A client asked if I could drop to one variant per brief because of Entity IDs. Their worry was that Meta would lump all three variants together as one ad, so two of them were wasted work.

My answer was yes and no. Entity IDs do exist, but if you significantly change the hook and bridge, Meta does see them as separate ads. And the hook matters a lot for results.

I told them I can do one per brief if they'd like, but they'll likely get less results. Even the sharpest 8-figure media buyers I know run multiple variants per batch.

That question covers most of Meta ads creative strategy right now. Andromeda reads the ad itself and groups the ones that look and sound alike. So ten caption swaps on one studio clip are one ad to Meta, while a new hook and bridge on that same footage is a real variant.

What Meta Andromeda Creative Strategy Changed for DTC Creative Testing

Years ago, media buyers spent half their week building interest stacks, layering lookalikes and excluding past buyers across dozens of ad sets. Targeting picked who saw the ad, and the video only had to convert whoever showed up. So you could run three or four headline swaps on one video and see which one won.

I haven't seen lookalike success in a long time. It's usually outperformed by broad.

A Meta Andromeda creative strategy works the other way around, because Meta reads the ad first and finds the buyers second. It looks at the frames, the spoken audio, the text overlays and the early engagement, then matches each ad to pockets of buyers across broad. If you upload five edits with the same creator, the same kitchen and the same opening two seconds, Meta gets almost nothing new to work with.

You feel it right away when you run meta ads creative testing inside a consolidated Campaign Budget Optimization setup, which I'll call a CBO from here. Meta compares the new uploads against what's already spending. If the new batch looks and sounds like an ad that's already live, it lands in the same lane as that ad. The winner already has purchase history behind it, so the near-copy either gets ignored or takes a few impressions from its parent, and your reach stays the same.

I've been part of $60M+ in managed ad spend, and the stalled accounts usually look the same. It's one creator in a bedroom cut into fifteen hook trims, and most of them collect pennies. So I'd rather ship five briefs a week with three variants each, where the hook and bridge really change.

Creative Is the Targeting

In almost every broad account I audit, the only targeting left is the country and an age floor. You optimize for purchases and the ad does the rest.

Your opening frame, your presenter and what they say decide who stops scrolling. Say you sell a joint recovery tool or protective gear, and 86% of your Shopify customers are 45 and older. If you put a 22-year-old fitness influencer in front of a ring light, Meta serves the ad to younger gym scrollers. Your CPM looks cheap and the click-through rate looks passable, but checkout stalls because the visual pulled in the wrong buyer.

The same thing happens when a winning angle runs out of people. I audited a US lash extension supplier spending around $3,000 a day, and 30-day frequency had climbed toward 10x across roughly 247,000 people. Every ad pitched veteran lash techs in the same salon vocabulary. A new interest stack would've just paid higher CPMs to hit the same quarter-million people again, so I'd brief new sub-avatars like cosmetology students, academy trainees and salon owners with multiple chairs who buy in bulk. Each new speaker and workplace gives Andromeda something new to match inside the same broad campaign.

That's why facebook ads creative testing works best when the campaign map stays boring and every brief puts a different speaker, setting or format on camera.

Visual Swing vs Copy Tweak

Once a founder accepts that the creative does the targeting, the next problem shows up in ad creative iteration. An ad works, so they ask the editor for ten iterations. Two weeks later they get ten cuts where the first two seconds look identical, the presenter's wearing the same shirt, and the only change is the white caption box going from four words to six.

If you want to know how to iterate winning ads without your new cuts eating the winner, sort every planned edit into this table before the editor touches a timeline.

A visual swing changes what the eye catches in the first frame. If your winner's an outdoor storytime video a customer filmed in a driveway, the swing puts a different buyer in a kitchen, or shows a side by side product demo on a workbench, or opens on a close-up of the problem in motion. You keep the reason the driveway video converted and change the visual container, so Andromeda treats it as a new ad for a new pocket of buyers.

The middle row is what I mean by three variants. The body and the mechanism stay the same, and you test three spoken openings at the same awareness stage. Take a hypothetical hip and joint chew for senior dogs, where the proven story is an older dog hesitating at the porch steps. All three hooks lead into the same footage of the dog's routine.

1. "My ten-year-old Golden stopped climbing our three back-porch steps."
2. "If your senior dog pauses at the bottom of the stairs, watch this."
3. "I thought my dog was just getting old until he stopped at the porch steps."

Each hook uses concrete nouns a buyer catches once on a phone speaker, then hands off to the same body and proof.

Skip the third row. Swapping caption colors, nudging a text box two pixels higher or trimming half a second off the end card doesn't teach you anything about the buyer, and it makes your own ads bid against each other. I show how I lay out three-hook briefs, clip sources and visual worlds for editors in my post on briefing visual iterations.

Seven Days in a Broad CBO: How I Read Spend Share

Even with real concepts going in, most founders misread the test inside the first 48 hours.

In a consolidated broad CBO, spend share is the test. When the budget piles onto one or two ads, that's the auction telling you it tried the batch on live impressions and picked a winner.

I give every new launch seven full days, never three, because ROAS on day one and day two is noise. One $110 order on $20 of spend shows a 5.5x ROAS in Ads Manager and it won't repeat. On the flip side, an ad at $170 of spend with one high-AOV order can pick up two more orders over days four to six once Meta settles on the right buyers.

