Sefinni
We moved budget out of the campaign with the best ROAS, and both channels came out ahead
12,874,353 RON
Attributed revenue
+46.8%, both platforms
+31.5%
Order volume
31,160 → 40,968
8.88x
ROAS on the added budget
every extra RON returned 8.88 RON
11.16x
Performance Max ROAS
up from 10.27x, on a 16.2% smaller budget
This case study was produced by the performance marketing agency BigConvert, based on data from the accounts we manage.
The data was checked and validated by the team at Sefinni.
- Position
- Leader in its category
- Services
- Google Ads, Meta Ads, Shopping
- Market
- Romania
- Period
- May 23, 2025 - Aug 13, 2026
Share of the Google budget, by campaign type
Google · before and after the reallocation.
At takeover
After reallocation
- Performance Max 10.27x → 11.16x
- Shopping 9.83x → 10.50x
- Search 19.26x → 14.37x
That's how much of the Google budget ran through a single campaign type at takeover. It wasn't a weak campaign - it was running at 10.27x. The problem was that nothing else ever got enough money to prove anything.
01 / The ceiling
It wasn't a performance problem, it was a distribution problem
We took over both accounts in May 2025. Neither was broken: together they were running at 10.59x. But an account that's doing well doesn't show you where its limit is - and here the limit was structural.
-
Concentration
78.2%
Of the budget, through a single campaign type
A campaign type that holds more than three quarters of the budget is a ceiling, however good its average looks.
-
Blocked channel
4.0%
What Shopping got of the Google budget
Too little to count as a test, let alone a channel. It had no way of proving anything.
-
Starting base
10.59x
Blended ROAS at takeover
We didn't start from zero: the account was already producing 31,160 orders in the previous window.
The decision that looks wrong in a table
We took money out of the more efficient campaign and put it into the less efficient one.
Where we took it from
5.70x
Performance Max
Revenue lost for every RON pulled out - far below the campaign's 10.27x average.
Budget share: 78.2% → 42.9%
Where we put it
10.55x
Shopping
Revenue produced by every RON added - almost double the place it was taken from.
Budget share: 4.0% → 36.7%
The difference between these two numbers is the whole argument. It doesn't show up in any standard report, because no standard report calculates the marginal return.
02 / The lever
The argument isn't the average, it's the marginal return
The average tells you how much a campaign's whole budget produced - a number about the past. The marginal return tells you how much the next RON produces.
We cut the better campaign's budget by 16.2% and moved it into the weaker one, which grew roughly fourteen-fold. Both came out ahead.
ROAS by campaign type, before and after the reallocation
Empty circle = before. Solid dot = after.Small value = before. Large, coloured value = after.
03 / Google
The numbers, straight from the account
The Google account generated 7,069,399 RON, 49.1% more, and every RON added produced 11.04 RON. Shopping revenue climbed from 155,864 to 2,354,101 RON.
| Campaign type | Budget share | Previous revenue | 2026 revenue | ROAS |
|---|---|---|---|---|
| Shopping | 4.0% → 36.7% | 155,864 RON | 2,354,101 RON | 9.83x → 10.50x |
| Performance Max | 78.2% → 42.9% | 3,213,256 RON | 2,923,947 RON | 10.27x → 11.16x |
| Search | 17.8% → 20.4% | 1,373,813 RON | 1,791,351 RON | 19.26x → 14.37x |
| Metric | Previous period | Period analyzed | Change |
|---|---|---|---|
| Attributed revenue | 4,742,932 RON | 7,069,399 RON | +49.1% |
| Budget | confidential | confidential | +52.7% |
| ROAS | 11.86x | 11.58x | −2.4% |
| Conversions | 16,981 | 22,480 | +32.4% |
| Clicks | 474,573 | 666,902 | +40.5% |
| Impressions | 29,030,481 | 35,080,496 | +20.8% |
| ROAS on the added budget | - | 11.04x | - |
Google Ads · account screenshot
04 / New demand
Brand budget went down, non-brand budget grew tenfold
Separating brand from non-brand shows whether an account brings in new customers or buys its own demand. Here the two moved in opposite directions, on purpose.
Share of Search revenue
Brand · budget −19.3%
Revenue from 1,290,343 to 1,048,112 RON, ROAS unchanged: 20.26x → 20.39x. Brand demand doesn't need more money, it just needs not to be missed.
Non-brand · revenue +790.4%
Revenue from 83,470 to 743,239 RON, with ROAS going from 10.92x to 10.15x. The only place in the account that buys demand that didn't already exist.
05 / Meta
Volume went up, the ratio went down
On Meta we scaled more aggressively and paid for it in efficiency: budget +58.5%, revenue +44.1%, ROAS from 9.40x to 8.54x. It is lower than where we started, and that needs to be said before anything else.
