Meta Ads for eCommerce Brands: A Creative Testing System That Scales

Monday, 8:15 a.m. The founder opens Ads Manager before coffee. Blended ROAS slid all weekend. The video that carried March, the one where the customer unboxes the product in her car, is sitting at a frequency of 5 and its cost per purchase has doubled.

Someone on the team says “we need new creative.” Three new videos go into the winning ad set by lunch. The ad set drops back into learning, spend lurches toward one of the new ads, and by Thursday nobody can say whether the new videos are bad or the account is just recovering from being poked.

That week is the problem this post solves. Most eCommerce brands don’t have a creative problem so much as a creative system problem: no fixed place to test, no rule for when a test is over, and no agreed way to move a winner into the campaign that scales.

Below is the system we run for DTC and Shopify brands on Meta, from campaign structure and test budgets to the metrics that tell you why an ad failed, plus the scaling rules, a weekly calendar and the brief template our team uses.

What changed in Meta’s delivery: Andromeda and creative as targeting

In December 2024 Meta published Meta’s engineering post on Andromeda, the system that picks which ads are even considered for a person before the auction ranks them. Meta said it delivered a 10,000x increase in retrieval model capacity, a 6% recall improvement and an 8% ads quality improvement on selected segments. Per Jon Loomer’s breakdown of Andromeda, retrieval narrows tens of millions of ads down to a few thousand candidates for each person.

The practical result is that your ad itself does more of the targeting than your audience settings do. Meta’s own guidance has shifted to calling creative diversification the best lever available, and the old recommendation of no more than six ads per ad set quietly disappeared, again per Loomer.

What “diversification” means matters. Five versions of the same video with different captions look like one ad to a retrieval system. Five genuinely different concepts (a founder story, a problem-and-fix demo, a customer testimonial, a comparison against the thing people use now, a static with a price and an offer) give Meta five different shots at five different kinds of buyer. Jon Loomer’s post on creative diversification makes the same point: variations that aren’t distinctly different are unlikely to change performance much.

So the unit of testing in this system is the concept. Hooks, edits and formats are iterations inside a concept, and we test those after a concept has earned it.

Advantage+ sales campaigns: what the name means now

If you learned Meta in 2023 or 2024, you learned Advantage+ shopping campaigns (ASC). That product has been folded into what Meta now calls Advantage+ sales campaigns. Per Meta’s developer announcement from June 2025, a sales campaign counts as “Advantage+ on” when it uses three things: Advantage+ campaign budget, Advantage+ audience in at least one ad set, and Advantage+ placements in all ad sets. Creating or updating the old ASC campaign type through the API ended across all versions with v25.0 in the first quarter of 2026, and the Existing Customer Budget Cap was deprecated along with it.

A few practical differences, as summarized by Jon Loomer: ad sets are back, there’s no fixed limit on how many you create, and each ad set holds up to 50 ads. Old ASC campaigns held up to 150 ads in one place. You can add audience suggestions and custom audience exclusions, which ASC never handled well.

In Ads Manager, the path is short. Click Create, choose the Sales objective, and keep the campaign budget, audience and placement settings on Advantage+ if you want the automated version. Ads Manager shows whether the campaign is running as Advantage+. Turn any of those off and you have a manual sales campaign, which is still useful for structured tests.

For most Shopify brands spending $20,000 to $300,000 a month, we want exactly one Advantage+ sales campaign doing the heavy lifting. Splitting budget across five similar campaigns just splits the learning.

The account structure that keeps tests readable

Testing only works if new ads have somewhere to prove themselves without disturbing the ads that pay the bills. This is the three-part structure we use:

Those percentages come from accounts we’ve run. They aren’t a Meta rule. A brand launching a new product line might push testing to 30% for a month, and a brand with one hero SKU and a deep library of proven ads might run testing at 10%.

There’s a budget ceiling to respect, too. In an October 2026 peak-season post, Meta said ad sets need roughly 50 conversions a week to learn, and that ad sets under that threshold ran about 28% more expensive per purchase. Meta published no method for that figure, as PPC Land’s report on Meta’s October 2026 peak-season post pointed out, but the same post gave a useful formula: weekly budget divided by (cost per result × 50) is about how many ad sets your budget can support. At $20,000 a week and a $40 cost per purchase, that’s 10 ad sets. At $100 per purchase, it’s four. Most accounts we inherit have twice as many ad sets as their budget can feed.

How many creatives, and how to structure the test

The right number of new ads per week depends on your test budget and your cost per purchase. The rule we use: each concept needs enough spend to buy two to three purchases at your target CPA before you judge it. Fewer than that and you’re reading noise.

An illustrative example. A brand spends $60,000 a month, puts 20% into testing ($12,000, or about $2,800 a week) and has a $45 target CPA. At roughly $135 per concept to reach a judgment, that budget supports about 20 ads a week. In practice we’d run four to six concepts with three or four executions each, rather than 20 unrelated ads.

