First, let’s clarify what Meta’s Andromeda update is: Launched in late 2024 with full global rollout completed by October 2025, Andromeda is Meta’s next-generation AI-powered “ad retrieval engine”—a foundational overhaul of how ads are selected, ranked, and delivered across Facebook and Instagram. Unlike older systems that relied on advertisers’ manual audience segmentation (e.g., stacking interests, 1% lookalikes), Andromeda processes millions of ad creatives and user signals in real time (powered by hardware like NVIDIA Grace Hopper) to answer a new question: “Which ad should this person see?” rather than “Who should see this ad?”
Its core goal? To handle the explosion of AI-generated ad creatives (Meta reports a 10,000x increase in creative volume post-generative AI) and deliver hyper-personalized ads. For marketers, this means the old playbook of micro-targeting and sparse creatives is obsolete—success now hinges on broad targeting, simplified account structures, and massive, diverse creative libraries. Below, we break down our real-world results with Andromeda, including key performance data and the strategy that drove 40% year-over-year (YoY) ROAS growth.
Over September–October 2025, our campaign ran with Andromeda’s algorithm at full capacity, delivering consistent efficiency across spend and conversions. Below are the unfiltered weekly performance trends (note: all figures reflect “Desired Actions” — adaptable to sales, leads, or app installs):
Across the full two months (Sep 1–Oct 31), total spend reached $180,000, driving 9,200+ results, an average cost per result of $19.57, and a total ROAS of 3.74—representing a 40% YoY improvement in ROAS and a 30% YoY lift in click-through rate (CTR, sustained above 3% in November 2025).
To test Andromeda’s adaptability to different spend levels, we ran parallel ad sets with varying daily budgets. The data below highlights how budget allocation and creative diversity impacted efficiency (all reflect “Website Purchases” as the desired action, but the pattern holds for other objectives):
Every decision was built to work with Andromeda’s AI logic—not against it. Here’s the framework that delivered 3.74 ROAS, applicable to any campaign objective (sales, leads, app installs, brand awareness):
Andromeda performs best when it has clean data to learn from—so we abandoned complex, segmented ad sets in favor of a 1-1-N framework: 1 campaign, 1 broad audience, N ad sets (each with unique creatives). Key choices:
- High Initial Budget (> $2,000/day) + ASC Fully Enabled: For “proven high-intent” objectives (e.g., repeat purchases, qualified leads), a bold launch budget accelerated the learning phase by driving rapid conversions (we included past engagers/purchasers to send strong signals to Andromeda). Meta’s Automated Speed Control (ASC) let the algorithm adjust delivery speed in real time, avoiding manual pacing bottlenecks.
- TCPA as a Safety Net: To prevent long-term inefficiency (a risk with high budgets), we set a “maximum acceptable Target Cost per Action (TCPA)” (Meta’s version, not Google’s). This flexible guardrail kept the algorithm focused on quality actions, even as the campaign scaled past 60 days.
Andromeda’s AI relies on diverse creatives to match ads to users—so we scaled production to 400–500 unique assets (5x more than 2024) and prioritized variety over “perfect” quality:
- Formats for Every User Behavior: 60% static images (quick attention for brand awareness/leads), 25% carousels (storytelling for complex conversions), 15% short videos (engagement for app installs).
- Continuous Freshness: We added 10–15 new assets weekly to avoid creative fatigue (Andromeda penalizes stale content). This kept CTR above 3% in November, even after months of running.
- Broad Targeting > Micro-Targeting: Andromeda’s AI finds better micro-segments than manual interest stacking—stick to broad audiences (e.g., “U.S. adults 25–54 interested in [your niche]”).
- Creative Volume = AI Accuracy: The more diverse assets you feed Andromeda, the better it matches ads to users. Even small teams can use Meta’s Advantage+ Creative tools (e.g., video splitting, AI-generated 文案) to scale efficiently.
- Mature Accounts Win First: This framework works best for accounts with 6+ months of data (Andromeda leverages history to optimize faster). New accounts should start small to build initial data.
By aligning with Andromeda’s strengths—simplified structure, massive creatives, and trust in AI—we turned $180k in spend into a 3.74 ROAS and 40% YoY growth. For marketers ready to adapt, this isn’t just a “trend”—it’s the new playbook for Meta advertising.
© Copyright notes
The copyright of the article belongs to the author, please do not reprint without permission.
