Framed AI beats human creative on conversion: the Currys case

A real client case, measured over three months, shows AI-generated creative clearly outperforming human selection on conversion. The number is brutal. The explanation less so.

Hard evidence on AI creative is rare outside of sales promises. The test run by retailer Currys with Movable Ink, on email and CRM personalization, provides some.

The measured results

After switching from manual visual selection to AI-generated creative, over a three-month test:

• +49% click-through rate
• +30% conversions
• +37% revenue per send

The direct AI versus human duel

More striking still: in a direct A/B test, AI against human selection, AI achieves 1.8 times the conversion rate and 2.1 times the revenue per send. On this specific ground, it is not close.

An honest creative should be able to look at that number rather than wave it away. So, has the machine won?

It is not AI beating humans. It is personalization at scale beating single selection, and only a machine can produce it.

The nuance everyone misses

No, for a precise reason. What this test measures is not creative quality: it is the ability to adapt a visual to each recipient. A human picks one visual for everyone. AI produces one per segment. On a personalization channel, the advantage is structural.

And above all: these results do not come from an AI let loose. They come from a framed setup, with defined segmentation, brand constraints and validated templates. The framing produces the result, not the AI. Remove the frame and you get generic content that sinks the numbers.

The lesson to draw

This case does not say creatives are replaceable. It says where each belongs: humans design the system, define the rules and templates, the machine declines at scale. That is exactly the shift from producer to orchestrator, and this test finally gives it numbers.

Frequently asked questions

Does AI outperform creatives?

On personalization at scale, yes, because it declines one visual per segment where a human picks one for all. On designing the system and the rules, no.

Can these results be reproduced without framing?

No. The gains come from a framed setup: segmentation, brand constraints, validated templates. Without a frame, AI produces generic work that lowers the numbers.

Are these figures verifiable?

They come from a client case documented by Movable Ink over a three-month test, with a direct A/B. That is more solid than most of the market's claims.

Florent DABERNAT · Art director and founder of IDSEED, based in Aix-en-Provence. I help my clients with branding, UX/UI and web, using a clear and documented method. Learn more ➞