Facebook ads: what a realistic budget buys

The auction, in plain terms

Meta does not run an auction for your audience. It runs an auction for each individual impression opportunity — one person, opening the app, at one moment. You are competing against every other advertiser who wants that slot.

You do not win by bidding the most. The auction weighs what you bid against how likely that person is to take the action you are optimising for, and against the quality and relevance of the ad itself. A high bid with an ad nobody engages with loses to a lower bid with an ad people respond to, because the platform makes more money from the second one.

Two consequences follow, and they explain most of what you see in an account. First, your cost per result is set as much by your creative as by your budget. Second, you cannot simply buy reach — you buy the platform's confidence that this person will act. If your targeting tells the system nothing and your creative is generic, you pay a premium for the same attention a competitor gets cheaply.

Creative is the lever, not targeting

Targeting has been narrowed by privacy changes and by the platform's own direction of travel. The practical result is that for most accounts, the difference between a good month and a bad one is the creative, not the audience definition.

This changes what you actually produce. Instead of one polished ad, you want several distinct angles — the same product argued from different motivations. A time-saving angle, a cost angle, a risk angle, a social-proof angle. Then variations within each angle: a different opening shot, a different first line, a different format. The algorithm needs options before it can find what works, and it needs new options before the current ones wear out.

Creative fatigue is not a mystery. The same person sees the same ad repeatedly, attention drops, engagement falls, and the platform charges more to keep serving it. The fix is a production cadence, not a bid adjustment. A business that can ship a handful of new concepts every month will beat one that spends the same money on a single hero video and then wonders why it stopped working in week three.

Audiences: what is actually left

There are four useful categories.

• Broad. Minimal targeting, letting the platform's delivery system find the buyers. Often the strongest option once you have reliable conversion data, because you are not guessing which interest correlates with purchase.
• Interest and demographic. Useful early, when you have no conversion signal, and useful for genuinely niche products. Otherwise usually a constraint rather than a precision tool.
• Lookalike. Built from a source audience — your customer list, your highest-value purchasers, your engaged visitors.
• Custom and remarketing. People who already interacted with you: site visitors, video viewers, page engagers, past customers.

A note on customer lists. Uploading your customer database to build a lookalike means processing personal data. In the EU you need a lawful basis, your privacy notice must describe it, and the platform acts as your processor. Never upload a list you bought.

Pixel-based audiences also shrink as cross-app tracking degrades, and models fill the gaps with estimates. Server-side event tracking recovers part of the signal and is now table stakes rather than an optimisation.

The pixel, events and what gets measured

The pixel is a piece of code on your site that reports what visitors do. The Conversions API does the same job from your server, which matters because browser-based tracking is blocked or degraded on a meaningful share of traffic. Running both and de-duplicating the events is the standard setup.

What you send determines what the platform can optimise for. A standard conversion event without a value teaches the system which people convert, but not which ones are worth having. Passing order value lets it optimise towards revenue, which is what you actually want. If you optimise for a form submission, you will get form submissions — including from people who will never buy.

Be clear-eyed about the numbers that come back. Attribution windows mean a conversion can be credited to an ad days after the click, and a share of conversions are modelled rather than observed, particularly on iOS. Platform results are an estimate with a bias towards the platform. We go into this properly in Attribution, honestly.

Retargeting is a multiplier, not a strategy

Retargeting almost always shows the best cost per acquisition in the account. This is not evidence that retargeting is brilliant. It is arithmetic: these people already visited, so some of the persuading is done. It is a multiplier applied to traffic you already paid for.

Two rules keep it honest. Exclude people who have already converted, unless you genuinely sell repeat purchases. And watch frequency — a small pool with a persistent budget produces the same ad twenty times, converting a warm audience into an annoyed one. When the pool is small, the budget should be small too.

How to read results without fooling yourself

The metrics stack in a fixed order. Cost per thousand impressions tells you what attention costs in your market and country. Click-through rate tells you whether the creative earns attention. Cost per result tells you whether the traffic is qualified. Return on ad spend tells you whether any of it is worth doing. Frequency and reach tell you whether you are buying new people or repeating yourself.

