What each level of the hierarchy decides
Settings on Meta live at a fixed level, and the delivery system learns at the ad set. That second point is the one that shapes everything else: the learning phaseThe period after launch or a significant edit when an ad set's delivery is still being optimised., the budget minimums and the stability of cost per result (CPACost of each conversion.) are all properties of the ad set, so the number of ad sets you create is the number of things the system has to learn separately.
| Level | Owns | Changes here affect |
|---|---|---|
| Campaign | Objective, buying type (auction or reservation), special ad category, campaign spending limit, Advantage+ campaign budget when enabled | Every ad set beneath it; the objective cannot be changed once published |
| Ad set | Budget and schedule (when not on the campaign), audience, placements, optimisation goal and conversion event, bid strategy, attribution setting | Learning: most ad-set edits restart it |
| Ad | Creative, format, copy, destination, URL parameters, Advantage+ creative options | Only that ad; adding ads to a set is a significant edit in Meta's terms |
Two consequences follow. Anything you want to compare fairly, such as two audiences or two budget levels, has to be separated at the ad set level or in a formal test, because the system will not report a clean split inside one ad set. Anything you want the system to optimise across, such as several creatives for one audience, should sit inside one ad set so it can move impressions towards the winner. The art is choosing which comparisons are worth the fragmentation they cost.
Consolidation versus segmentation: the governing principle
Segmentation feels like control. An ad set per age band, per interest, per placement, per creative angle gives you a report that says exactly where the results came from. The cost is that each of those ad sets has to find enough conversion events on its own to exit learning, and most will not: the budget is split, the events are split, and the system spends its time exploring instead of exploiting. Meta's own guidance for the learning phase points the same way, recommending that advertisers combine ad sets and avoid frequent edits so that each one accumulates events faster.
Consolidation hands the choice of who sees what to the delivery system and asks it to spend where results are cheapest. It gives up reporting granularity in exchange for faster learning and lower cost per result, and on a platform where the system is predicting from behaviour rather than matching a declared query, that trade is usually right. The exceptions are real and specific: markets with separate budgets, audiences that must be spoken to in different languages, retargeting pools that need their own frequency control, and tests designed to answer a structural question. Segment for one of those reasons, not for the comfort of a tidy report. The targeting hub covers how audience choices interact with this principle, including why broad targeting with strong creative often beats a stack of interest-based ad sets.
The core guides
The core guides cover one mechanism each, and they are written to be read in order.
Meta campaign objectives explains what the objective tells the delivery system, the optimisation goals and conversion events that sit beneath each objective, how the choice changes who sees the ad and what you are charged for, and how to match an objective to a business goal. It is the page to read before building anything, because a campaign optimised for the wrong event is efficiently wrong.
Campaign versus ad set budgets compares Advantage+ campaign budget, where the system moves spend between ad sets in real time, with budgets set per ad set, where spend is guaranteed to each. It covers how each distributes money, the spend limits and minimums, daily against lifetime budgets, and how to switch between the two without breaking a working campaign.
The learning phase describes the documented learning behaviour: what the phase is, what "learning limited" means, the edits that reset it, and how to structure ad sets and budgets so they exit it. Most of the structural advice elsewhere in this hub traces back to this guide.
Beyond this section, the optimisation hub covers what to do once a structure is running, including scaling and fatigue, and the measurement hub covers the conversion signal every structural decision here depends on.
Further guides
Further guides cover the layer above the first campaign and the agency or freelancer hired to work in that layer, the bid strategy and the messaging route that sit on top of a structure, and running Meta Ads when the budget is small:
- Setting up a Meta business portfolio is a tutorial for the ownership and access layer before the first campaign: who owns the portfolio, the admins, the Page and Instagram account, the ad account, the dataset and domain, and who gets access.
- Hiring and evaluating a Meta Ads agency covers what an agency or freelancer is hired to do, the assets that stay in the business's portfolio, the questions about creative supply, structure and the Conversions API, warning signs, reports checked against orders and changing provider without losing the account.
