How many days should a Meta ad run before I judge it?

About 7 days, unless it is clearly broken. Meta's learning phase needs around 50 optimisation events in about 7 days, and a full week also covers both weekday and weekend buyers. Judge by spend too: an ad that has spent less than your target cost per sale has not had a fair test.

Does pausing an ad restart the learning phase?

A short pause does not. Meta lists a pause of 7 days or more as a significant edit, so a long pause sends the ad set back into learning when you turn it on again.

Should I delete a bad ad or pause it?

Pause it. Deleting is permanent: you cannot turn a deleted ad back on. A paused ad keeps its numbers in view, and you can turn it back on later.

When to stop, keep or change a Meta ad

Small shops lose money on ads in two ways. Some stop a good ad on day one because it spent RM 15 with no sale. Others leave a tired ad running for weeks because it used to work. This lesson gives you rules for both.

First, give the ad a fair test

Wait about a week. When an ad set starts, Meta is still learning who to show it to. Meta says an ad set needs about 50 optimisation events (sales, leads, or whatever you picked) in about 7 days to leave this learning phase. Results jump around until then. A full week also covers both weekday and weekend buyers.

Do not edit it while you wait. Changing the audience, the ad, the goal or the bid, or making a big budget change, restarts learning. So does adding an ad, or pausing for 7 days or more. Meta calls these significant edits. Every edit throws away what the ad set learned.

Judge by spend, not by hours. A common rule among ad buyers (not a Meta rule): an ad has had a fair test when it has spent one to two times your target cost per sale. If your target is RM 20, an ad with RM 12 spent and no sale has told you nothing. An ad with RM 40 spent and no sale has.

If you do not have a target yet, set one first. See Good numbers in Malaysia.

The three calls: keep, change or stop

After the fair test, every ad gets one of three calls.

Call	When	What you do
Keep	It sells at or under your target	Leave it alone. No edits.
Change	People click (CTR is fine) but each sale costs more than your target. Or it was a winner and is getting tired.	Make a new version with one change and run it beside the old one.
Stop	It spent two times your target with no sale. Or it gets far fewer clicks than your other ads.	Pause it.

"Change" is where most of your best ads come from. A good hook with a weak offer, or a good offer with a dull picture, is one change away from working. Change one thing at a time, so you know what helped. The Test one change lesson shows how.

Stop it the same day if something is broken

Some problems do not need a week:

The link goes to the wrong page, or the page is down.
The price or offer in the ad is wrong.
Sales are coming in but Meta shows none, so tracking is broken.
Meta rejected the ad (see Rejected ads).

Fix the cause, then start again.

Know when a winner is getting tired

A good ad does not last forever. Watch for three numbers moving together: frequency up, CTR down, cost per sale up. Search Engine Land's 2025 guidance puts the practical ceiling for frequency at about 3 for new audiences and about 5 for retargeting.

Do not edit the tired winner. An edit restarts learning and wastes what is left of it. Launch a fresh version beside it, and pause the old one once the new one sells. The Ad fatigue lesson covers this.

Pause, do not delete

Deleting an ad is permanent: you cannot turn it back on. Pause it instead. A paused ad keeps its numbers in view, and you can turn it back on later.

A worked example

Say you sell kuih raya in tins online at RM 45 and keep RM 20 a tin. Your target is RM 15 per sale. You run three ads in one ad set at RM 60 a day. Meta does not split the budget evenly, so after 7 days:

Ad	Spent	Sales	Cost per sale	Call
Family at the open house	RM 180	14	RM 12.86	Keep
Close-up of the tin with the price	RM 160	6	RM 26.67	Change: same picture, test a "buy 3, free delivery" offer
Flat lay of six kuih	RM 80	0	none	Stop: RM 80 is more than two times the RM 15 target

The total is RM 420, seven days at RM 60. Next week you run the family ad, the new tin offer and one new idea.

How AdPlay.ai helps

Insights in AdPlay.ai uses rules much like these, on the last 7 full days. Its Copilot says column gives each ad a verdict:

Too new to judge: the ad has run under 7 days.
Selling well: 3 or more sales, at or under your target.
Selling: 3 or more sales, but above your target.
No sales yet: no sale, and spend of at least two times your target.
Tiring (shown beside the verdict): seen 2.5 times or more per person, with clicks per view down 20% on the week before.

The What works table in Insights, with a Copilot says verdict for each ad: Selling well, Selling, No sales yet, Rejected by Meta, Tiring and Too new to judge

See the Insights guide.

For a Change call on an image ad you made in AdPlay.ai, open it from Ads and use Create variant to make a new ad with one thing changed. See Create variant.

On Pro, Copilot reads your Meta account every morning and can suggest pausing an ad as a card. Nothing changes until you tap the card's button (Pause ad). You can undo it for 7 days, and Copilot never pauses the last running ad in an ad set. See Copilot.

Checklist

You waited about 7 days and did not edit the ad.
The ad spent at least your target cost per sale before you judged it.
Each ad has a call: Keep, Change or Stop.
Each Change has one change, not three.
You paused, not deleted.

Your homework

List every ad that ran last week. Next to each one, write its spend, its results, its cost per result and your call: Keep, Change or Stop. Then open Insights in AdPlay.ai and compare your calls with the Copilot says column. Where you disagree, write down why.