Impressions grew by 40.1%, reach by only 9.1%. The account talked more often to the same people - from 4.5 to 5.8 impressions per person - which is why efficiency dropped.
| Metric | Previous period | Period analyzed | Change |
|---|---|---|---|
| Attributed revenue | 4,029,772 RON | 5,804,954 RON | +44.1% |
| Budget | confidential | confidential | +58.5% |
| ROAS | 9.40x | 8.54x | −9.1% |
| Purchases | 14,179 | 18,488 | +30.4% |
| Link clicks | 689,282 | 1,096,016 | +59.0% |
| Impressions | 38,481,642 | 53,903,013 | +40.1% |
| Reach | 8,544,462 | 9,324,150 | +9.1% |
| ROAS on the added budget | - | 7.08x | - |
Meta Ads Manager · account screenshot
06 / Results
The blended picture, and what we gave up for it
Blended efficiency went down. It is not an efficiency gain and we don't present it as one.
Budget and revenue, indexed to 100
ROAS points, from 10.59x to 9.98x. The budget grew faster than revenue, and the distance between the lines is the cost of the decision, not its gain.
| Metric | Previous period | Period analyzed | Change |
|---|---|---|---|
| Attributed revenue | 8,772,704 RON | 12,874,353 RON | +46.8% |
| Budget | confidential | confidential | +55.7% |
| ROAS | 10.59x | 9.98x | −5.8% |
| Order volume | 31,160 | 40,968 | +31.5% |
| ROAS on the added budget | - | 8.88x | - |
-
What we got
+9,808
More orders and 4.1 million RON in revenue
From 31,160 to 40,968 orders, i.e. +31.5%.
-
What we gave
−0.61
ROAS points, from 10.59x to 9.98x
At almost 10x, volume is worth more than the ratio - but it is a choice, not an automatic outcome.
07 / The system
What can be repeated
-
Phase 01
You look for concentration
The first thing checked isn't performance, it's how the budget is distributed.
-
Phase 02
You judge on the marginal return
You compare what you lose by pulling money out of one place with what you gain by putting it in another.
-
Phase 03
You agree the trade in advance
If scaling requires a drop in the ratio, the number is agreed beforehand, not explained afterwards.
The average tells you what you did. The marginal return tells you what to do tomorrow.
08 / Limits
What we don't take credit for
-
Basis of comparison
18.5%
Of the previous window already falls after the takeover
The period analyzed is longer than a year, so the base isn't entirely prior to our collaboration.
-
Would have come anyway
14.8%
Of Google revenue came from brand
1,048,112 RON at 20.39x, on demand that already existed and on a budget that went down.
-
What the numbers are not
Revenue, not profit
Both accounts run on classic ROAS
All amounts are revenue reported by the platforms, not margin and not profit.
See more about the methodology
The period, the sources, the units of measurement, the other two limits and what can't be added together.
The other two limits
- Both accounts were already working. We didn't start anything from zero: at takeover, the blended account was running at 10.59x and producing 31,160 orders in the previous window. Growth is measured from a base that was already working.
- We can't separate seasonality and category growth from our own work. The window analyzed is longer than a year and contains two seasonal peaks. Platform exports don't contain the information needed to isolate this effect.
Methodology and data notes
- Period analyzed: May 23, 2025 - August 13, 2026, from the account takeover to three days before this was written, compared with May 23, 2024 - August 13, 2025. The two windows are the same length, 448 days.
- Overlap: because the period analyzed is longer than a year, the comparison window overlaps with the first 83 days of the collaboration, i.e. 18.5% of it. The basis of comparison isn't entirely prior to the takeover.
- Source: campaign-level exports from Google Ads and Meta Ads. Google totals are calculated exclusively from campaign rows, after removing 6 Total rows, which would otherwise have multiplied the account.
- The currency is the Romanian leu (RON) on both platforms.
- Conversion value: both accounts run on classic ROAS, so conversion value is sales value. All amounts in this document are revenue reported by the platforms; no claim is made about margin or profit, because platform exports don't contain that information.
- Budgets: at the client's request, absolute figures for cost, cost per result and CPM are not published. Percentage changes in budget are.
- Comparability: the comparison is made within each platform, on the same account, the same calendar window and the same attribution model in both periods (Meta: 7-day click or 1-day view), so methodology differences between Google and Meta don't affect the growth percentages.
- Attribution: because both platforms can attribute the same order, the blended total expresses revenue reported by the platforms, not the store's net revenue. Blended growth remains valid, since the attribution overlap exists in both periods compared.
- Blended volume: adds up the conversions reported by Google and the purchases reported by Meta, two different counting systems. Average order value is practically identical across platforms, 314 RON on Google and 314 RON on Meta, which confirms that the two count comparable events.
- Exclusions: the Display channel was excluded from the channel analysis, as it recorded no spend in either period.
- Screenshots: the platform screenshots are presented as proof of the data in the tables. The cost columns are covered, at the client's request.
- Limitation: the results reflect one specific account, category and period. They are not a promise of performance.