There are two clean ways to run the test.

• A separate testing campaign. A manual sales campaign, ad set budgets, one ad set per concept, broad targeting with your existing customers excluded if you want a cleaner read on new buyers. Winners get moved into the scaling campaign using the existing post ID, so the likes and comments come with them.
• Meta’s Creative Testing tool. Inside an ad’s settings there’s now a Creative Testing section with a Set up test button. Per Jon Loomer’s walkthrough of the Creative Testing tool, you can test two to five ads, Meta recommends setting the test budget at no more than 20% of the campaign or ad set budget, the default length is seven days, it requires the Highest Volume bid strategy, and each person sees only one of the test ads. Results show up under Experiments.

We use the built-in tool for head-to-head questions inside a concept (hook A versus hook B) and the separate campaign for new concepts. Each test gets one written hypothesis before launch, such as “A 30-second founder story will beat our UGC testimonial on new-customer CPA.” If you can’t write the sentence, you’re not testing anything.

Reading results: the metrics and the formulas

ROAS tells you whether an ad is working. It doesn’t tell you why. The creative metrics do, and most of them are simple ratios you can build as custom columns in Ads Manager. The formulas below follow the definitions used by Runneth (formerly Motion), plus the business metrics we hold every account to.

Read them in order, like a funnel. Each stage points to a different fix:

A low hook rate means you rewrite or reshoot the first three seconds and leave the rest alone. A strong hook with weak hold usually means the video promised one thing and then wandered off into brand talk. Good hold and weak CTR is an offer or call-to-action problem. Strong CTR with a weak conversion rate is almost always the product page, the price or shipping cost, not the ad.

We don’t publish universal “good” numbers for these, because they swing with category, price point and placement. Benchmark against your own account’s last 90 days instead. Your top quartile of ads is the bar.

And keep MER above the ad-level numbers. Meta’s attributed ROAS can rise while store revenue stays flat, especially when spend drifts toward people who were going to buy anyway. If MER and new-customer CPA hold while you scale, the platform numbers are telling the truth. Our guide to marketing metrics covers how to set targets from your margins.

Kill, iterate or graduate: the decision rules

Every test needs a pre-agreed exit. These are the defaults we start with, adjusted per account:

• Kill when an ad has spent two to three times your target CPA with no purchase, or when hook rate sits in your account’s bottom quartile after a few thousand impressions.
• Iterate when one stage is strong and another is weak. A great hook with a bad middle gets a re-edit. A great video with a weak CTR gets a new offer or end card.
• Graduate when CPA is at or under target across at least five to ten purchases (more for low-price products), and the new-customer share looks healthy.

Graduated ads go into the scaling campaign using the same post ID. Don’t duplicate the winning ad set to “scale” it. And don’t turn off the old winner the same day; let Meta shift spend between them.

UGC vs. studio creative, and where creators fit

We get asked this every week. The honest answer is that the format matters less than the concept, and most winning accounts run both.

UGC-style video (a real person, a phone camera, talking straight to the viewer) tends to win on hook rate and on cold audiences, because it looks like the rest of the feed. Studio work wins when the product is visual (apparel, jewelry, home goods) or when price and positioning need it to look expensive. Static images are underrated. A clean product shot with a price, a review quote and an offer is cheap to make and often beats video on CPA for repeat buyers.

Creators add something else: their own audience and their own handle on the ad. On our Vital Proteins launch, we seeded product to more than 1,200 wellness, fitness and food creators, then ran the best posts as whitelisted creator ads (Spark and partnership ads). Those creator ads returned 6.2x ROAS. The lesson we took from it wasn’t “creators always work.” It was that a large pool of creator content gives the testing system far more distinct concepts to choose from than any single shoot could.

Two practical notes. Get usage rights in writing before you put a creator’s video behind paid spend, and keep the disclosure intact; our post on FTC rules for influencer partnerships covers the details. And our UGC vs. influencers guide goes deeper on cost and sourcing.

Catalog ads: the quiet workhorse

Advantage+ catalog ads (Meta’s name for what used to be dynamic ads) pull products from your catalog and show people the items they viewed, or items Meta predicts they’ll want. Since May 2024 they can include product video as well as images, per Social Media Today.

For most Shopify stores, catalog ads do two jobs. They retarget product viewers and cart abandoners (the classic use), and they prospect broadly when the catalog is large enough for Meta to find matches. A store with 12 products gets less from broad catalog prospecting than a store with 1,200.

Catalog ads are only as good as the feed. Connect the catalog through Shopify’s Facebook & Instagram channel, check Commerce Manager for rejected items weekly, and write titles a human would want to read. Add frames or overlays with a review score or free-shipping line if your category is crowded. The feed work overlaps heavily with Google Shopping, so if you’ve done our Google Shopping setup for Shopify, most of it carries over. And our retargeting explainer covers the basics if this is new.

