Meta ads: Facebook and Instagram

Why creative is now the targeting on Meta

The biggest shift in Meta advertising is that the creative decides who sees the ad. Meta's delivery system reads the response to your ad and finds more people who behave like the ones who responded. That means a video aimed at new parents will find new parents, whether or not you told the platform to target them.

This changes the job. Ten years ago the skill was audience research and layered interest stacks. Now the skill is producing enough distinct creative angles that the system has something to work with. Two ads with the same message tell Meta almost nothing. Six ads with genuinely different hooks, formats and offers tell it a lot.

It also changes where the money should go. Accounts that spend their time rebuilding audiences and their budget on one hero video tend to plateau. Accounts that keep a steady flow of new creative into a simple structure tend to keep improving. We plan production capacity first, then the campaign structure around it.

Broad targeting, custom audiences and lookalikes

Broad targeting now beats narrow stacks in most accounts. Give the system a wide pool and good conversion signal and it will find buyers faster than a hand-built interest list can. Narrow targeting mainly makes sense when there is a genuine constraint, such as a licence area, a niche profession or a legal restriction on who can buy.

Custom audiences still matter, because they are built from things you actually know. Website visitors, customer lists, video viewers, people who engaged with your page or saved a post. These are the raw material for retargeting and for exclusions, and they are more reliable than any interest guess.

Lookalikes have become less critical than they were, since broad targeting with a strong signal often lands in a similar place. Where they still earn their keep is with a high-quality seed list, such as your best customers by lifetime value rather than all purchasers. A one percent lookalike of your top spenders is a different thing to a one percent lookalike of everyone who ever bought.

• Start broad unless there is a real reason not to
• Build custom audiences from behaviour you can verify
• Seed lookalikes with your best customers, not all of them
• Always exclude recent purchasers from prospecting campaigns

Campaign budget optimisation and account structure

Campaign budget optimisation puts the budget at the campaign level and lets Meta move it between ad sets. It usually outperforms manual ad set budgets, because it reacts faster than a human checking in each morning. The trade-off is that it will starve ad sets you wanted to test, so it is a scaling tool more than a testing tool.

The structure that works for most advertisers is simpler than people expect. Fewer campaigns, fewer ad sets, more ads inside them. Consolidating spend means each ad set clears the conversion volume it needs to exit the learning phase, which is roughly fifty conversions a week. Split the same budget across eight ad sets and none of them learn properly.

Where you do want separation, base it on a decision you need to control: cold versus retargeting, one country versus another, or a product line with a different margin. Do not split for the sake of tidiness. Every extra ad set divides the signal.

• Consolidate so each ad set can clear the learning phase
• Separate prospecting and retargeting, because the economics differ
• Use campaign budget optimisation to scale, not to run tests
• Avoid overlapping audiences inside one campaign, they bid against each other

Advantage+ or manual campaigns?

Advantage+ campaigns hand audience selection, placement and creative combination to Meta's automation. For ecommerce accounts with a clean product catalog and steady conversion volume, they often perform as well as or better than manual builds, and they take far less time to run.

The limits are worth knowing before you commit. You get less control over who is excluded, less visibility into which segment drove the result, and you rely on Meta's existing customer setting to keep the balance between new and returning buyers honest. If you do not set that, an Advantage+ campaign can report a healthy return that is mostly repeat customers you would have kept anyway.

We tend to run both. Advantage+ for efficient scale where the catalog and data support it, and a manual campaign alongside for testing new angles and for audiences the automation will not naturally reach. The manual side also produces the learning that makes the automated side better.

Tracking after the iOS changes, and the Conversions API

Apple's app tracking prompt in 2021 cut the amount of browser-side data Meta receives, and browser restrictions have tightened since. The visible effect is that the pixel under-reports conversions, attribution windows shortened, and the numbers in Ads Manager stopped matching the numbers in your shop.

The Conversions API is the fix. It sends conversion events from your server to Meta directly, rather than relying on the browser. Combined with the pixel and proper event deduplication, it restores a good deal of the lost signal. That matters twice over: for reporting, and more importantly for delivery, because Meta optimises against the events it can see.

Two things make it work properly. Sending the customer information parameters you legitimately hold, hashed, so events can be matched. And event deduplication set up correctly, so a purchase sent by both the pixel and the server is counted once. Get either wrong and the reporting becomes worse, not better.

• Run pixel and Conversions API together, with deduplication configured
• Send hashed customer parameters so more events can be matched
• Expect platform numbers and shop numbers to differ, and reconcile deliberately
• Check the event match quality score, it is the fastest health indicator

A creative testing framework that produces answers

Test one variable at a time and test it at the level that matters most. On Meta the hook is the biggest lever, so it is the first thing to isolate. Same offer, same product, five openings. Once you know which hook holds attention, test the format, then the offer, then the details.

Give each test enough budget and time to mean something. A test that ends after two hundred impressions has told you nothing except which ad got lucky. Set a threshold before you start, usually a spend figure roughly equal to a few times your target cost per acquisition, and do not call it early.

Keep a record. The value of testing compounds only if you write down what won and why. Over a few months you build a set of hooks, formats and angles that reliably work for your audience, and new creative starts from that rather than from a blank page.

• Test hooks first, they carry the most weight
• Change one variable per test so the result is readable
• Set the spend threshold before launch and hold to it
• Keep a written log of winners so learning accumulates

Frequency, fatigue and audience saturation

Creative fatigue shows up in the metrics before it shows up in the return. Watch click-through rate falling while frequency rises. That combination means the same people are seeing the same ad repeatedly and responding less. Cost per acquisition follows a week or two later, by which point you have paid for the lesson.

How fast this happens depends on audience size relative to spend. A small country and a large budget will exhaust an audience in days. A large audience and a modest budget can run the same creative for weeks. There is no universal frequency number that means trouble, so watch the trend in your own account rather than a rule from a blog post.

The fix is new creative, not a new audience. Rotating the same three ads through different interest stacks reaches largely the same people wearing a different label. Refreshing the hook, the format or the angle genuinely resets the response. This is why creative production capacity, not audience research, is the real constraint on scaling Meta.