We get accounts that have been spending for months, with good weeks and losing weeks, and nobody can say why. The platform is not broken. Meta Ads works, and works well. What is missing is the structure that makes a result repeatable.
One question shows the difference. If someone asks why last week's ROAS was 4.1x and this week's is 1.8x, do you have an answer or a guess? If it is a guess, you are paying to learn without keeping anything.
What is Meta Ads, and how does it work?
Meta Ads is Meta's advertising system: one account, one Ads Manager, and ads that can run on Facebook, Instagram, Messenger, Threads and Audience Network, the network of outside apps. The available placements are listed in Meta's placement documentation.
Every campaign has three levels. The campaign sets the objective (sales, leads, traffic, awareness). The ad set decides audience, placements and budget. The ad is the creative people see.
Who sees what is decided by an auction, at every single impression. Meta's help center puts it plainly: for each ad it calculates a "total value" from three factors, the advertiser's bid, the estimated chance that this person takes the action, and the ad's quality. The highest total value wins. So a more relevant ad can beat a higher bid, and a weak creative costs more even on the same budget.
How much does Meta Ads cost?
There is no price list. You set the budget, daily or lifetime, and the price of each result comes out of the auction. What you pay per thousand impressions, per click or per purchase depends on how contested the audience is, how good the creative is and how many results the campaign can collect. The ranges we see in the Italian market are in the table further down.
One threshold matters more than budget. Meta says an ad set leaves the learning phase once it gets about 50 results in the week after its last significant edit. Below that, the numbers stay unstable whatever you do.
Which objective makes the algorithm work for you?
Meta optimises for exactly what you ask. Pick "Traffic" in Ads Manager and the algorithm finds people who click: it brings them to you, and they are the people who click on everything. Pick "Purchases" and it looks for people who buy.
This is the most common mistake, and the most expensive, because it does not show at first. Clicks come in, CPC looks fine, the report looks healthy. Then you check revenue and it does not add up.
Placement works the same way. A Reel catches people scrolling to switch off: if the message does not land in two seconds, it is over. The Facebook Feed reaches people who stop to read, and carries longer copy. Stories are vertical and immersive, and an image built for Feed looks like a glitch there. This is not taste: the same ad performs in one place and fails in another.
Why does tracking matter more than budget?
The Meta Pixel tracks from the browser. The Conversions API tracks from the server. You need both, sent with the same `event_id` and the same event name: Meta's deduplication docs explain that when the two arrive within 48 hours, it counts one event instead of two.
This is not redundancy. Between declined cookie consent, browser blocking and ad blockers, the Pixel alone loses part of the events, and loses them unevenly, so the data is not just incomplete, it is skewed. The algorithm optimises on what it receives: a skewed signal gives badly optimised campaigns, and you see the cost in CPA.
The events that matter are few. For ecommerce: `ViewContent`, `AddToCart`, `InitiateCheckout`, `Purchase`. For lead generation: `Lead` and `CompleteRegistration`. On Shopify, Meta's official integration turns on server-side sending in a few steps; on WooCommerce you need an extension or custom work.
In Europe consent adds a layer. Server-side tracking recovers some of the events limited consent loses, but it exempts you from nothing: you configure only the events you need, connect a consent management platform, and check everything in the test events tool before spending the first euro. An account that starts with a broken signal is not fixed by reading reports. It gets rebuilt.
Three audiences, three messages, not three copies of one
There are three audiences and they are built separately. Cold works on interests and behaviours. Warm uses lookalikes built on your buyers or leads. Hot is retargeting people who viewed key pages or abandoned the cart.
Almost everyone puts the budget on cold and runs the same creative for all three. That is the most frequent reason spend does not scale: you are asking a message written for strangers to convert people who already know you, and the other way round.
Formats follow the stage, not taste. Vertical 9:16 Reels and Stories (1080×1920) work in discovery because they do not look like ads. Carousels of 2-10 cards work in consideration, when you need to show several products or one benefit at a time. A static image with a clear headline and a direct call to action holds up on hot audiences, where nobody needs convincing that you exist. For direct-response video, 15 to 60 seconds is the length that works.
Having campaigns live on all three stages at once is the difference between an account that scales and one that burns budget. How the full path is built, stage by stage, is in the guide on how to build a funnel.
Which metrics should you watch, and in what order?
