Guide · Kaliber Creative Diversity Model
How to brief ideas that are genuinely different
Meta asks advertisers for creative that is genuinely different. Most accounts answer with new colours, crops and headlines on the same idea. This is how Kaliber tells a new idea from a new outfit, and how we brief a batch that gives Meta real choice.
- Why it matters
- What counts as a new idea
- Same idea or new idea
- How to brief a batch
- Idea types to draw from
Why it matters now: Andromeda
In December 2024 Meta described Andromeda, its ad-retrieval system. Before any auction, it shortlists which of all the ads on Meta each person could be shown. An ad that isn't shortlisted never competes, at any bid.
Meta's engineers write that “increased ad diversity can improve people's experience with ads and drive better advertiser outcomes.” More on Andromeda. What Meta doesn't publish is a way to tell whether your own ads are actually different from each other. The model below is Kaliber's answer. It is our way of reading creative, not a description of how Meta's systems represent ads.
Three things change in an ad. Only one creates a new idea.
| What changes | The question | What it adds |
|---|---|---|
| What it says | What does the ad claim, and who is it for? | A new creative identity, the count that matters |
| How it persuades | How is the claim made believable: a demo, a testimonial, a price? | A new identity only when it changes what the viewer is asked to believe |
| How it looks | Setting, presenter, visual style, format | Execution range, never a new identity on its own |
One distinct proposition: what an ad claims, and who it is for.
The same claim for the same buyer, shot in a bedroom and then on a beach, is one identity with more execution range. The same product pitched to a finance lead on risk and to a marketer on hours saved is two identities.
Execution range still has value. It keeps an idea fresh and lets it run in more places. It just doesn't give Meta a new idea to choose from. Twelve ads can be one idea in twelve outfits.
Same idea or new idea?
Merge: these are variants
- The same proposition applied to different products or models.
- The same offer on different occasions or amounts. A Mid-Autumn “save $2,500” and a September “save $1,600” are one idea.
- A new colour, headline, background, crop, format, setting, model or creator delivering the same proposition.
- Close persuasion styles making the same claim: one customer's testimonial and a “thousands of customers” ad arguing the same benefit.
Split: these are new ideas
- A different reason to buy: convenience, health, price, status, less risk.
- A different buyer or need: the finance lead on risk versus the marketer on time saved.
- A different kind of argument: showing the product working versus a price promotion versus a comparison with the alternatives.
Three checks
- Sanity check. If your count of ideas is close to your count of ads, you have counted variants as ideas. Group again.
- Tie-break. Still unsure after the rules? Count it as separate.
- Two readers. Grouping is a judgement. Have a second person group the same ads without seeing your answer, and compare. The Autopsy groups every account three times for the same reason.
How to brief a batch: explore, then exploit
Maximum variety is not the goal. Meta optimises for results, not for a balanced portfolio, so every batch has two jobs:
- Explore: genuinely new ideas, to find what works.
- Exploit: more depth around ideas already proven on your own results.
The rules we brief to
- Start from the diagnosis. Know how many ideas you run today, which one dominates, and what's missing.
- Split the batch. With no results yet, the whole batch explores. Once results show winners, a typical start is 60% explore, 40% exploit. Kaliber benchmark
- Keep new ideas distant. Each should differ in reason to buy, buyer or kind of argument. A new setting for an old claim is exploitation, not exploration.
- Around a proven idea, keep the proposition fixed. Vary the hook, execution and format, so a result can be read as the idea's result.
- Put promising ideas into more than one format, so they can run in more placements.
- Watch offer-led creative. If most of a batch leads with a discount, the brand teaches buyers that it is a price. Keep it a minority unless the offer is the strategy. Kaliber benchmark
- Write every idea's angle in eight words or fewer. If you can't, the ad is arguing two things. Split it or cut it.
- Count before you ship. Group the batch, ideally blind by two people, and check the count is clearly below the number of ads.
Idea types to draw from
When an account is short of ideas, these are where we start. Each one is a different kind of argument, so each can carry a new idea.
| Idea type | What it is |
|---|---|
| Customer testimonial | A real client says why they chose you. |
| Founder or expert on camera | Someone with a stake explains the product to camera. |
| Demonstration | Show it working. The footage is the argument. |
| Problem first | Lead with the buyer's problem. Reaches people not yet comparing options. |
| Comparison | Against the old way, a competitor, or doing nothing. |
| Education or how-to | Teach something the buyer keeps. |
| Social proof at scale | Numbers, reviews, accreditations. |
| UGC, shot like a post | Native to the feed. On its own this adds execution range; pair it with a new claim to add an idea. |
How many ideas does your account run?
The Andromeda Autopsy reads your live Meta Ad Library and groups your ads with these rules. Free, no login, no account access.
Evidence labels: is recommended publicly by Meta. Kaliber benchmark is a threshold Kaliber set: a starting point, calibrated as audits accumulate. Everything else here is Kaliber's model of reading creative, not an official Meta method. Source for Andromeda: Engineering at Meta, 2 Dec 2024. Not affiliated with Meta.