta Ads Manager and Events Manager for the ad account you'll use.
Enough budget to run two test arms for 14 days. Step 5 shows how to calculate it.
Step 1: Check your tracking and catalog
Most "templates don't work" results are actually tracking or catalog problems. Check these before you build anything.
Check the catalog match rate. In Events Manager, open your catalog's Events tab. The
content_idssent by your Pixel and CAPI should match catalog product IDs at 90% or higher. If they don't, Meta can't tie events to products and delivery is effectively guessing.Check event match quality. Your Purchase event should score 6 or higher.
Clean your product set. Remove out-of-stock items, products with broken images and any SKUs you don't want to advertise.
Make sure no other feed is syncing to the catalog. In Meta Commerce Manager, open the catalog's Data sources and check that Hunch Light is the only feed updating it. Another feed can overwrite product data and images during the test.
Step 2: QA your template across the whole catalog
A template that looks great on your hero product but breaks on 15% of the catalog will lose the test. That's a QA failure, not a template failure.
Use the preview tool in Hunch to check the template on the difficult cases, not just your best products. See Dynamic templates preview & review.
Very long product titles
Products with no sale price, or with a sale price
Products with transparent PNG images
Images in unusual aspect ratios
Both 1:1 (Feed) and 9:16 (Stories and Reels) renders
For 9:16, make sure logos, prices and badges stay out of the top and bottom areas covered by the Stories and Reels interface.
Step 3: Set up two arms that differ only in the image
Your test has two arms:
Control: raw catalog images
Test: templated images from Hunch Light
Hunch Light syncs your templated images to the same claimed Meta catalog, so both arms run from one catalog and you don't need a second catalog or feed. Use the same product set in both arms.
Keep everything else identical:
Setting | Must be the same in both arms |
Catalog and product set | Exactly the same |
Primary text, headline, CTA | Identical |
Audience | Identical |
Placements | Identical |
Optimization event | Identical |
Attribution window | Identical (for example, 7-day click, 1-day view) |
Important: Turn off Advantage+ creative enhancements on both arms. This includes automatic overlays, brightness and contrast adjustments, and dynamic media options. If Meta modifies your images on top of the template, the test is no longer clean.
Tip: Test template vs. raw image first, not template A vs. template B. You need a baseline before any template-to-template comparison means anything.
Step 4: Use Meta's A/B test, not two ads in one ad set
Don't put both ads in the same ad set. The delivery system picks a favorite within a day or two, often based on early cheap clicks, and stops serving the other ad. You'd learn what the algorithm preferred on thin data, not which creative performs better.
Instead:
In Ads Manager, create a Sales campaign using your claimed catalog.
Set up the control ad set with raw catalog images.
Use A/B Test (or Experiments) to create the test arm, choosing Creative as the variable, and use your Hunch template collection in the test ad. To add Hunch ads to an existing campaign, see Publishing catalog ads with Connect Ads in Hunch Light.
Meta splits the audience so the two arms don't overlap, and reports a confidence level at the end.
Step 5: Set a budget that gets each arm out of learning
Each arm needs about 50 optimization events in 7 days to stabilize. Calculate the daily budget per arm like this:
Target CPA × 50 ÷ 7 = daily budget per arm
Example: with a target CPA of €20, that's €20 × 50 ÷ 7 ≈ €140 per day, per arm.
If you can't afford that while optimizing for Purchase, optimize for AddToCart instead and treat purchase CPA as a secondary metric. An underfunded arm that never leaves learning gives you random results that look like insights.
Step 6: Choose one audience
Pick one audience type for the test. Don't mix both.
Broad / Advantage+ audience (prospecting). Recommended. This is where templates usually matter most, because cold users don't know your brand and the price badge, logo and benefit callout do real persuasion work. It's also where you'll scale.
Retargeting (viewed or added to cart in the last 14 days). Faster and cheaper signal, but warm users convert mostly on product recognition, so the template effect tends to be smaller. Use this if your budget is tight.
Step 7: Launch and leave it alone
Run the test for 14 days.
Don't edit budgets, audiences or creative while it runs. Edits reset learning.
Ignore days 1–3. Early results are noise.
Step 8: Read the results
Use CPA or ROAS as the primary metric. Use the other metrics to understand why one arm won or lost.
What you see | What it means |
CTR up, conversion rate down | The template grabs attention but over-promises. A loud discount badge, for example, can attract clicks from people who don't buy. |
CPM lower | Meta's quality and engagement signals favor the creative. A good sign, even if CPA hasn't moved yet. |
CTR flat | The template isn't stopping the scroll. The design isn't different enough from a raw image. |
Calling a winner: look for a CPA difference of 10–15% or more at 80%+ confidence in Meta's test results. Anything less is a tie.
A tie is still useful. It tells you this template isn't adding value yet, so iterate on the design rather than giving up on templates.
Step 9: Iterate
Once you have a baseline winner, test template vs. template, changing one element at a time:
Price treatment
Background
Badge vs. no badge
Brand frame
Run each new test the same way: same products, same copy, enhancements off, A/B test, enough budget, 14 days.
Troubleshooting
One arm barely spent. You probably ran both ads in one ad set instead of using A/B Test. Set it up again as described in Step 4.
Both arms are stuck in learning. Budget is too low for your optimization event. Raise the budget or switch to AddToCart (Step 5).
Templated images look different in Ads Manager than in Hunch. Advantage+ creative enhancements are turned on. Turn them off on both arms (Step 3).
Templated images aren't showing, or products show the wrong images. Check that no other feed is syncing to the same Meta catalog (Step 1). Another feed can overwrite the images Hunch Light syncs. Also check that syncing isn't paused. See Why is syncing paused in Hunch Light?
Results swing a lot day to day. That's normal early on. Don't judge before day 7, and don't decide before day 14.
Related articles
Questions? Contact us at [email protected].