How Meta Actually Works in 2026

**Tony Chopp** (0:00)
Meta's own documentation states this explicitly. Relationship between historical performance and future delivery is not deterministic.
And so the reason why this matters for account management is because understanding these systems changes everything about how you manage a meta account. If the algorithm can evaluate 10,000x more ad candidates per impression, your job is to give it more options.

**SPEAKER_2** (0:23)
This episode is brought to you by Tapcart, the mobile app platform for Shopify brands. Here's something worth sitting with. Seventy-one percent of impulse purchases happen on mobile. But for most Shopify brands, mobile is still their worst converting channel despite high traffic volume. Tapcart fixes that.
Brands with Tapcart apps see 63% higher conversion rates than mobile web and 82% more revenue per session. And with the recent launch of Tapcart AI, brands get AI push notifications that automatically send thousands of personalized messages to each shopper. A for you feed that adapts to their individual behavior, and an app your customers actually want on their home screen.
Princess Paul used Tapcart's AI push notifications to hit 11x session conversion versus industry benchmarks with triple-digit revenue growth from abandonment flows alone. Book a free demo at tapcart.com/ctc to learn how brands like Aviator Nation, BEIS, Athletic Brewing and thousands more are winning mobile with Tapcart. That's tapcart.com/ctc.

**Tony Chopp** (1:23)
I'm here today to talk to you about a component of the CTC canon, specifically how we think about modern meta advertising.
Meta is the most powerful advertising platform that's ever been created and as of this moment in time continues to represent the vast majority of share of wallet for ecommerce advertisers.
The power of Meta's advertising engine is explicitly a function of two core systems that are relatively new in the ecosystem.
Number one, Andromeda, Creative First Ad Retrieval Engine and GEM, which is short for the Generative Evaluation Model, which is Meta's ranking intelligence layer operating at LLM scale capacity. Together, these two systems process billions of interactions daily, making probabilistic allocation decisions that no human operator can match.
And it's in the interaction of these two systems that the CTC canon, how we think about structuring and operating in a Meta ads account is generated from.
So I want to talk first a little bit about how Meta optimizes and targets in the new world of Andromeda and GEM. And before you can build an effective Meta advertising strategy, you have to understand this fundamental shift that has occurred and how the platform operates. And Meta moved from an audience first world to a creative first world. The system no longer relies primarily on your targeting selections. Instead, it evaluates creative elements to determine who sees your ads.
The infrastructure is Andromeda and GEM.
And Andromeda operates on a creative first paradigm, using creative attributes, user context and behavioral signals to build a candidate set of eligible ads for each impression opportunity. Meta's engineering team describes Andromeda as enabling 10,000 X increase in model capacity, allowing the system to evaluate far more ad candidates per impression than was previously possible. Paired with GEM, the ranking intelligence layer, it determines what should be shown from that candidate set. And these two things are the underpinning of what is often colloquially referred to in our industry around the need for more creative into the meta ads ecosystem. And these two technological engines are the reason for that. Another super important concept to understand, to take into consideration for modern meta advertising, is something called the breakdown effect. And it's one of the most important things for us to internalize. Historical ROAS does not predict future ROAS. This always reminds me of, some of you may be familiar with the Gambler's Fallacy. This concept is always related to me, to the Gambler's Fallacy. So if you flip a coin 10 times in a row, and it lands on heads 10 times, it is not more likely to land on tails. So this is, it's not a perfect analogy, because the meta probabilistic engine is actually taking into consideration signals that are actually real, as opposed to a coin flip. It sort of helps me detach from the idea that historical ROAS is predictive of future ROAS. And ultimately, this is why making decisions based on yesterday's ROAS is fundamentally flawed. You're operating on incomplete information using a small sample set of historical outcomes to predict a future that a system models far more accurately than you can. And MetaZone documentation states this explicitly. The relationship between historical performance and future delivery is not deterministic. And so, the reason why this matters for account management is because understanding these systems changes everything about how you manage a Meta account. If the algorithm can evaluate 10,000x more ad candidates per impression, your job is to give it more options. If the ranking intelligence operates on signals you cannot see, your job is to trust that allocation decision it's making within your defined cost controls. If historical performance does not predict future performance, your job is to stop making ad-level optimization decisions based on backward-looking dashboards. And trust me, as a lifetime career media buyer, I feel intimately how challenging these ideas are. But the power is in the system. And our job at CTC is to set the constraints, feed the machine, and let it work. And as a result of that, the Canon, CTC's Meta Canon, uses a very simple framework for our campaign structure. Basically, we have a three-tier hierarchy. At the campaign level, we're using Sales Objective in 95% of the cases, Value Optimized with Minimum ROAS as our starting point, ASC or CBO to leverage Meta's budget allocation, and Inflated Budgets with Cost Control. At the ad set level, 7-day click optimization window is our default with broad targeting, with ACQ and retention exclusions, and ROAS targets are set here, not at the campaign level. And at the ad level, we're using AI enhancements where our brands are willing to test, specifically dynamic tests and creative enhancements and multiple URL testing. We're leaning into machine learning optimizations as much as possible.