At day seven the rule is blunt. If a new ad got under 10% of the campaign's spend across those seven days, it lost. A lot of founders get on a call convinced Meta made a mistake and skipped their new ads. What happened is the auction gave those ads early impressions, saw weak scroll-stop, soft clicks or poor purchase response next to the current champion, and moved the money to where buyers converted. I go through that and how to fix the creative in my guide on CBO ads not getting spend.

Here's the decision table I use at the end of a seven-day cycle.

The scoreboard behind that table is Triple Whale on 7-day click, or Meta purchases checked against Shopify new-customer orders. I never cut creative off a 1-day view window, because it over-credits retargeting impressions and hides what's bringing in cold buyers. And platform ROAS only points you in a direction, since refunds, product cost, tariffs and subscription math decide whether a 2.5x ROAS makes money on the first order.

Before You Pause a Top Spender

The bottom two rows of that table are the ones that save accounts. One of the costliest mistakes in meta ads creative testing is pausing your highest spender on a short dip before a replacement exists.

I audited a clinical ketone brain and migraine supplement brand where customer acquisition cost went from $150 to $400 over five weeks. The team thought broad had stopped working and their messaging had worn out. I sorted Ads Manager by 90-day spend and checked the change log, and the cause was right there. Five weeks earlier, someone on the team had paused one hero ad carrying roughly 41% of account spend after a brief ROAS wobble. On top of that, the campaigns were optimizing on an unmapped custom conversion event, so delivery was chasing cheap clicks.

With that 41% ad gone, Meta pushed the budget into weaker ads and CPA went up more than two and a half times while the team blamed the market.

Before you pause a high spender that looks soft on 7-day click, run three checks in order.

1. Check the 14-day and 30-day attribution windows. Higher-consideration supplements, footwear and ergonomic hardware often take one to three weeks from first click to checkout.
2. Check blended store results and new-customer CAC. If Shopify new-customer revenue held steady while Meta's attribution lagged, pausing the ad cuts real sales you just aren't crediting.
3. Check mid-funnel feeder lift. A lot of story ads introduce the problem, take the early clicks and hand the buyer to an offer static that closes four days later. If you pause the feeder, the static drops a week later.

Even when an aging hero ad softens on all three over 14 to 30 days, I keep it live and feed new visual swings into the CBO. A challenger has to take majority spend on its own before the old one goes off. There's a longer checklist in my breakdown on when to pause Meta ads.

Where New Concepts Go

I keep one consolidated broad CBO per major market or product line, so the purchase data builds up in one place.

Winners stay in their existing ad sets. Don't duplicate a live champion out of its ad set just to tidy up the naming, because moving it resets the real history tied to that ad set.

When the next batch comes back from your editor, put it in a new ad set inside that same CBO. Give the ad set a date and a batch label, keep targeting open with only your country and age floor, and let the new concepts compete against the champion for seven days. I'd launch three to five distinct concepts, or one concept with three sharp hook variants, so you're not flooding one ad set with twenty half-baked cuts. If you're sizing a launch against your daily budget, I cover that in my guide on how many ads to test in a CBO.

From there the week repeats. Review 7-day spend share, drop anything under 10%, brief new visual swings off what's winning, and ship the next ad set into the same CBO.

Three Structures That Break the Test

When an account comes to me for an audit, the problem usually traces back to a setup that stops Meta from testing the creative cleanly. These three break the test faster than anything else.

1. Fragmented ABO Ad Sets

I audited a senior mobility equipment founder whose business had scaled fine on a consolidated setup, then handed the account to an outside agency. The agency split it into separate Ad Set Budget Optimization ad sets, the ABO setup, and lost money for 90 straight days while defending the complexity. New creative for their 45+ buyers stalled the whole time.

When you lock small daily budgets into dozens of ABO ad sets, you're paying Meta to force impressions onto weak ads a CBO would've dropped in four days. And none of those ad sets get enough purchases in a week to stabilize.

The fix was collapsing the ABOs back into one consolidated CBO, hiring a dedicated editor and shipping joint-relief demonstrations for those 45+ buyers.

2. Using a CBO to Split-Test Landing Pages

A multi-country oral anti-sweat supplement brand on Shopify came to me spending 2,000 SEK per day while running a six-way landing page split test inside one CBO. That can't work, because a CBO picks the winning ad and doesn't split traffic evenly across six URLs. Whichever ad got cheaper early engagement took most of the impressions, and the other pages got tiny, lopsided traffic that told the team nothing about conversion rate.

To be honest, the creative mix was the bigger risk. 51.4% of spend sat on uncredited ripped TikTok clips, which can get the account banned for good. Ripped content should be 3 to 5% of what you run at most and heading to zero, it only works short term. A claymation surgery ad had also burned 15,509.86 SEK for 20 purchases at a 0.73 ROAS.

So I'd move the URL tests onto Shopify where traffic splits cleanly, pause the ripped clips right away, and shoot owned creator POVs around real sweat moments like handshakes and dates. Then the proven markets like Canada, Norway and Ireland go into one clean CBO.

3. Cost Caps, Bid Caps, and Manual "Rescues"

I don't usually run cost caps or bid caps, and I don't set ad set spend minimums in a testing CBO either. They stop Andromeda from testing concepts cleanly across broad.

Pulling a 7-day loser out of the CBO into its own ABO campaign to force spend onto it does the same damage, because you're overriding the auction's verdict with your own budget.

If your account's stuck in fragmented tests or copy-only iterations and you'd like an audit or five briefs a week with three variants each, book a discovery call or see live creative strategy work.