27 paid-ad results, each credited and linked to the source.
Across the 196 results in this hub, the median CPA drop was 35%, counted from the 144 results that state a percentage or a before-and-after figure for CPA, CAC or cost per purchase.
Angle changes, format switches and landing-page changes had the biggest median drops; budget automation had the smallest.
Be aware that 114 of the 196 results (58%) are vendor-reported, and only 36 are controlled split tests.
To rank levers, we took every result with a computable CPA change and grouped it by the levers it is tagged with.
Many results change several things at once, so one result can count toward more than one lever.
Levers with fewer than 10 results are left out.
| Lever | Results with a CPA % | Median CPA drop |
|---|---|---|
| Angles | 14 | 54% |
| Format switch | 23 | 50% |
| Landing pages | 10 | 49% |
| Placements | 10 | 42% |
| Audiences | 37 | 41% |
| Campaign structure | 47 | 38% |
| Bidding | 20 | 38% |
| Measurement and signals | 30 | 33% |
| Offers | 10 | 31% |
| Creative volume | 31 | 31% |
| Visuals | 25 | 30% |
| Budgets | 21 | 22% |
Structure is the most common lever (47 results), and Meta is the most common platform (64 of the 144, median drop 33%), ahead of Google (34, median 38%) and TikTok (33, median 35%).
Three single-account Meta stories shared by @SamPiliero show how big a change one lever can make: a price cut with the same ads took CPA €170.34 → €93.51 (−45%), better ads on a flat budget moved Cost per order €84.38 → €58.63 (−30.5%), and fresh angles in a tired account took CPA $101.29 → $60.50 (−40%).
On bidding, @EcomRedneck reported Sales +328% in 5 days on cost caps, though that post gives no CPA figure.
For the full ranking across metrics, see change types ranked, or browse all results.
Creative was the most common way people cut CPA.
The biggest drops came from new angles, which means a new reason to buy rather than a new edit of the same ad.
AdQuantum's burnout angle for a mindfulness app reached a €23.85 CPA in its best funnel, against €57.15 in the earlier Meta campaign.
Hustler Marketing's "3 reasons why" UGC concept for Tao Clean cut CPA by 50%.
Browse angle & message changes and hook changes for more.
Testing more creative helps, but by less: creative-volume results have a median drop of 31%.
The cleanest example is a split test. Marpipe reports −77% CPA in one week for the top ad in an AgelessRx test of its current copy against 4 new value propositions (vendor-reported).
More in creative volume & testing cadence changes.
Only 10 results with a CPA figure change the offer, and 10 change the landing page, so treat these medians as rough.
Landing-page changes come out at 49%. In Admiral Media's App Store test for ImmoScout24, a product page matched to renter intent took CPA from €2.15 to €0.80. Offer changes come out at 31%.
Gill's Point S, a local tire shop, went from CPA $15.18 to $3.14 with a holiday offer on TikTok (vendor-reported by TikTok). See offer & price changes.
Account changes make up a large share of this hub, and most of them come from the platforms themselves.
In a Meta split test, adding Advantage+ shopping gave Jenny Bird 14% lower cost per purchase.
A Google test for Kia UK found value-based bidding produced a 10% cheaper CPA plus 20% more value than target CPA (Think with Google).
Both are vendor-reported. Of the agency stories, Flighted consolidated K-Republiq into one ASC campaign with cost caps and reported CPA $40 → $27. See campaign structure changes and bid strategy changes.
Budget automation gave the smallest median drop, 22%.
Smartly's split test for HungerStation reported CPA −15%.
Among the 144 results with a CPA percentage, the median drop is 35% and the middle half falls between 21% and 60%.
48 results claim a drop of 50% or more, and 29 report less than 20%.
How the result was measured matters. Split tests have a median of 30% (33 results), campaign reports 29% (28), and before/after comparisons 41% (82).
Vendor-reported results have a median of 30% (85), compared with 43% for independent ones (59).
Most of the independent stories in this hub are before/after comparisons (49 of all 82 independent results), and those can't separate the change from seasonality or rising spend.
Results that changed three or more levers at once have a median of 41%, versus 30% for single-lever results.
That bigger number comes partly from changing everything at once, which also makes it impossible to say which change did the work.
This hub has almost no losses: 195 of 196 results are marked as wins.
That is selection bias, because people publish their wins, so don't read it as a 99% success rate.
The one mixed result is Obvi's GLP-1 angle, which ran 123 ads at ROAS 1.06 and CPA $58.96, so tracked revenue roughly matched ad spend.
Some "wins" barely moved CPA. Nespresso's Performance Max launch reported CPA −3%, Dr. Squatch reported CAC −2.9%, and Catawiki's consolidation added conversions at the same cost of acquisition.
A Google AI Max pilot for UNIQA raised conversions 13% while CPA rose 3%.
These changes are least likely to work when CPA is already close to what your offer and margins allow.
Small-percentage wins like these also tend to disappear in noise without a holdout.
If you need help, here are agencies who publish results like these.
CPA is ad spend divided by conversions, but this hub mixes cost per purchase, CAC, cost per install and CPL, and they don't compare directly.
Pick one conversion event and one attribution window before you test, and keep them the same before and after.
Where you can, run a split or holdout rather than a before/after comparison, and track ROAS or contribution margin too, because a lower CPA on smaller orders is not a win. See the CPA formula and definition.
In this hub's 144 results with a CPA figure, the biggest median drops came from new angles (54%), format switches (50%) and landing-page changes (49%).
The overall median drop was 35%. Change one thing at a time, and use a split test if you can, so you know what caused the drop.
64 of the CPA results here are on Meta, with a median drop of 33%.
The repeat winners were new angles in tired accounts, better creative on a flat budget, offer or price changes, and consolidated Advantage+ structures.
Several of these are single-account before/after stories, so test them in your own account.
The results here point to tired angles, an offer that doesn't convert, landing pages that don't match the ad, and fragmented campaigns.
Check that you're optimizing for the right conversion event before you change creative.
A few results cut CPA by fixing measurement alone, with a median drop of 33%.
The median drop here is 35%, and the middle half of results falls between 21% and 60%.
Controlled split tests show a median of 30%.
Because published results lean toward wins, a 15-30% drop from one well-tested change is a sensible planning figure.