36 paid-ad results, each credited and linked to the source.
We have 247 published Meta ad results here, and 245 of them report an improvement.
The 49 results that give a CPA or CAC cut as a percentage have a median drop of 33%.
The 42 that give a Meta ROAS gain as a percentage have a median rise of about 55%.
Read those numbers with care. Only 26 results come from split tests or studies, 96 (39%) are vendor-reported, and almost nobody publishes a Meta test that lost.
Every result on this page ran on Meta, and 53 of them also used Google or TikTok.
We grade each one by its evidence, using the same evidence grades as the rest of the site.
Here is how this hub's 247 results break down:
| Evidence grade | Results |
|---|---|
| Before/after (account compared with an earlier period) | 123 |
| Campaign result (one campaign, no comparison) | 69 |
| Split test (control and variant run at the same time) | 23 |
| Screenshot snapshot | 23 |
| Unsourced claim | 6 |
| Study across many accounts | 3 |
Of the 23 split tests, 19 are vendor-reported, mostly from Smartly and Marpipe catalog tools.
In total, 96 results come from a tool or platform describing its own feature: Madgicx (20), Smartly (19), Marpipe (18), Insense (14), Billo (11), Northbeam (7) and Meta (5) supply most of them.
If you want the cross-platform view, see all paid ad results.
The results we trust most change one thing and keep the rest fixed.
In CPA €170.34 → €93.51 (−45%), @SamPiliero cut the product price by €10 while keeping the same ads and budget.
In Cost per order €84.38 → €58.63 (−30.5%), the budget stayed flat and only the ad quality changed.
Two other @SamPiliero before/after results: a restructure took CPM from CPM $50 → $15, and a Labor Day 15%-off promotion for a bedding brand returned Revenue +102%, blended ROAS 10.11x → 13.17x.
Power Digital reported incremental ROAS +220% and CAC −32% for YoungLA after spend moved from drop days to always-on.
Geo holdout tests guided the shift, but those two figures are a before/after of the whole program.
Many results changed more than one thing, so these lever counts add up to more than 247. Audiences were tagged most often (82), then campaign structure (58), budgets (47), creative volume (44), measurement (44), visuals (41) and switching formats (34).
Angles (29), offers (23), bidding (23), landing pages (19), copy (16) and hooks (15) were tagged less often.
Nearly half of the results (122) carry no format tag, because they changed structure or targeting instead.
Among the tagged ones, UGC leads with 60 results.
Then come video (51), catalog (38), partnership or whitelisted ads (23), static (20), AI-generated (8), carousel (6) and founder-led (4).
Catalog results are the most vendor-heavy group: 25 of 38 come from Smartly or Marpipe.
UGC and partnership-ad results come from agencies such as Flighted (20) as well as creator platforms Insense (14) and Billo (11); treat the platform figures as vendor-reported.
For agency-reported UGC results, start with facebook ugc ads.
The Advantage+ shopping results agree with each other but vary a lot in size.
Flighted's side-by-side comparison for Ando (not a formal split test) found 40% higher ROAS on ASC vs broad CBO.
JENNY BIRD reported 14% lower cost per purchase after adding Advantage+ shopping, and Grover reported CPA −70%.
The last two are both vendor-reported. All of them are collected under advantage plus shopping campaigns.
General ecommerce (57) and apparel (46) make up 42% of the hub.
Home (19), accessories (19), supplements (18), beauty (17), apps (11), and food and beverage (11) come next.
SaaS, finance and telehealth together account for 19 results, many of them Flighted lead-gen accounts that switched their optimization event to qualified leads.
See apparel & fashion ads for the biggest single vertical.
CPA is the metric most often reported as a percentage change.
In the 49 results that report one, the cuts run from 5.6% (DFS) to 88% (Natural Acne Clinic, vendor-reported by Madgicx), with a median of 33%.
In the 42 results that report a ROAS gain as a percentage, the median is about +55%.
We left out headlines where the percentage referred to Google or to spend.
Absolute numbers vary too much to compare across accounts: the same hub has a $8.43 cost per purchase and a $385.80 cost per conversion.
For typical costs, see what Meta ads cost.
No result in this hub is tagged as a loss, and only two are mixed.
Obvi's GLP-1 angle ran 123 ads and reached only ROAS 1.06 at a $58.96 CPA.
A UK merch store hit ROAS 2.03 but spent £700+ in countries where nobody bought.
Some small results are worth noting. Meta's in-app product browsing study found +3% conversions across about 117,700 advertisers.
NP Digital's chart showed CPA only 6.8% lower with 20+ ads per campaign than with 1–2 ads.
One snapshot shows a ROAS of 51,287 that is just an attribution quirk on $0.02 of spend.
Expect smaller gains than the median when your account is already well structured, the same change has already been made, or the result came from a holiday period.
The patterns that repeat across the winners: fresh angles in a tired account, fewer and broader ad sets, an optimization event closer to revenue, and offers tested with the creative held fixed.
Test one of these at a time, keep the budget flat, and compare against the same weeks rather than a holiday.
Our creative testing guide covers setup and how long to run a test.
Yes. This hub has 247 Meta results with their credited sources.
The 49 that report a CPA cut as a percentage have a median drop of 33%.
Only 26 are split tests or multi-account studies, and 96 are vendor-reported.
Audiences are the most common lever (82 results), followed by campaign structure (58), budgets (47), creative volume (44) and measurement (44).
Many results changed several of these at once.
General ecommerce (57) and apparel (46) are the two biggest groups.
Typical ecommerce wins come from catalog template changes, consolidating into Advantage+ shopping, and offer tests such as a €10 price cut that dropped CPA 45% with the same ads.
Treat them with caution. Half are before/after comparisons, 39% come from vendors describing their own tools, and only 2 of 247 report a mixed result.
Use them to choose what to test, not to forecast your own results.