100+ video ad variations a month, not once a quarter.

Why creative volume is the real bottleneck now

Every performance channel now optimises delivery automatically. Meta, Google and AppLovin all spend a budget the way their own algorithm decides, not the way a media buyer manually schedules it. What is left for a team to control is the creative itself: which hooks, formats and angles get fed into that algorithm, and how fast a fatigued one gets replaced.

That shifts the constraint from media strategy to production capacity. A studio producing a handful of ads a month is handing the algorithm a small, ageing pool to optimise against. A studio that can ship dozens of fresh, on-brand variants every week keeps the pool alive, and the algorithm keeps finding new winners inside it.

Fatigue is the real cost, not the extra ads

A single ad set does not fail gradually. It works, then it saturates the audience that responds to it, and cost per result climbs fast. The fix is not one better ad. It is a standing supply of new ones, tested continuously, so a saturating ad already has a replacement queued before performance actually drops.

Manual production has a hard ceiling

A design team can turn out a strong ad. What it usually cannot do is turn out fifty strong ads a month at a cost that still makes sense against the media budget it is meant to serve. That ceiling, not creative talent, is what a production system is built to remove.

Inside the AI Creative Factory

Admiral Media’s AI Creative Factory is a closed loop, not a one-off generation tool. Each stage feeds the next, so output keeps improving instead of drifting off-brand as volume goes up.

A human strategist and QA pass sits on top of every stage. The system produces the volume; people decide what is on-brand and what ships. Formats span UGC-style, static and motion, spec’d for Meta, TikTok, YouTube and the other channels a given account runs.

The proof: Fastic and Star Chef 2

Two client results carry this page, both public case studies with named attribution.

Fastic: −70% cost per result

Fastic is a subscription fasting app that needs a constant supply of fresh hooks to keep its feed from fatiguing. The AI Creative Factory took over its ad production and cut cost per result 70%, at a volume no hand-built studio could match. The work won Silver in the AI category at The Drum Awards for Marketing EMEA 2026.

Star Chef 2: +45% ROAS, +55% CTR, −18% CAC

Star Chef 2, a cooking and restaurant management game from 99games, needed 60+ creative variations every week across multiple gameplay concepts and channels, a volume its in-house team could not sustain without risking the game’s visual style. The Factory ran a structured A/B testing matrix across those variations to isolate winning hooks, visuals and formats, and fed results back into the next batch.

The volume, in real numbers

These are Admiral’s own production numbers, not industry averages.

The system is trained on 500M+ ad performance data points, so new variants start from what has already worked rather than a blank brief. Plans run from 20 video ads a batch up to 80, all one month rolling, so output scales with a brand’s media spend rather than locking it into a fixed retainer.

How to evaluate a creative-at-scale partner

Volume alone is not the pitch. A partner that generates hundreds of variants with no discipline around brand, testing or reporting just produces noise faster. Ask about the following before you sign anything.

Testing cadence

How many new creatives actually enter live testing each week, and what triggers a variant getting pulled? A partner should have a specific number and a specific fatigue signal, not a general promise of “a lot.”

Fatigue detection

Ask how they know a creative is fatiguing before cost per result visibly climbs. A reactive process waits for the number to move. A good one is already queuing the replacement.

Brand guardrails

Ask what stops AI output from drifting off-model as volume increases. There should be a named process, templates, prompt frameworks, human QA, not just an assurance that someone “checks it.”

Format coverage

Confirm the partner covers UGC-style, static and motion formats, spec’d correctly per channel. A partner that only does one format is solving part of the fatigue problem, not all of it.

The feedback loop

Ask whether performance data from this week’s batch actually changes what gets generated next week, or whether every batch starts from the same brief. The loop is what compounds results over months instead of resetting each time.

Creative at scale, answered

How many video ad variations should we be testing a month?

It depends on spend and channel mix, but as a floor, a brand actively scaling paid media should have new creative entering testing every week, not once a quarter. Admiral’s own batches run from 20 video ads up to 80 a month depending on plan, with 100 to 150 variants generated per creative cycle to select from.

Does AI creative actually outperform manually produced UGC?

It outperforms it on speed and volume, which is where most testing programs are actually constrained. Star Chef 2 needed 60+ variations a week and could not sustain that by hand; running that volume through the AI Creative Factory delivered +45% ROAS, +55% CTR and −18% CAC. The comparison that matters is not one AI ad against one UGC ad, it is whether your testing program can run at the pace the algorithm needs.

How fast can a batch turn around?

Generation itself takes hours, not weeks, because it is not manual production. Turnaround on a full batch, brief through QA’d, ready-to-launch assets, depends on batch size and how much brand-guideline setup is needed up front.

What’s included in a batch?

AI-produced, human-refined, channel-ready creative built on the brand’s own guidelines and trained with competitor research, plus a dedicated creative strategist. Batches are one month rolling, so a brand can scale volume up or down as spend changes.

Does this work for game apps, or only subscription and ecommerce apps?

Both. Fastic is a subscription app; Star Chef 2 is a mobile game. The same generate, learn, lock-brand, compound loop runs for either, the difference is which hooks, formats and channels the creative strategist targets.

Is this fully automated, or does a person actually touch the output?

Both. AI produces the volume. A human strategist and QA pass sits on every stage, deciding what stays on-brand and what ships. The brand-rule layer is what prevents output from drifting as volume goes up, and that layer is set and checked by people.

How is creative fatigue actually detected?

By scoring every live variant against real performance data continuously, rather than waiting for a scheduled review. That is the “learn in real time” stage of the loop: a fatiguing ad’s replacement is already in the pipeline before cost per result visibly climbs.

How do we start?

Talk to the team about your current spend, channel mix and testing cadence. Admiral scopes a batch size against your actual media budget rather than a fixed package.