Budget scaling rules that don’t break the account

Meta doesn’t publish a percentage for how big a budget change resets learning. It gives examples instead: going from $100 to $101 isn’t likely to put an ad set back into learning, and going from $100 to $1,000 may, per Meta’s own examples, collected by Jon Loomer. Everything between those is judgment.

Ours, from accounts we’ve managed:

1. Raise the scaling campaign budget 15% to 20% at a time, no more often than every two to three days, while CPA and MER hold.
2. Scale on the business numbers. If MER drops more than about 10% after a raise and stays there for a week, step back to the last budget that worked.
3. Batch your edits. One change on Tuesday is better than four changes spread across the week, because every edit costs learning time.
4. Feed the scaling campaign new graduates every week. Budget increases without new creative are how frequency climbs and CPA follows.
5. Plan peak season early. Meta’s own peak-season advice was to pick your biggest campaign, set it up and leave it alone through the run-up.

Houston brands get one extra seasonal wrinkle. When a hurricane or a hard freeze hits the Gulf Coast, local attention and delivery promises both change for a week or two. If a meaningful share of your customers are in Texas, have a pause-or-adjust plan ready rather than discovering it in the CPA chart.

The weekly testing calendar and the brief template

A system is mostly a calendar people actually follow. This is the week we run for eCommerce clients:

Every new concept starts with a one-page brief. Ours has these fields:

The “why we think it will win” line is the one people skip, and it’s the one that makes the Monday review useful. When a test loses, you learn whether the idea was wrong or the execution was.

Mistakes we see in eCommerce Meta accounts

• Adding new ads to the winning ad set. It resets learning on the thing that’s working. Test somewhere else.
• Testing tiny variations. Changing the button color isn’t a test Andromeda can tell apart. Test concepts.
• Judging on day two. Meta reallocates spend in the first days. Set the test length before launch and stick to it.
• Ignoring the product page. When CTR is strong and purchases aren’t, the ad has done its job. Look at the page, price and shipping.
• Running everything off platform ROAS. Without MER and new-customer CPA, it’s easy to scale spend into people who would have bought anyway.
• No creative pipeline. Brands that brief on Wednesday because the numbers dropped on Monday are always two weeks behind.

A first month on this system looks like this: week one, rebuild the structure and set up custom metric columns; week two, launch the first four to six concepts; weeks three and four, run the calendar and graduate the first winners. By day 30 you should know which two or three concepts your customers respond to, which is worth more than any single winning ad.

We run this for DTC and Shopify brands through our paid social team, and our eCommerce page shows the rest of the playbook. If you’d like a senior strategist to look at your account and build a testing plan around your margins and catalog, get in touch.

Frequently asked questions

How many creatives should I test on Meta each week?

It depends on your test budget and target CPA. Give each concept enough spend to buy two to three purchases at your target CPA before judging it. A brand spending $60,000 a month with 20% in testing and a $45 CPA can support about 20 new ads a week, which in practice means four to six distinct concepts with a few executions each. Concepts matter more than raw ad count.

What is the difference between Advantage+ shopping and Advantage+ sales campaigns?

Advantage+ shopping campaigns were folded into Advantage+ sales campaigns. A sales campaign runs as Advantage+ when it uses Advantage+ campaign budget, Advantage+ audience in at least one ad set and Advantage+ placements in all ad sets. Ad sets are back, each holding up to 50 ads, the Existing Customer Budget Cap was deprecated, and the old campaign type could no longer be created through the API after early 2026.

What is a good hook rate for Facebook and Instagram video ads?

Hook rate is 3-second video plays divided by impressions. Good varies a lot by category, price point and placement, so we benchmark against the account’s own last 90 days rather than a universal number. Treat your top quartile of ads as the bar. If a new ad sits in your bottom quartile after a few thousand impressions, rework the first three seconds before changing anything else.

How fast can I increase my Meta ads budget?

Meta doesn’t publish a threshold. Its examples say a $100 to $101 change is unlikely to restart learning while $100 to $1,000 may. In accounts we run, we raise budgets 15% to 20% at a time, no more than every two to three days, while CPA and MER hold. If MER drops about 10% and stays there for a week, we step back.

What is MER and why does it matter more than ROAS?

MER, or marketing efficiency ratio, is total store revenue divided by total marketing spend. It’s sometimes called blended ROAS. Platform ROAS only counts sales Meta attributes to itself, which can rise even while the business stays flat. MER shows whether total revenue is growing with spend. Pair it with new-customer CPA so you know the spend is buying new customers rather than repeat orders.

Written by Sean Melton

Sean Melton is the founder of First Byte, a digital marketing agency in The Woodlands, TX. A former global CMO of a $250M software company, he has spent his career inside the businesses he markets. His team runs paid social, search, CTV and creative for brands like Equinox, Kroger and Viceroy, and shares what works here.