- Bid strategies on Meta covers how highest volume, highest value, a cost per result goal, a ROAS goal and a bid cap work, and how to choose one from your own economics.
- Click-to-message ads covers choosing Messenger, Instagram or WhatsApp, the goal, a reply flow that qualifies, response times and counting the conversations that became customers.
- Meta Ads when the budget is small covers what small means against your own event costs, one consolidated ad set, an event the budget can feed and reading results when events are few.
A first campaign: the decisions in order
For a business building its first Meta campaign, the decisions arrive in a fixed order, and taking them out of order is how accounts end up rebuilt.
- Confirm the signal. Is the pixel firing, is the Conversions API sending the same events, are they deduplicated, and does the event you want to optimise for actually occur often enough to learn from? If the answer is no, fix measurement first.
- Choose the objective from the business goal: Sales for purchases, Leads for enquiries, Awareness for reach. Then choose the optimisation event beneath it, as close to revenue as the volume allows.
- Choose the budget level. One ad set means the distinction barely matters; several ad sets sharing a goal favour Advantage+ campaign budget; ad sets that must each spend a fixed amount need their own budgets.
- Set the budget at a level the optimisation event can support, using the CPA calculator to work back from target cost per result and conversion rate (CVRShare of visits or clicks that convert.) to the spend required.
- Build one ad set, broad or with one well-sourced audience, all placements on, and several creatives.
- Leave it alone through the learning phase, adding creative rather than editing settings.
A language school's first lead campaign
A language school in Manchester wants enrolment enquiries for evening courses. The team confirms the pixel fires a Lead event on the enquiry form's thank-you page and that the Conversions API duplicates it with a matching event ID. They build one Leads campaign with Advantage+ campaign budget at £45 a day, one ad set targeting the city and its commuter belt with no interest layering, and four ads: two short videos of a class in progress, one testimonial image and one plain offer graphic. They expect 2 enquiries a day at about £22 each based on the form's historical conversion rate of 6% and a planned cost per click of £1.30. The first ten days average £31 per enquiry while the ad set learns, then settle at £19 as the two videos take most of the spend. The team's only edits in the first month are adding two more videos and retiring the offer graphic.
Common structure problems
Most structural problems show up as a symptom in the reporting rather than as an obvious mistake in the settings. The table maps each symptom to the guide that explains it.
| Symptom | Likely structural cause | Guide |
|---|---|---|
| Lots of link clicks, few sales | Optimised for clicks or landing page views rather than purchase | Campaign objectives |
| Ad sets stuck in "learning limited" | Too many ad sets, too little budget each, or a rare optimisation event | The learning phase |
| One ad set takes all the spend, another gets none | Advantage+ campaign budget doing its job; add minimums only if that ad set must spend | Campaign vs ad set budgets |
| Cost per result jumps after every change | Edits resetting learning | The learning phase |
| Cost per thousand impressions (CPMCost per 1,000 impressions.) rising while audiences overlap | Ad sets bidding against each other for the same people | Targeting hub |
| Return on ad spend (ROASRevenue generated per unit of ad spend.) looks fine in the platform, poor in the books | Attribution and value tracking, not structure | Measurement hub |
| Clicks from launch but no recorded sales or leads | Wrong optimisation goal, missing events, placements, landing page or attribution window | Clicks but no conversions, diagnosed |
| A day's spend lands above the daily budget | Documented daily flexibility, campaign budget, lifetime pacing or a mid-day edit | Spending more than the daily budget |
| Spend concentrates on one ad even within a single ad set | Delivery predictions and a structure that cannot split fairly | When all the budget goes to one ad |
| A small budget split across several ad sets, none of which gives a readable result | Too many ad sets, or an optimisation event the budget cannot feed | Meta Ads when the budget is small |
| Message ads start plenty of conversations that never become customers | No reply flow that qualifies, slow responses, or conversations counted instead of customers | Click-to-message ads |
How the author applies these decisions on live accounts, and which account types that experience covers, is described on the Meta Ads expertise page; this section teaches the mechanisms.
