22 Jul Why Your Best Meta Ad Loses Budget the Next Week
If you’re running a large automated ad set — 10, 20, or more ads pooled together and left to Meta’s algorithm — you’ve likely seen this: one ad performs beautifully for a week, strong CTR, solid cost per result, and then the following week, Meta quietly shifts most of the budget elsewhere to test other ads in the pool. It’s not random, and it’s not necessarily a sign anything’s broken. It’s a specific, well-documented consequence of how the current delivery system actually works.
What’s actually happening
In a large ad set, a small number of ads — often just two or three out of ten — typically absorb most of the budget. That part is expected: the system is trying to find and reward what’s working. The frustrating part is that this allocation isn’t sticky. An ad performing well one week can see its budget cut sharply the next, as the system continues testing across the pool rather than locking in on a proven winner. This creates real volatility and unpredictability in accounts running this structure, even when nothing about the ad itself has changed.
The mechanism behind part of this: near-duplicate creative competing with itself
One specific, current explanation worth understanding: Meta’s delivery system groups creative that looks functionally identical — same scene, same core message, same visual structure, with only minor variations like color or a few words of copy — under the same internal grouping. When several “different” ads in your pool are actually near-duplicates of each other in the system’s eyes, they end up competing against each other for the same auction slot rather than each reaching distinct people. That internal competition can drive costs up without improving results, and it can look a lot like the same “budget keeps shifting around” volatility described above.
The practical implication: adding more ads to a pool doesn’t help if they’re all variations on the same underlying creative. Genuine creative diversity — different formats, different angles, different messaging approaches — matters more than raw quantity.
Two different ways advertisers are responding
There’s real, honest disagreement here, and it’s worth knowing both sides rather than assuming one is settled.
One approach: keep structure simple, lean into diversity, trust the system. This view holds that the fix isn’t fighting the algorithm’s behavior — it’s feeding it better input. Simplify account structure, avoid narrow manual targeting layers that restrict the signal the system has to work with, and continuously supply genuinely varied creative rather than minor tweaks on the same concept.
A different approach: hybrid structure, isolate what’s proven. This is a more hands-on response: split budget between the large automated ad set and a smaller, parallel testing pipeline of single-ad campaigns. A single ad, isolated in its own ad set, tends to show noticeably more stable week-to-week performance than the same ad competing inside a large pool. Once an ad proves itself over a couple of weeks in that isolated environment, it gets promoted into a dedicated single-ad campaign for scaling — more predictable to manage, because it isn’t subject to the same internal reshuffling. Ads that don’t prove out get retired and replaced with new tests. Over time, this can build up a bench of several stable, proven single-ad campaigns running alongside the main automated structure.
Neither approach is universally right. Results genuinely vary by account — some see strong, consistent performance sticking with the fully automated structure; others see meaningfully more stability after shifting toward the hybrid model. The honest position is that this needs testing on your own account rather than adopting either framework on faith.
The mindset that matters more than the framework
The goal isn’t proving which structure is philosophically correct — it’s finding whichever one actually produces better results for a specific account, and staying open to switching if the data says otherwise. Treating either approach as the permanent answer, rather than something to keep testing against real performance, is the more common mistake than picking the “wrong” structure in the first place.
This breakdown is written by Jason Gan, a Meta Certified Professional and Badged Meta Business Partner who has personally audited over 1,000 advertiser accounts since 2010. You can see real account breakdowns on the Jason Gan YouTube channel.
Frequently Asked Questions
Why does Meta stop spending on my best-performing ad? In large automated ad sets, Meta’s delivery system continuously re-tests the pool of ads rather than locking in permanently on whichever one performed best last week. A winning ad can see its budget cut the following week as the system shifts spend to test others, even without a clear performance drop.
Why do many similar-looking ads in one ad set underperform? Meta’s delivery system groups near-duplicate creative — same scene, same message, same structure with only minor differences — under the same internal identifier. Those near-duplicates end up competing against each other for the same auction slot instead of expanding reach, which can waste budget and push costs up without improving results.
Should I run many ads in one automated ad set or isolate winners into single-ad campaigns? There’s genuine disagreement here, and it depends on the account. One approach keeps structure simple and feeds the automated system a wide range of genuinely different creative. A different approach splits budget — running the automated structure alongside gradual single-ad testing, then moving proven winners into dedicated single-ad campaigns for more predictable, stable scaling. Testing both on your own account is the only reliable way to know which fits.
If your ad spend keeps shifting in ways you can’t predict
Book a Meta Ads Audit — $80 / 30 Minutes →
No login required. Just a screen share, your real account, and a clear read on whether a hybrid structure would actually help your specific situation.