Every performance marketing playbook that built the great DTC brands of the last decade is now actively working against you.

The high-volume creative testing model, the iterate-on-your-winners approach, the "ship 50 variations and let the algorithm pick" framework that every growth team learned in 2020 through 2023, all of it is now costing brands money instead of making it.

And the worst part is, most operators are just noticing now.

You can still upload 50 variations of an ad, hit publish, and watch the data come in. Everything feels the same. But underneath the surface, Meta swapped out the entire engine in late 2024, and the rules that determined which creative scaled in 2023 are no longer the rules that determine which creative scales in 2026.

Brands following the old playbook are bleeding 20 to 30% of their performance ceiling to a structural inefficiency they can't see inside their reporting.

That inefficiency has a name: it’s called entity ID.

It's easily one of the single biggest reason creative budgets are underperforming this year.

The brands beating it aren't the ones with bigger budgets or better designers, they're the ones who quietly figured out how to feed Meta's new algorithm a fundamentally different type of creative signal.

Graza is one of the cleanest examples of this in CPG, and once you see what they're actually doing, you can't unsee it.

Here's what's happening, why your account is probably losing to it right now, and the exact framework for fixing it.

In late 2024 Meta rolled out a new delivery architecture called Andromeda, a system that processes roughly 10,000 times more data per impression than the one it replaced. The piece that matters most for you sits underneath everything: the entity ID.

An entity ID is essentially a fingerprint, assigned by Meta's computer vision the moment you upload an ad. When two ads share a fingerprint, they share an entity ID, which means they also share a learning phase, an audience pool, and a single ticket to the auction. When two ads carry different fingerprints, each gets its own.

The uncomfortable part is that Meta doesn't show you entity IDs directly, and there's no field in Ads Manager you can pull. You can see proxies inside the Creative Similarity Report (Similarity Score, Fatigue Score, Top Creative Themes), but the underlying fingerprint stays invisible. You only feel its presence when your CPMs rise and your scale stalls for no clear reason.

Since Andromeda shipped, the pattern's been consistent across the accounts we've worked with: brands ignoring entity IDs lose meaningful performance to internal competition they can't see, often in the 20 to 30% range. Inside accounts spending six and seven figures monthly, that's real money walking out the door every week.

If your Similarity Score's above 60, your library's clustering, and if your Top Creative Themes show 70% or more of spend concentrated in one or two categories, you're not running the number of tests you think you're running.

Most brands hear "diversify creative" and respond by producing more variations of what they already have, which usually means headline swaps, background color changes, and new hooks layered onto the same UGC base. The algorithm clusters all of it together.

To unlock a genuinely new fingerprint, your creative needs to vary across four structural dimensions:

The unsayable part: if your performance creative agency can't tell you how many entity IDs your account's running this week, you're paying them to produce ads the algorithm's ignoring.

Volume packages without structural diversity are the most common form of waste inside Meta budgets right now, and most brands don't realize they're funding it.

Graza launched in January 2022, sold out in week one, hit $65 million in sales by year two, and by mid-2025 had become the fifth-largest national olive oil brand in the United States, all while keeping paid ad spend reportedly between 5 and 12 percent of gross revenue. They didn't outspend the category, they out-diversified it.

Pull their Meta Ad Library right now and you'll see seven examples currently running, each one doing something structurally different from the others:

Across these seven ads, Graza's hitting every format, different performer archetypes, environments, and changing up the angle that maps to a different stage of the funnel, which means each one almost certainly gets its own entity ID and explores its own audience pocket inside Meta's delivery system.

That's the structural advantage compounding behind their growth, and it isn't about budget size. It's about building a library Meta can't cluster.

This is the framework we run across our client portfolio, which includes Bespoke Post, Cuts Clothing, Topo Designs, MìLà, and Newton Baby.

1. Audit your entity ID footprint first. Pull the Creative Similarity Report and diagnose your Similarity Score and Top Creative Themes before you produce a single new asset, because if you don't know your current footprint, every new ad you ship's a guess in the dark.

2. Build a four-by-four creative matrix. Pick four formats, four performer archetypes, four environments, and four angles, with every cell representing a structurally distinct concept. That's sixteen unique entity IDs from a single planning cycle. Aim for five fully different concepts per week, each produced in two to three formats.

3. Use creators to multiply diversity. One creator partnership can produce three to five structurally different concepts in a single shoot day, while studio production typically produces one concept in three variations, so the math favors creators every time. Build a creator roster across at least four demographic archetypes so you're never shipping the same face into the auction twice.

4. Budget for entity IDs, not ads. Each entity ID needs roughly $50 to $100 minimum to exit learning cleanly, which means a $5,000 weekly test budget actually funds 50 to 100 unique entity IDs at the floor, not 500 ads that consolidate into 10. Set your testing budget by entity ID count, not by ad count, and you'll stop fragmenting spend overnight.

5. Run the system, not the ad. The brands winning Andromeda aren't the ones with the one viral ad, they're the ones with the repeatable production system that ships structurally diverse creative every single week. The system's the asset, the individual ads are just the output.

We're doing a limited number of free entity ID audits this month for brands spending $50K or more monthly on Meta. We'll pull your library, run the Similarity and Theme analysis, identify where your spend's collapsing into clusters, and send you a one-page breakdown of exactly what to ship next.

Twelve audit slots open. First come, first served.

Until next time
✌️, Conrad