For years, Meta was primarily seen as a social media company, but these days that definition falls short to describe Meta and all their work. In practice, Facebook's Meta has become one of the leading innovators in the development of artificial intelligence on the planet. And one of the systems that best represents this shift is Andromeda, its AI and Machine Learning (ML) powered ad retrieval and personalization engine.
Although Andromeda is rarely presented as a visible “product” for the end user, its impact is profound: it decides which ads are shown, to whom, when, and why, using predictive algorithms to select the best ads for each individual user.
Meta and its transition to an AI company.
To understand Andromeda, it is important to first understand Meta's strategic shift, one that led to the development of this AI. In recent years, Meta has aggressively invested in AI infrastructure, exemplified by specialized data centers and high-end GPU accelerators, such as the NVIDIA Grace Hopper superchips, or proprietary language models like LLaMA (Large Language Model Meta AI), and increasingly complex recommendation systems. Advertising has always been at the core of Meta's business. However, the traditional model—manual segmentation, advertiser-defined interests, and human control—began to show limits. Too much data, too many variables, and an increasingly restrictive regulatory environment in terms of privacy. The solution that Meta found was to automate decision-making through artificial intelligence. Andromeda was born exactly from that need.
What is Andromeda?
Andromeda is a next-generation ad retrieval engine designed to operate at massive scale and in near real-time. Its main function is to select, from among millions of possible ads, which ones are most likely to be relevant to a specific user at a given moment.
Unlike previous advertising systems, Andromeda does not primarily rely on fixed rules or manual segmentations. Instead, it uses deep learning models that analyze different aspects of a user's browsing, such as behavioral signals, device, historical ad performance by user, and hundreds of additional variables that allow Andromeda to select the ad that has the highest likelihood of impacting the user receiving it, all in just fractions of a second.
From a technical standpoint, Andromeda combines several advanced AI approaches. Notably, the neural networks developed to work with NVIDIA Grace Hopper superchips. These networks allow for the evaluation of a potential ad's relevance not only through explicit data but also through complex relationships between seemingly unrelated signals. Andromeda does not evaluate all ads one by one. Instead, it progressively filters large sets of ads until it is left with the most promising ones.
Every user interaction feeds the system. Clicks, views, conversions, and omissions become training data for Andromeda, constantly learning without the need for direct human input. In simple terms: the system gets better with use. It learns from users and the trends they explore with them, constantly readjusting to keep up with people's changing tastes.
One of the most visible effects of Andromeda is the reduction of direct control for advertisers. Old strategies based on interests, very specific audiences, or micro-segmentation have lost relevance. The logic is simple: if the AI is good enough, it doesn't need humans to tell it who to show the ad to. It only needs good data and good creative pieces. The more unique and eye-catching an ad is, the better chances it has of being selected by the Andromeda algorithm and, consequently, it will receive greater visibility compared to a generic ad. For some advertisers, this has been a liberation. For others, a nightmare.
From Meta's perspective, Andromeda offers clear advantages for advertisers and users. In well-structured campaigns, many advertisers have reported improvements in cost per conversion and effective reach. AI can detect patterns that no human team could identify in time. Additionally, Meta's engineers assure that the Andromeda system reduces reliance on explicit personal data, relying more on contextual signals and aggregated behavior.
However, not everything is enthusiasm regarding Andromeda. It has also generated significant criticism, as advertisers no longer know exactly why an ad works or doesn't work. The explanation of 'the AI decided' is ambiguous, at best, and can prevent smaller companies from gaining visibility through digital advertising because Andromeda does not believe that their product, image, or ad is valuable enough for the group of users they are trying to target.
For larger companies, on the other hand, optimizing becomes more reactive than strategic. Adjusting campaigns is, in many cases, an exercise in trial and error. Investing in Digital Marketing has become a risk overnight, which can lead to less bold, more cautious advertising. Companies have few incentives to invest large amounts of money in advertising without the guarantee of success. Furthermore, the more automated a system or algorithm like Andromeda is, the less transferable the knowledge is. What works in Meta does not necessarily work in another ecosystem. There is a certain skepticism around Andromeda and Meta for the above reasons, with some professionals questioning whether the reported improvements are real or simply the result of changes in attribution and measurement.
Beyond advertising, Andromeda raises broader questions: To what extent is it acceptable for an opaque system to decide what commercial content we see? How can we prevent the reinforcement of algorithmic biases? What happens when optimization prioritizes only conversion, ignoring social or psychological impact?
Meta claims to apply constant ethical controls and evaluations to keep Andromeda impartial and under control, avoiding political, religious, or cultural directions. However, the size and complexity of the system make total oversight difficult, at best.
Andromeda is not an isolated project. It is neither the first nor will it be the last of its kind in the world. Andromeda is part of a larger strategy where AI becomes the invisible layer that supports everything in our times: advertising, recommendations, content, virtual assistants, and eventually, immersive experiences.
Andromeda represents both the best and the most unsettling aspects of large-scale applied artificial intelligence. It is efficient, sophisticated, and clearly profitable. But it is also opaque, centralized, and difficult to question. For advertisers, the message is clear: adapt or fall behind. For users, the experience will remain 'relevant,' though increasingly less understandable. And for the digital ecosystem in general, Andromeda marks a point of no return. The question is no longer whether AI will decide what we see. The question is who decides how that AI is trained, with what, and for what purpose. And that question, curiously, is still not answered by any algorithm.
Brenda Maldonado, a digital marketing expert with 15 years of experience in creative direction, marketing analysis, audiovisual production, graphic design, relativity, and data analytics across various media, shares her opinion on AI in modern marketing and its sustainability, as well as the impact it has on professionals and clients.
“From my perspective, this is an opportunity for professionals such as artists, designers, managers, brand directors, and developers in the field to showcase their talent. Advertising and connection with the end customer are undergoing reform, or a renaissance. Reaching a customer is no longer measured by a price, but by the real value where connecting an artist's creativity with good practices and ethics in the proper use of AI will be reflected in real value, both for the brand and for the consumer in self-sustaining flywheel models.”