Daily budget and the learning phase

The first days of a new ad set are noisy. On Monday a lead costs RM 8. On Tuesday it costs RM 25. The urge is to fix it: raise the budget, change the audience, rewrite the headline. That urge is the most expensive habit in Meta ads, because most fixes send the ad set back to the start.

What the learning phase is

When you launch an ad set, Meta does not yet know who will respond. So it tries different people, places and times, and watches what happens. Meta calls this the learning phase. While it runs, costs swing from day to day. That is Meta testing, not your ad failing.

Meta says an ad set leaves the learning phase after about 50 results in 7 days from its last big change. A "result" is whatever you asked Meta to look for: purchases, leads or chats. After that, delivery settles and costs become more steady.

You can see the status in Ads Manager, in the Delivery column of each ad set. "Learning" means it is still learning. "Active" means it is done.

What Learning Limited means

Sometimes the status reads Learning Limited. It means Meta does not expect this ad set to reach about 50 results in 7 days. The usual causes are a budget too small for your cost per result, an audience too narrow, or a result that happens too rarely.

It is not a penalty, and a small test can still be read. But if you want the ad set to learn fully, there are three honest fixes:

1. Raise the budget so it can buy about 50 results a week. The maths is in How much to spend on Meta ads, in ringgit.
2. Pick a result that happens more often. A chat or an add to cart happens more often than a sale.
3. Merge ad sets. Two ad sets at RM 25 a day learn slower than one at RM 50.

The edits that restart learning

Meta lists the changes it treats as big. Each one restarts the learning phase and throws away what the ad set learned so far.

On budget, Meta gives no exact percentage. It only says that moving from 100 to 101 is unlikely to restart learning, while 100 to 1,000 may. The "20%" you see online is a rule of thumb from advertisers, not from Meta. The learning phase guide goes deeper, including when adding an ad does and does not reset it in practice.

The first-week rule

Set the budget once. Then leave the ad set alone for 7 days. Do not judge it on day 2.

Two exceptions are worth a restart:

• Something is broken. A wrong price, a dead link or a spelling mistake in the offer. Fix it at once. A restart costs less than paying for a broken ad.
• The spend is far past what you can afford. Pause it, and plan a smaller test.

The best way to avoid the first exception is to check every ad before launch. On the AdPlay.ai ad page, Ad check looks for problems before you click Launch.

A worked example: the Penang tuition centre

Say you run a tuition centre in Penang. You want parents to book a free trial class, so your result is a lead. You can spend RM 50 a day, and you hope a lead costs about RM 15.

1. Can it learn? RM 50 a day is RM 350 a week. At RM 15 a lead, that buys about 23 leads, short of 50. Expect Learning Limited.
2. Is that fine? For a first test, yes. 23 leads a week is enough to see which ad brings bookings at the lowest cost.
3. Day 2. One ad has spent RM 40 with no leads. You leave it.
4. Day 4. The same ad has 4 leads, and the cost per lead has dropped. Leaving it was right.
5. Day 8. You read the results for the full week. Now you can act: turn off the weakest ad, and keep the rest.

If you later want the ad set to learn fully at RM 15 a lead, you need about RM 107 a day (50 × RM 15 ÷ 7). Raise the budget step by step once you have a winner, as shown in Raise the budget safely.

How Copilot keeps its hands off

On Pro, Copilot in AdPlay.ai reads your Meta account every morning and suggests changes as cards you approve. It follows the same rule. It does not suggest raising the budget of an ad set that is still learning. When it does suggest a raise, each step is at most 20%, with at least 72 hours between budget changes on the same ad set or campaign. Nothing changes until you approve the card. See Copilot for how the cards work.

Checklist

• I know my target cost per result.
• I know if my budget can buy about 50 results in 7 days, or I accept Learning Limited for this test.
• I checked every ad before launch.
• I wrote down the date 7 days after launch.
• I will not change the audience, the ads or the budget before that date unless something is broken.

Your homework

Write three things in your calendar: your daily budget, your target cost per result, and the date 7 days after your launch. Work out if the budget can buy about 50 results a week. If it cannot, decide now: accept Learning Limited for this test, or pick a result that happens more often.

Then open Campaigns in AdPlay.ai and check that the goal line at the top shows the same target. If it does not, click Change goal and fix it.