A Facebook ads campaign structure falls apart for a predictable reason. Every new idea becomes a new campaign. After a few months nobody can tell whether a result came from the audience, the creative, the offer, or the fact that three campaigns are bidding on the same person.
Structure is not a template you download. It is a way to make the next decision readable. If you cannot say what a campaign is for, it should not exist.
Give each campaign one job
A useful split for a store that already sells is simple. One place finds new customers. One place speaks to people who already know you. One place, if the economics support it, sells to existing customers with a different offer. Country and catalog complexity can force more splits. A new campaign every Monday does not.
- Prospecting has one job: pay to reach people who have not bought.
- Retargeting has one job: reach people who already showed intent, with a message that assumes that.
- Existing customers get their own treatment when the product, margin, or replenishment cycle is different from acquisition.
Inside those jobs, creative is usually the variable that moves the account. A new audience stack with the same five ads is not a test. It is a rearrangement.
Read the account before you rebuild it
I do not open a new brand and paste in a structure from the last one. The numbers decide. Before I move budget I want to know four things.
- Is the purchase event real? A broken pixel, a duplicated Conversions API event, or a purchase that includes a view-content by mistake will make every structure look smart or stupid for the wrong reason.
- What is already carrying the account? An ad or ad set with a long profitable history should not be reset so the account looks tidy.
- Where does the revenue come from? New customers and returning customers can share one ROAS number and be two different businesses.
- How many distinct ads can the brand actually make this month? Structure that assumes 20 new concepts a week will sit empty.
That last point is the one agencies skip. A structure is only as good as the creative you can feed it. If the bottleneck is photos, edits, and offers, more campaigns will not fix it.
When to split further
Split by country when the offer, price, language, or shipping reality is different enough that one ad set would hide a losing market inside a winning one. Split by product when a hero SKU has its own margin and creative, and a secondary product would only steal delivery. Do not split because a blog post from 2021 listed eight campaign types.
Advantage+ and broader delivery changed where the control sits. You still decide the job of the campaign, the exclusions, the creative, and the budget. You do not get that control back by building a lattice of interests that Meta will ignore.
Exclusions, catalog, and the ads you can actually make
Structure fails in the settings people treat as paperwork. Exclusions decide whether prospecting is buying people who already purchased. The catalog decides whether a dynamic ad is showing the product you have in stock, at the price on the site. The pixel and Conversions API decide whether any of the ROAS you will quote next month is real. I will not design a pretty campaign tree on top of a purchase event that double-fires.
Count the ads the brand can make, not the ads a framework assumes. If the team can produce four distinct concepts a month, the account should be built to learn from four concepts. A structure that needs a new winner every few days will sit on the same two videos and call it testing. Creative volume is an input. Campaign count is not a substitute for it.
- Prospecting should exclude recent purchasers, or you will congratulate yourself for sales that were coming back anyway.
- A catalog ad needs a feed that matches the site: price, availability, and the product you want pushed.
- One offer per ad. A discount, a bundle, and a new arrival in the same creative is three tests pretending to be one.
- Name campaigns for the job, not the date. “Prospecting – US – hero SKU” still means something in October. “Test 14 final v3” does not.
Country splits belong here too, but only when the customer is actually different. Shipping cost, language, price, and returns policy are reasons. A hunch that “Canada might work” is not. I would rather one country with a readable account than five countries sharing one budget and hiding a loser inside a winner.
Do not copy another brand’s account
The structure that held a supplement subscription will punish a high-ticket furniture brand, and the reverse is just as messy. Repeat purchase, average order value, and how long a person takes to decide are the bones of the account. If the last brand I worked on replenishes every 45 days and yours is a considered purchase, I am not pasting their campaigns into your ad account.
This is also why a recycled agency playbook shows up so often. The junior has a checklist: prospecting campaign, retargeting campaign, a lookalike from 2019, three interest stacks. The checklist does not know your margin. When I open an account I want the last 60 to 90 days left intact until I can see what is already paying for itself. Tidying a profitable ad set so the left nav looks organized is how people delete the business.
If you are comparing account structures from two buyers, ask each of them what they would refuse to change in the first week. The useful answer names a specific ad, audience, or offer and says why. The weak answer is “we’ll rebuild it properly.” Proper is not a structure. Proper is a decision you can explain after the change.
A week inside the account
Once the jobs are clear, the work is repetitive on purpose. I look at spend against purchase value, which ads are tiring, which country or product is quietly losing, and whether creative is the constraint or the site is. Then one change, with a reason, and a date when we will read it. A Facebook ads campaign structure is only useful if Thursday still makes sense.
- Check that yesterday’s purchases landed in Ads Manager and in the store. A gap is a tracking problem until proven otherwise.
- Read prospecting separately from warm traffic. A blended ROAS will flatter the ads that talked to people who already knew you.
- Pause a loser only when it has had enough spend to be wrong. Fatigue is a frequency and a result, not a feeling.
- Put the next creative in before you need it. The account should not discover on Friday that nothing new is ready.
That rhythm is the point of a simple structure. You are not building a monument. You are building a place where a bad week has a cause. The companion to this is what a good ROAS is, because a clean account still gets misread if the only number anyone quotes is the ratio.
What “clean” actually means
A clean account is one where a bad week has a cause. You can point at the creative that fatigued, the country that stopped converting, or the offer that only worked on warm traffic. You can pause a thing without pausing the business.
If you are comparing buyers, ask them to walk through an account this way. The walkthrough matters more than the chart. I wrote a companion note on what a good ROAS is so the number on that chart does not get the last word.
One last filter before you adopt anyone’s diagram. Ask what breaks if creative slows down for a month. A structure that only works when a studio feeds it twenty new ads is a studio plan, not an account plan. Ask what you would pause on a bad Tuesday without stopping the ads that pay the bills. If the answer is “we’d have to rebuild,” the account is not structured. It is piled.
I keep the diagram small on purpose. Prospecting, people who already know the brand, and existing customers when the offer is actually different. Inside those, the ads change and the job does not. When a founder asks for a more sophisticated setup, I ask what decision the extra campaign would make readable. If they cannot name it, we do not build it. Sophistication that nobody can operate on a Thursday is just clutter with a legend. The account should still make sense if I am the only person in it, because on the accounts I run, I am.
If you inherit a messy account, do not start by deleting it. Export the last 90 days, name what each live campaign is actually doing, and only then turn off the ones that duplicate a job or spend without a purchase you trust. Deleting history to get a clean left nav throws away the only evidence you have. Rename first. Pause second. Rebuild last, and only the part that failed the test.