Case study · Mobile games UA

Tremex Games: 16 Google campaigns cut to 4, and ROAS was back at its June level in one month

Cooking Rage's Google D30 ROAS slid from 66.4% in June 2024 to 47.1% in September. In October, the first month of the new structure, it was 66.3%. Then I planned Cooking Rage's iOS launch, and later helped prepare Cooking Blast's.

Client
Tremex Games
Games
Cooking Rage, Cooking Blast
Model
Free to play, ads and in-app purchases
My role
UA lead, hands-on on Google and Meta, plus measurement
Engagement
Three rounds of work over two years
Results window
Google, Android, Sep vs Oct 2024
47.1% to66.3%
Google D30 ROAS
+19.2 points. Before: 47.1% in September
31.9% to40.4%
Google D7 ROAS
+8.5 points. Before: 31.9% in September
16 to4
Campaigns carrying the spend
The 4 new campaigns took 91% of October's Google budget

September vs October 2024, all Cooking Rage Google campaigns. ROAS counts ad and in-app revenue. The budget was smaller in October. How each number is measured is at the bottom of the page.

The challenge

A Google account that kept growing, while its return kept sliding

Cooking Rage is a cooking time-management game. It earns from ads and from in-app purchases. By September 2024 its Google account had 16 campaigns spending, one per region tier and goal, and D30 ROAS had dropped three months in a row.

The goal was simple: get D7 ROAS back up, and do it on a structure that could scale again.

My role

I ran UA hands-on. Gabriel Dumitru and the Tremex team owned the game, the budget and the creatives.

  • Rebuilt and ran Google App Campaigns on Android.
  • Launched Meta as the second channel.
  • Planned the iOS launch: Apple Search Ads, SKAN and Singular setup.
  • Fixed the revenue data the ad networks learn from.
  • Weekly calls with the founders, and advice on a third title in development.
What I covered
UA strategyGoogle App CampaignsMeta Advantage+Apple Search AdsUnity AdsSingularSKAN conversion schemaROAS goals on ad + IAP revenueiOS launch planPlayer-review analysisPlayable ads partner
Two years, two games

They kept coming back

Sep to Nov 2024

Cooking Rage, Android

Took over UA on Google, rebuilt the account and launched Meta.

The results on this page
Jan 2025

Cooking Rage, iOS

Apple Search Ads plan, SKAN and Singular setup, and a Google campaign held until the revenue data was right.

Setup
Jul to Oct 2026

Cooking Blast

iOS launch plan, the iOS conversion schema in Singular, a Unity Ads revenue fix and a player-review analysis.

Setup and advice
Results

Google ROAS by month, June to December 2024

Before the rebuildNew structure, under my contractAfter the contract ended

Month (2024)Google D30 ROASGoogle D7 ROASCampaigns spending
June66.4%41.8%not pulled
July62.5%39.7%not pulled
August54.5%34.4%11
September47.1%31.9%16
October (new structure)66.3%40.4%7 (4 new carry 91% of spend)
November52.5%31.2%7
December (after contract)54.4%37.7%5

The budget was smaller in October and November while the new structure learned. November gave back part of the gain as spend rose again. A month-on-month comparison, not a controlled test.

"Uri did an awesome job helping us with managing the UA side for our project. He managed to increase the ROI on GoogleAds with more than 10% compared to what we usually had."

Gabriel Dumitru, Tremex Games. Cooking Rage UA, 2024

"He understands performance marketing very well and more importantly, he tries to understand what is actually driving the results."

Gabriel Dumitru, Tremex Games. Cooking Rage iOS setup, 2025
What I did

Four moves

The options
  • Keep the 16 and tune bids tier by tier.
  • Cut the tiers that missed the target.
  • Rebuild into a few campaigns with more data each.
The call

Rebuild. Four campaigns: US, the rest of the world, and tier-one countries, each on a D7 ROAS goal that counts ad and in-app revenue together, with a "cooking game lovers" audience signal and only the top creatives.

Why

Sixteen campaigns split one budget so thin that most of them never got enough conversions to learn a ROAS goal. And a game that earns from ads loses most of its signal when the goal only counts purchases.

