Meta has pushed its underlying ad retrieval and delivery architecture into a new era for 2025. This updated system, widely referred to in the industry as the Andromeda algorithm, fundamentally changes how ads are served. In short, Andromeda shifts the core mechanism of ad delivery from an audience-first approach to a creative-first neural retrieval and ranking system.

While this may sound abstract, its practical impact is immediate: the platform will no longer rely primarily on manual, granular audience targeting. Instead, a large-scale model trained to match the right user with the right creative in the right context will dictate ad exposure. As this update rolls out globally, many advertisers are experiencing significant performance fluctuations, sparking widespread discussion across performance marketing communities.

The Three Pillars of the Andromeda Algorithm

According to Meta's technical documentation, the Andromeda update is built on three core pillars:

  • Neural Retriever: This system analyzes massive pools of candidate creatives and user historical data in real time to identify the combinations with the highest probability of conversion.
  • Increased Weight on Creative Signals: Creative elements—including images, short-form videos, ad copy, and music—are now directly encoded into the model as primary signals.
  • End-to-End Objective Optimization: Rather than focusing solely on short-term clicks or immediate Cost Per Acquisition (CPA), the model trains its ranking system to optimize for long-term customer value.

Why Advertisers are Feeling the Shift

Advertisers in highly competitive verticals—such as e-commerce, gaming, and digital entertainment—are noticing a sudden shift in performance. The traditional playbook of using hyper-precise targeting and frequent manual adjustments is becoming less effective under Andromeda.

Meta's system now heavily favors broad audiences paired with a large creative pool, allowing the algorithm to explore which creative resonates best with which user segment.

During this transition phase, performance has diverged sharply. Advertisers who quickly feed the system with diverse, high-quality creatives are seeing their ROAS double. Conversely, those relying on outdated strategies are experiencing sudden drops in traffic, rising Cost Per Click (CPC), lower conversion rates, and difficulty spending their budgets. These issues stem from a mismatch between legacy campaign structures and the new algorithm's requirements.

How to Adapt Your Meta Ad Strategy for Andromeda

To succeed under the Andromeda algorithm, performance marketers must shift their resources from granular audience segmentation to creative production and data infrastructure. Industry experts recommend the following playbook:

  1. Scale Creative Production: Prepare 8 to 12 distinct creative variations per campaign to give the neural retriever enough data points to test.
  2. Prioritize Data Quality: Ensure your Meta Pixel and Conversions API (CAPI) are seamlessly integrated and reporting clean, real-time event data.
  3. Respect the Learning Phase: Give the algorithm a 7-to-14-day learning window without manual interventions to allow the model to optimize delivery.

The key to winning with Andromeda is simple: stop trying to outsmart the algorithm with manual targeting. Instead, feed the model high-quality creative assets and clean conversion data, and let the neural retriever do the heavy lifting.