None of it is meaningful below a certain volume. The platform needs a stable flow of the event you are optimising for before delivery settles; until then you are in a learning period where results swing wildly and costs are inflated. Meta's own documented guidance is on the order of fifty optimisation events per ad set per week to exit it. That number is the whole argument of this guide, so it is worth sitting with.

It means your required budget is not a matter of taste. It is: events needed per week × your cost per event. If you need fifty purchases a week to exit learning and your cost per purchase is high, the budget required is enormous, and paid social may simply not be viable for that offer. If you can optimise for a cheaper, earlier event that still correlates with revenue — a qualified lead, a booked call, an add-to-cart — the arithmetic changes completely.

What a realistic budget buys

Ignore benchmark tables. They are averages across wildly different countries, categories and offers, and they tell you nothing about your unit economics. What budget actually buys is capability, and it breaks down like this.

Small budgets: proof of concept

At a low monthly spend, you can run one or two audiences and a handful of creative variations. What you get is a read on whether the message lands and what a click and a lead cost in your market. What you do not get is a stable cost per acquisition, because you are not generating enough events to exit learning. Treat this as research, not as a channel. If it produces a real signal, the next step is more budget, not more optimisation.

Mid budgets: a channel that can be managed

At a moderate spend, you can run a genuine test cadence — new creative every week or two, a retargeting layer, and enough volume that costs stabilise. This is the first band where paid social behaves like a system rather than an experiment. It is also the band where most businesses discover that their landing page, not their advertising, is the limiting factor.

Large budgets: the constraint moves

Past a certain point, more money does not buy more results — it buys more of the same results with declining efficiency. What limits you then is creative production and audience size. You need a pipeline of new concepts, a system for producing them, and a clear view of how many people in your market can realistically buy from you. Scaling budget without scaling creative is the most reliable way to watch costs climb.

In every band, the binding constraint is your unit economics. Paid social does not fix a business where acquisition costs more than the customer is worth. It accelerates the discovery of that fact.

When paid social is genuinely the wrong answer

• When search demand already exists and you are not capturing it. Someone searching for your category is further along than someone scrolling past it. Capture that first; it is cheaper and it converts.
• When your total market is tiny. If there are a few hundred plausible buyers worldwide, the platform cannot find them efficiently and frequency becomes unbearable. Direct outreach beats advertising here.
• When the offer cannot survive the cost of attention. Paid social is interruption marketing. It works when the value of a customer is high enough to absorb the cost of interrupting many people.
• When you have no creative capacity. An account with one ad and no plan for the next one is not a channel, it is a countdown.
• When you cannot measure anything. Without a working pixel, events and a downstream view of revenue, you are buying impressions and guessing at the rest.

Where most people get this wrong

The defining error is judging paid social by the wrong scoreboard. Likes, shares and comments are not results. An ad with a thousand reactions and no purchases has failed, and an ad that looks dull but produces cheap qualified leads has succeeded. If the report does not reach revenue, it is decoration.

The second error is fiddling. Turning campaigns on and off, changing budgets daily, and restructuring ad sets every week guarantees that nothing ever exits learning. Most of the "optimisation" happening in small accounts is the operator resetting the algorithm's education. Change one thing, give it enough volume to judge, then decide.

The third is trusting retargeting numbers as evidence that the whole system works. Retargeting reports a flattering cost per acquisition because it is credited with sales from people who were already close. Cut off the top of the funnel for a month and watch what happens to those numbers — that experiment is more informative than any dashboard.

The short version

• You are bidding in a per-impression auction judged on estimated action rate. Creative drives that rate as much as targeting does.
• Pass conversion value, not just conversion events, or you are teaching the platform to find cheap buyers of the wrong thing.
• Required budget is events needed × cost per event. Work that out before you commit a monthly figure.
• Low budgets buy research; mid budgets buy a manageable channel; large budgets make creative production the bottleneck.
• Skip paid social when search demand is uncaptured, the market is tiny, or nobody can produce new creative every month.