use case

Creative clustering, explained: why your ads cluster and what it costs you

the short answer

Creative clustering is when near-duplicate ad creatives, especially those sharing a hook, collapse into a single auction slot under Meta's Andromeda retrieval and split delivery instead of multiplying it; Omniscia surfaces these Entity ID clusters and the spend they waste.

Creative clustering is the mechanism behind a problem most performance teams feel but cannot name: you launch a fresh batch of creatives and total delivery barely moves. The reason is that Meta's retrieval engine groups creatives by how similar they look in the opening, and ads that fall into the same group, the same cluster, compete for one auction slot rather than expanding your reach.

Understanding clustering changes how you build a set. The question stops being 'how many creatives can I ship' and becomes 'how many genuinely distinct entities am I giving the system to place'. Omniscia makes that visible by clustering your own ads the way Meta does and putting a number on the waste.

Entity ID clustersOmniscia groups your ads the way Meta's retrieval does

What an Entity ID cluster is

An Entity ID cluster is a group of your creatives that Meta's retrieval treats as effectively interchangeable, usually because they share a hook or open in a near-identical way. Within a cluster, the creatives do not each win their own delivery; they queue behind a single slot and divide whatever that slot earns.

Omniscia builds these clusters from similarity analysis of your live creatives and tells you why each one formed, the shared hook, angle or format that pulled the ads together. That 'why' is what lets you fix it, because you can see precisely which dimension to vary.

The efficiency figure and what it means

On top of the clusters, Omniscia reports an efficiency figure: the percentage of your creatives that are redundant in retrieval terms. A set that is 70% efficient has roughly a third of its creatives adding no incremental reach, just splitting the delivery the rest would have won anyway.

This is a far more honest measure of a creative library than a raw count. Two teams can each be running thirty ads, but the one with high clustering is effectively running far fewer distinct entities and paying to maintain the duplicates.

Tying clustering to wasted spend

Clustering is not just an efficiency abstraction; it costs money. Omniscia's Spend Protection analysis uses the same similarity clustering to estimate the budget lost to ads cannibalising each other, so you can see the overlap expressed as wasted spend rather than a percentage.

Paired with the Untapped Branches view, which shows the hook and format combinations you have not covered, you get both halves of the picture: where your set is wastefully dense and where it is thin. The fix is rarely 'make more ads', it is 'make different ones, and stop paying for the duplicates'.

frequently asked

What causes creative clustering?
Similarity in the opening of your creatives. Meta's Andromeda retrieval reads roughly the first 3 seconds, so ads sharing a hook or near-identical opening collapse into one auction slot and form a cluster.
How is an efficiency figure useful?
It tells you what share of your creatives is redundant. A low efficiency figure means a large portion of your set is splitting delivery rather than adding reach, which a raw creative count would hide.
Does Omniscia show the money lost to clustering?
Yes. Its Spend Protection analysis uses the same similarity clustering to estimate budget lost to ads cannibalising each other, so clustering is expressed as wasted spend, not just a percentage.
How do I reduce clustering?
Vary the dimension that pulled the cluster together, usually the hook, and use the Untapped Branches list to add genuinely distinct angles and formats rather than more near-duplicates.

Last updated June 6, 2026

ready to try omniscia?

open omniscia