CPM, CPC, CPA and ROAS measure four different things and are not judged one at a time. CPM tells you how contested the auction is for the audience you picked. CPC tells you whether the creative does its job. CPA is the acquisition metric. ROAS is the final verdict on profitability.
The recurring mistake is looking at CPC and deciding. An ad at €1.20 per click converting at 5% gives a €24 CPA. An ad at €0.40 per click converting at 1% gives a €40 CPA. The second looks three times more efficient and costs almost double.
| Metric | Order of magnitude, Italian market | When to look at it |
|---|---|---|
| CPM | €4-12, up to €25 in finance and insurance | If it rises for no reason, the audience is too narrow |
| CPC | €0.30-1.50 | If it is high with a good CPA, leave it alone |
| CPA | €10-60, depends on product and objective | The metric you set budget on |
| ROAS | 3x-8x on structured ecommerce, below 1x is a loss | Read it against your margin, not a benchmark |
Those figures are orders of magnitude from the accounts we manage, not targets. The threshold that counts is yours: a store with 25% margin breaks even at a different ROAS from one with 50%. Work out the minimum acceptable ROAS before switching campaigns on, not after the first losing week.
What to check before going live
Two things outside the account move ROAS more than half the in-account tweaks. First: people who click and do not buy can enter an email sequence, which lowers overall CPA without spending another euro. Second: customer lists uploaded as excluded audiences stop you paying acquisition on people you already have.
And the landing page is not a separate piece of the funnel; it sits inside it. An ad that works, sent to a weak page, wastes budget with precision.
When should you hand Meta Ads to an agency, and when not?
Not everyone needs an agency. If you spend €300 a month, the management fee costs more than the mistakes it would prevent: better to learn on a small budget, and there is a guide for that, Meta Ads on a small budget.
It makes sense to talk when your monthly budget is at the point where one bad month costs more than the management, when nobody in-house can run tests systematically, or when you need to grow faster than it takes to build a team.
Our method on Meta is the one we use everywhere: AI agents do the volume — creative variants, reading performance, killing what does not convert quickly — and people keep strategy, message and quality control. In practice a test cycle closes in 10-14 days instead of 4-6 weeks. What working this way means, and where it breaks, is in what is vibe marketing.
The campaigns that scale are not the ones with the biggest budget. They are the ones where every euro spent produces data for the next decision.
Frequently asked questions
What is a good ROAS on Meta Ads?
There is no good value in absolute terms: it depends on your margin. On structured ecommerce the working range is between 3x and 8x, and below 1x the campaign loses money. But a business with 25% gross margin breaks even at a much higher ROAS than one with 50%, so the first thing to calculate is your own minimum acceptable ROAS, before switching on any campaign.
Do I really need the Conversions API if I already have the Pixel?
Yes, and it is not redundant. The Pixel tracks from the browser and loses events to declined consent, blocking and ad blockers; the Conversions API sends the same events from the server. With the same `event_id` and event name Meta deduplicates them instead of counting twice. The point is not more data: it is an undistorted signal, because the algorithm optimises on what it receives.
Which campaign objective should I pick if I sell online?
Purchases, not Traffic. Meta optimises for what you ask: ask for clicks and it brings people who click, who are not the people who buy. Traffic makes sense in specific cases, such as pushing volume to a piece of content, but if the result you care about is revenue, that objective trains the algorithm in the opposite direction.
How much budget do I need to start on Meta Ads?
The right question is not the absolute minimum but how many conversions a week an ad set can collect: Meta's guidance is about 50 in the week after the last significant edit to leave the learning phase. At a €10 cost per conversion that means roughly €70 a day to get out in normal time; you can start with less, but concentrate it on one campaign instead of spreading it across six.
How often should creatives be changed?
When the numbers say so, not on a calendar. The signal to watch is frequency on the hot audience and CPA rising while CPM stays flat: it means the audience has seen that ad enough. In our process a test cycle closes in 10-14 days, so new variants come in while the old ones are still profitable.
Want to know where your account is leaking?
We look at structure, tracking signal and the last eight weeks of numbers together, and tell you which piece is missing. We take on few clients at a time, so the review also tells us whether we are the right fit.
Request the free reviewKeep reading
Founder of As Growth, an AI-first growth partner. He runs growth and strategy across partner businesses: OroEtic, Giojello, Equity Plus and others. Apply if you want to work with us.