-66%
cost per subscription
Plus 162% more subscriptions for FET across paid social.
Read the FET case study
Mobile UA framework
The Admiral Performance Loop™
Five phases that feed each other: signal, creative, audience, forecast and scale. One system instead of five separate reports.
4.9 on Clutch from 48 reviews€500M+ ad spend managed150+ brands grown
Why a loop
Most UA programs run in five lanes.
The Admiral Performance Loop is a closed-loop UA framework. It connects signal capture, creative testing, audience modeling, ROAS forecasting and scale decisions into one compounding system.
Five workstreams
- Creative is briefed by one team
- Audiences are set by another
- ROAS arrives in a monthly report, already 30 days stale
Learnings from scale never reach creative, and each launch starts from the same assumptions as the last.
One loop
- Signal→
- Creative→
- Audience→
- Forecast→
- Scale→
- Signal, sharper
Phase 5 output is Phase 1 input. Each scale decision adds cohort data, which sharpens LTV curves, which improves audience seeds, which improves the signal creative is judged on.
The five phases
How the Loop runs in practice.
Pick a phase. Each one is the actual sequence of work, from onboarding to steady-state management.
Phase 1: Signal Capture
Measure what actually pays.
You cannot optimise against a signal you are not measuring accurately. This phase runs before launch and sets the quality ceiling for everything after it.
- Attribution setupPostback windows matched to your monetisation curve, not the default D7. Set up through AppsFlyer, Adjust or Singular.
- Cohort events as targetsD1, D7 and D30 retention and revenue events flow back to Meta, Google, TikTok and Apple Ads, so platforms learn user quality, not install volume.
- Creative taggingEvery asset is tagged by format, hook type and audience angle, so learnings carry across campaigns and account managers.
- Why it comes firstThe common failure is not weak creative. It is scaling a campaign that delivers cheap installs from low-value users.
Phase 2: Creative Velocity
Out-test, do not out-guess.
Platforms have converged on similar optimisation logic. What separates accounts at scale is the quality and volume of creative inputs, so we build a system to produce and test them fast.
- Testing volumeAt least 4 to 6 new variants a week per active account, tested across hook, format and call to action. One winning hook can grow into 8 format variants.
- Kill thresholdA creative that misses the account's IPM benchmark within 5 to 7 days is paused, not iterated. Iterating before proof burns production capacity.
- Concept familiesA creative that clears the benchmark with positive D7 signals grows into 5 to 8 variants across format, length, call to action and audience angle.
- Fatigue signalsWatched per creative: CTR down more than 25% week over week, or IPM down more than 20% with no change in bid, audience or placement.
Phase 3: Audience Modeling
Seed from your best users.
A lookalike is only as good as its seed. Seeding from all installs imports your worst segments, a flaw that stays invisible in early metrics and compounds over time.
- High-value seedingLookalikes are seeded from D7 payers and D30 retained users. A seed needs at least 1,000 qualifying users before launch.
- Exclusion logicProspecting suppresses users who already have the app, churned users and low-LTV cohorts. Re-acquiring them is waste.
- Broad in parallelNarrow 1 to 2% lookalikes run next to broad delivery for scale and discovery. Broad typically takes 60 to 70% of the split after 14 or more days of cohort data.
- Same logic per platformOn Meta these are Custom Audiences built from purchase events with a D7 window. TikTok gets the equivalent event-based audiences.
Phase 4: ROAS Forecasting
Forecast payback, do not react to D7.
A campaign at 40% D7 ROAS can be profitable or deeply unprofitable, depending on the shape of its LTV curve. Forecasting turns early signals into forward-looking bid and budget decisions.
- LTV curve calibrationNew accounts start from genre benchmark curves for D7 to D90. Account cohort data replaces the assumptions as it matures, so the model keeps updating.
- Per-channel paybackEach channel brings users with a different LTV profile, so each gets its own ROAS target instead of one number for all.
- Predictive biddingWhere the platform supports it, such as Meta value optimisation or Google tROAS, we bid toward predicted D30 revenue, fed by event-level postbacks from Phase 1.
Payback check
An illustration, not a prediction. Real curves differ by genre, platform and geo.
