AI creative production performance: What can we learn from it

Claire Rozain is CEO of RZain Consulting.

We talk a lot about AI-generating creatives. But the real opportunity? Teaching AI - and our teams - to understand what works, and why.

AI for now is really fragmented, and I created a new framework to automate creative production and leverage User acquisition; here is what I learnt from it.

After analysing the 2025 AppsFlyer Creative Optimisation Report (based on 1.1M creatives and $2.4B in spend), I’m convinced: If we want better performance, we don’t just need more content - we need better learning automated AI loops.

AI isn’t just a production tool - it’s a feedback engine

AI has made it easy to mass-produce creatives. But that only works if we train it with the right signals. As a result, the top 2% of creatives only get 43% of the spend, showing a much higher adoption rate of diversification against ad fatigue. 

So how do we build a system - whether manual or AI-assisted - that actually learns? My lesson from launching hundreds of AI ads produced content by the Rzain algorithm. 

Lesson 1: Train your AI (and team) on outcome-based learning

Top 2% of creatives in Gaming = 53% of spend. In Non-Gaming = 43%.Source: AppsFlyer Creative report

That’s not just about winners - it’s a signal. My app clients experiment with an average of 20 concepts per week, while gaming clients still rely on existing best-performing concepts and segments. 

Gaming optimises for short-term performance. Non-gaming spreads risk, tests more and expands the audience to reach scale with in-depth creative testing mechanisms, allowing more experimentation to win a share of voice in the market.

As a result, app hook and hold rates are, on average, 15% higher than gaming companies across my clients. 

Takeaway: Tag and analyse not just your winners but your “long tail” creatives. Many underfunded ads show high retention - meaning your AI (or team) might be overlooking quiet champions.

If a creative is not pushed on the algorithm but has good performance - this means you can improve your hook or hold rate in order to improve your IPM and push this creative toward a best-performing concept.

Lesson 2: Don’t feed AI with just visuals - feed it motivations

“Instant gratification” drives installs in Finance

“Serious relationships” drive retention in Dating

“Failure-to-success” stories beat “Pure Success” in Casual Games

AI can now detect colors, hooks, and layouts. But what separates high-retention creatives is motivation alignment. Building your creative to appeal in the first 3 sec.

Different motivators are key to achieving your goal. Some top practices can be using a UGC influencer, asking a question, or directly communicating with the user in the first three seconds.

If AI can do this, it means the network creative algorithm can use your creative as a targeting object appealing to the right user, depending on the hook. 

Takeaway: Start labelling your creatives not just by format but by emotional trigger. Train models (and humans!) to spot why something works - not just what it looks like.

Lesson 3: Creativity needs context - platforms matter

For Casino games according to Appsflyer Creative report: 

• On Ad Networks, “Power” wins.

• On Social platforms? It’s “Completion” drives an IPM lift.

Same creative. Different results. Why?

Because context is content, and content is your new target for accessing inventory.

This is where AI falls short - unless we build it in.

Takeaway: Train your teams (and scripts) to analyse by platform. A TikTok winner might flop on YouTube. Build channel-level feedback into your AI loop.

Ask yourself why it worked and how to evolve this concept to prevent ad fatigue and frequency capping.  

Lesson 4: AI needs creative feedback - and so do we

At RZain Consulting, I use an internal script to identify top-performing ad structures from competitors - then test and optimise format by channel with AppsFlyer’s Creative Optimisation Suite.

It’s not about automating everything - it’s about creating smarter feedback loops.

Here’s what I recommend:

• Use SoC (share of cost) + IPM + D7 retention together and use different weights depending on your campaign type.

• Label creatives with emotional and behavioural drivers

• Test UGC vs. polished content- and track by segment

• Track all IEC across the channels with AppsFlyer's creative solution.

•  Let data guide production - not replace storytelling

The new creative feedback loop

1. Generate at scale (AI or otherwise)

2. Tag and track by motivator, channel, format

3. Test across touchpoints (Meta vs. DSP vs. UAC)

4. Learn from signals beyond CTR: IPM + ROAS + Retention

5. Create again - but smarter