Google campaigns spending
US ad ROAS D7 (18 Sep)
US total ROAS D7
US CPA reward
US ad ROAS D7 (26 Sep)
UK CA AU EU T1 total ROAS
All T1 ad ROAS D14
WW ad ROAS D7
EU T2 total ROAS
EU T3 total ROAS
Asia T2 total ROAS
Asia T3 total ROAS
Asia T3 ad ROAS D7
Asia T3 ad ROAS D14
AML total ROAS
AML total ROAS v2
Africa total ROAS
Rest of world: IAP + ad ROAS D7
US: IAP + ad ROAS D7
Tier one: ad ROAS D7
US: total ROAS D7
16campaigns spending
The biggest US campaign, D30 ROAS
Sep, old setup
30.1%
Oct, new setup
78.7%

Real campaign names from Singular, shortened. The US bars compare one campaign each month, the largest US campaign by spend.

The call

Meta became the second channel from 21 October 2024. It started on purchase and rewarded-ad events, then moved to value optimization on ad and in-app revenue together from 2 November.

Why

An install goal finds people who install. For a game that earns from ads, the goal has to see ad revenue, or Meta finds the wrong players.

What the data said

D30 ROAS went from 30.3% in its first month to 40.5% in its second. Still below Google: a launch, not a finished result.

What Meta was told to find
Installsthe usual start
A purchase21 Oct
Rewarded-ad views25 Oct
Value: ad + IAP revenue2 Nov

Diagram of the setup, not values. Dates from the campaign names.

The call
  • Apple Search Ads: exact match only, brand and competitor keywords in separate ad groups, Search Match off, custom product pages, a small US test first.
  • Google: the new campaign stayed paused until Singular received the ad revenue. Only then a ROAS goal.
Why

A ROAS campaign can only learn from the revenue it sees. If the revenue isn't arriving, every euro trains it on the wrong users.

Apple Search Ads structure
Brand keywords
→
Own ad group, exact
Competitor keywords
→
Own ad group, exact
Search Match
→
Off
Google, in order
Campaign built, paused
→
Ad revenue reaches Singular
→
Turn on, then ROAS goal

Setup only. No result is claimed for this round.

The call

Before the iOS launch: a channel plan, the iOS conversion schema set up in Singular, and a read of what players said in their reviews. On Android, Tremex had started Unity Ads and Unity showed $0 revenue. I switched Singular to send all revenue events, ads and purchases, mapped to purchase, so Unity could optimize for total ROAS.

Why

Same rule as on Cooking Rage: the network can only find valuable players if it sees where the money comes from.

I also introduced a playable-ads studio, and advised the founders on a third title in development.

What Unity Ads could see
Purchases
sent as revenue
Ad revenue
sent as revenue
Unity goal
total ROAS
  • iOS launch plan
  • iOS conversion schema in Singular
  • Player-review analysis
  • Playable-ads partner intro

Setup and advice. No result is claimed: the iOS launch was moved by Tremex while the game team worked on product KPIs.

Key learnings

What holds for any mobile game

1

Fewer campaigns, more signal.

A ROAS goal needs enough conversions per campaign to learn. Split by every region tier and none of them get there.

2

On a hybrid game, optimize on total revenue.

If ads pay a big share of the bills, a goal that only counts purchases points the network at the wrong players.

3

Fix the revenue pipe before the first euro.

Check what your MMP sends to each network. A campaign learns from the data it gets, right or wrong.

4

Read ROAS next to the budget.

A jump on a smaller budget proves the structure. Holding it as spend grows is the next test.

How the numbers are measured
Google D30 and D7 ROAS
Revenue from the players acquired in a month, counted to their day 30 (or day 7) after install, divided by Google's cost for that month. Revenue is total revenue: ad revenue plus in-app purchases. All Cooking Rage Google campaigns, Android, every campaign included.
Points, not percent
"+19.2 points" means 47.1% became 66.3%. As a relative change that is +41%.
Campaigns spending, 16 to 4
Campaigns with Google cost in the month: 16 in September 2024, 7 in October. In October the 4 new campaigns carried 91% of Google's cost; 3 old ones ran out the month on the rest.
Biggest US campaign
The largest US campaign by cost each month: the old US ad ROAS D7 campaign in September, the new US campaign in October.
Meta D30 ROAS
Same definition, Meta's cost, October and November 2024.
What this is not
A month-on-month comparison, not a controlled test. The budget was smaller in October and November while the new structure learned, and November gave back part of the gain as spend rose. Budget figures stay off this page.
Gabriel's "more than 10%"
His own words. It is not a Singular figure.
Source
Singular, Tremex Games' measurement platform, read on 9 October 2026.