Setting the budget wrong cripples the campaign before it even starts. The difference between daily and lifetime budget, why the learning phase breaks, and the safe way to raise the budget.
The most-debated topic in an ad account is the amount of the budget. Yet what cripples most campaigns is not the amount but how the budget is set and how often it's changed. In this article we explain budget setup in Meta (Facebook and Instagram) ads, why the learning phase breaks, and the safe way to raise the budget.
The difference between daily and lifetime budget
The daily budget is the amount the ad set aims to spend in one day. Meta doesn't spread this evenly through the day; it spends more in the hours where it sees opportunity and less when it sees less. It may go over the budget on one day and balance out by staying under the next. So looking at a single day's report and saying "it overspent" is misleading; you have to look at the weekly total.
The lifetime budget is a single amount for the entire date range you set. Meta spends this amount by shifting it toward the days it judges most likely to convert. It sounds smarter, but it has a cost: it stops when the campaign period ends, and changing the budget during the period resets learning.
The practical distinction: for campaigns that will run indefinitely and constantly target sales, use a daily budget; for a period with a defined start and end (sale week, launch, season), a lifetime budget is easier to manage.
The learning phase: the campaign's most fragile period
When a new ad set goes live, Meta doesn't know who will convert. The period until it gathers signal is called the learning phase. In Meta's own documentation, closing this phase requires the ad set to gather about 50 conversions within 7 days. This threshold is set by the platform, not by AGO.
Three conclusions follow:
- If the ad set runs on a budget too narrow to produce 50 conversions a week, learning never closes. The result: volatile cost, unpredictable outcome.
- Splitting the budget across many ad sets drops each below the threshold. One strong set usually works better than five weak ones.
- If your conversion event happens very rarely (for example the purchase of an expensive service), you need to target a more frequent event: form fill, call start, add-to-cart, and the like.
Why does changing the budget break learning?
When you make a significant change to the ad set, Meta restarts the learning phase. Large jumps in budget count among these changes. So doubling the budget in the morning can render useless the information the system has gathered up to that day.
The widely accepted practice in the industry is not to raise the budget by more than 20–30% at once and to wait at least 2–3 days between increases. This is not a platform rule but the practical way to grow without breaking learning. In accounts that want to grow in a hurry, you typically see this: the budget doubles, the cost triples, panic sets in, the budget is pulled back, and the campaign never recovers.
Campaign-level budget (Advantage campaign budget)
You can give the budget not to individual ad sets but to the whole campaign and leave the distribution to Meta. The advantage is that the system gives more share to the set that works well. The disadvantage is that your ability to say "this much must definitely be spent on this audience" decreases.
The general tendency is this: in the testing phase, give ad sets separate budgets to see which audience works; after the winner is clear, move to campaign level. Doing the reverse means asking the system to decide when there's no data yet.
Don't confuse the ad budget with the agency fee
A frequent conceptual confusion: the ad budget is the money paid directly to Meta that makes the ad show. The agency fee is in return for setting up the campaign, producing creatives and daily management. These are separate items, and usually the ad budget concerns the advertiser directly; the agency fee doesn't have to rise automatically when the budget rises.
A reasonable setup for the start
A simple order that works in a new account:
- One campaign, a clear goal (sales or leads).
- Two, at most three ad sets — a broad audience and one remarketing audience.
- 3–5 different creatives per set; variety of image and text is more useful than variety of audience.
- A daily budget of at least 5 times your target conversion cost. If you expect 100 TL per conversion, under 500 TL a day is too narrow for the system to learn.
- For the first 7 days, don't touch anything. This is the hardest step.
Five common mistakes
- Looking at the report every day and intervening. Daily fluctuation is normal; decisions are made with weekly data.
- Opening several sets for the same audience. Your own ads compete with each other and cost rises.
- Launching a sales campaign without conversion tracking set up. The system can't know what to optimize.
- Never refreshing the creative. When the same image is shown to the same audience repeatedly, response drops.
- Turning the campaign off at the weekend. Interruptions can restart learning.
Frequently asked questions
Can I raise the daily budget on weekdays and lower it at the weekend?
Technically possible, but frequent, large changes break learning. If you want to spend more at certain hours, ad scheduling (dayparting) is a more suitable path and works with a lifetime budget.
Does lowering the budget also break learning?
Yes. Direction doesn't matter; a significant change can restart the learning phase. Progressing in small, infrequent steps applies in both directions.
How much should I start with?
There's no single right figure; the measure is the conversion cost in your sector. A daily budget of 5 times the target cost is considered a reasonable start for learning to close. As product price and competition rise, so does this amount.
If I pause and restart the campaign, does it continue where it left off?
Short pauses may not cause problems; but after long interruptions the ad set can return to the learning phase. If the budget isn't finished, lowering the budget gradually is safer than stopping the campaign.
