£3.5mAd spend analysed
344.4mImpressions
1,784Unique ecommerce ads
189Data points per ad

The short version

What did we find when we tested combinations, not single parts?

Most creative testing changes one thing at a time. The platform judges your whole ad at once. When we scored combinations instead of single parts, the results doubled.

  1. Creative is the biggest lever you have.

    It drives around half of the sales lift from advertising. Most marketers think it's closer to a fifth.

  2. Meta reads your whole ad, not its parts.

    Andromeda judges the hook, message, offer and format together. Testing one part at a time misses most of what's going on.

  3. Combinations beat single ideas.

    The best combination we found performed twice as well as the best single idea.

  4. Winners mix different parts of the ad.

    Every winning combination paired two different parts. None paired two versions of the same part.

  5. The hook is a multiplier.

    Hooks never won on their own, but they appeared in six of the eight winning combinations.

Signal in the noise

Why is creative still undervalued?

"Half the money I spend on advertising is wasted. The trouble is, I don't know which half."

Attributed to John Wanamaker, early 1900s

The data has caught up. Creative quality drives almost half of the sales lift from advertising. Most creative reports still grade ads one part at a time, which is too blurry to show what actually drives results.

49%

Share of sales lift driven by creative

Source: Nielsen / NCS

20%

What marketers think it drives

Source: Industry survey, 2024

Decoded

How does Meta read your ads now?

Before Andromeda

You picked the audience.

The platform delivered to it. Creative and targeting were two separate jobs.

Now

The platform reads your creative and picks the audience.

It weighs up tens of millions of ad variations for every auction. Meta says creative variety is now the main way to reach the right people, not niche targeting.

10,000×

Increase in the capacity of the models Meta uses to shortlist which ads to show

Source: Meta Engineering, Dec 2024

8%

Improvement in ad quality from Andromeda, in the segments Meta measured

Source: Meta Engineering, Dec 2024

Your creative strategy is now your targeting strategy.

The recipe

Which ad combinations won?

The best single idea reached up to 1.5× a brand's typical ad. The right combination reached 2× to 3×. These eight pairs beat the stronger of their two parts on its own.

Limited-time offer + Product demoOffer × Format
3.0×
Pattern interrupt + Value propositionHook × Angle
2.3×
Urgency + Percentage offAngle × Offer
2.4×
Relatable scenario + Value propositionHook × Angle
2.5×
Screen recording + Direct addressProduction × Hook
2.4×
Text hook + LifestyleHook × Format · Best evidence
2.2×
Screen recording + Text hookProduction × Hook · Best evidence
2.1×
Bold claim + AspirationHook × Angle
2.0×

Score is click-through rate compared with the brand's own typical ad. Pairs are listed in the paper's order, ranked by how far each beat its stronger half. Best evidence means the pairing was seen across the most accounts (4 to 5 each). These are patterns in real spend, not controlled experiments, so use them to decide what to test first.

Law one

Combining beats choosing.

The top pairing scored 3.0×, twice the best single idea.

Law two

The hook never won alone, but it's in three-quarters of the winners.

Hook6/8
Angle4/8
Offer2/8
Format2/8
Production2/8

Law three

Every winner mixed different parts. Eight out of eight.

Not one winning pair combined two ideas from the same part of the ad, yet that's how most creative variation gets made.

Five hooks on one angle isn't five ideas. It's one idea in five costumes.The Signal · Issue 1

Proof outside retail

Does this only work for ecommerce?

No. We ran the same engine on a sports app sold through video creators: 1,764 ads from 63 creators, with no products, no unboxing and no before-and-after.

20%

Cheaper per purchase when the ad opens with the viewer's problem

Held across 12 creators

10%

Cheaper per purchase with medium pacing instead of fast

Rapid edits grabbed attention, then lost it

Before · Bold claim

Three things you need to do…

After · Problem first

I've got a big doubt.
43%cheaper per purchase
48%more people stopped to watch

One line changed. Same creator, same product. The opening line is a hook, and once again the hook was the lever that moved.

Get the full paper

What else is in the full white paper?

This page is the summary. The full 19-page paper includes the method, the mistakes and a checklist your team can use before Q4.

  • How Creative IQ reads every ad, and the four checks every finding must pass 03 · Inside the engine
  • The three single ideas that beat the brand's own average 04 · Ingredients
  • The full combination grid, including the one pairing nobody has tested 05 · The recipe
  • Five mistakes that waste your budget 07 · Mistakes
  • Nine questions to ask before Q4 08 · Next move

Download The Signal · Issue 1

We'll send the PDF to your inbox.

Case Study

Brands we've helped grow

Futureproof your paid search campaigns

TALK TO US