100%projected D30 ROAS
RelatedUA agency pricing
Phase 5: Scale Architecture
Scale without paying for it twice.
Unstructured scale means CPI inflation, creative fatigue and dependence on one platform. Scale Architecture is the rule set that prevents all three.
- Diversification triggersIf the main channel's CPI rises more than 20% week over week without better LTV signals, budget moves to secondary channels. A rule, not a judgement call.
- Platform sequencingMeta and TikTok first. Apple Ads for iOS once brand search and category benchmarks are validated. Google for Android once Meta D30 data can calibrate tROAS. DSPs such as Moloco, Unity Ads and Digital Turbine add reach.
- Geo layeringThree tiers with separate CPI and LTV models: Tier 1 (US, UK, AU), Tier 2 (DE, FR, JP, CA) and Tier 3 (LATAM, SEA, MEA).
- Creative-level budgetsWinners get more budget within 48 to 72 hours of reaching significance. Underperformers pause on the same clock.
Where the Loop has run
Results you can read in full.
Three published case studies from app categories we scale every day.
117%
ROAS on Day 30
Up 3x on Android and 190% on iOS for a dating app scaled on Moloco.
Read the Moloco case study-74%
cost per install
PURE lowered CPI and beat its ROAS goals by scaling through Moloco.
Read the PURE case studyFigures as published in each case study. See all case studies.
Quick check
Find where your loop leaks.
Tick what is true for your account today.
0
Tick what is true for you.
Five yes or no questions, one per phase. Nothing is stored.
Where it fits
One structure, four categories.
The Loop is the operating method on every engagement. The setup of each phase changes with the app.
Mobile gamesCasual, midcore, hypercasual and hardcore. The phase setup changes by genre.See our approach Subscription appsPayback modeling is central because subscription LTV curves are long.See our approach Fintech and utilityD7 events are shallow, so D30 to D90 LTV modeling carries the bid decisions.See our approach E-commerceCreative velocity and audience suppression are the highest-leverage phases.See our approach
FAQ
The Performance Loop, answered.
What is the Admiral Performance Loop?
The Admiral Performance Loop is Admiral Media’s closed-loop mobile user acquisition framework. It connects signal capture, creative velocity, audience modeling, ROAS forecasting and scale architecture into one system where each phase feeds the next.
What are the five phases of the Performance Loop?
Phase 1 is Signal Capture, the measurement foundation. Phase 2 is Creative Velocity, structured creative testing. Phase 3 is Audience Modeling, lookalikes seeded from high-value cohorts. Phase 4 is ROAS Forecasting, LTV curves and payback targets per channel. Phase 5 is Scale Architecture, rule-based channel, geo and budget expansion.
How is the Loop different from running these workstreams separately?
Most UA programs brief creative, set audiences and report ROAS in separate cycles, so learnings from scale decisions never reach creative strategy. In the Loop, Phase 5 output feeds Phase 1 input, and every cycle starts from what the last one learned.
Which apps is the Performance Loop built for?
It is the working method across Admiral engagements: mobile games, subscription apps, fintech and utility apps and mobile-first e-commerce. The setup of each phase changes by category.
Which channels does the Loop cover?
Meta, TikTok, Google, Apple Ads and DSPs such as Moloco, Unity Ads and Digital Turbine. Channels are sequenced by data volume: Meta and TikTok first, Apple Ads for iOS, Google for Android, then DSPs for incremental reach.
Which attribution tools does Admiral work with?
AppsFlyer, Adjust or Singular, depending on the client’s existing stack. The key setup decision is postback window length, matched to the app’s actual monetisation curve.
How long does it take to see results?
Phase 1 happens before launch. After that it depends on the app’s monetisation curve: a D30 cohort needs 30 days to mature, and forecasts sharpen as account cohort data replaces benchmark assumptions.
How do I start?
Book a meeting with our team. We look at your attribution setup, creative output, audience structure and LTV modeling and show which phases are weakest. Our engagement and pricing page explains how we work and charge.
Book a meeting
Close the loop on your UA.
- Send your appYour store link, markets and goal. Two minutes.
- We read your enquiryA senior specialist reads it and replies with next steps.
- Talk it throughA call on which phases are weakest, what to fix first and what it costs.