The machines are constantly evolving, and so must we. As performance marketers, we have always lived in the liminal space where the only constant is change. Meta’s new ad retrieval AI, Andromeda, marks another in an ongoing series of shifts in the way in which ads are served by machines and optimized by us. The algorithms keep evolving too, especially as creative AI features advance and demand more processing power. But success is still ultimately dependent on having skilled human ad managers and strategists who understand these systems and know how to make them work for a business’s bottom line. This means knowing when (and how) to adapt! For advertisers, changes like these present both an opportunity and a challenge. 

The Evolution of Ads Optimization

At OpenSail, we’ve spent years optimizing for efficiency – eliminating spend bleed, maximizing conversions, and holding every dollar accountable to optimize our clients’ ROAS. This discipline has always been especially important for clients who are strategically conservative with their ad budgets (because not everyone needs to spend $10K a day to scale!).  

Ten years ago, optimizing meant waiting for statistical relevance before pausing an underperforming ad. We had to wait for “enough” data to prove a loss or failed performance. But, in the last few years, machine learning has changed the rules. Meta’s ad delivery algorithms have become exponentially more accurate at identifying winning ads early – allowing us to make faster, smarter decisions to prevent wasted spend, time, and drive stronger results with the same budget. 

Now, Meta is ushering in a new era of ad retrieval in the form of Andromeda. 

Andromeda: Ads in the Age of AI Ad Retrieval 

The introduction of Andromeda marks the beginning of a new chapter in Meta advertising. As an AI-driven ad retrieval system, Andromeda is designed to handle the staggering scale of AI-generated ad diversity. With millions of ad variations competing in every auction, Meta has designed a new retrieval engine capable of processing, analyzing and selecting the right ad to serve your audience in seconds. 

For months, you couldn’t be blamed for thinking Andromeda was just a buzzword going around marketing circles. There wasn’t much available by way of hard data to suggest how ad buyers should adjust to its introduction. But now, new performance graphs and 3rd-party data have begun to show an interesting shift in ad retrieval and delivery behaviours. 

From “Underperforming” Ad to Sleeper Hit

One graph in particular caught my eye! It showed an ad that had performed initially, dipped for an extended period of time, and then suddenly spiked and outdid its original performance by 10x. If that ad had been paused prematurely, those later conversions would have been lost forever. 

This fundamentally challenges one of our longstanding beliefs and best practices: cutting “underperforming” ads quickly to prevent spending wastage. Andromeda’s new ad retrieval logic may result in the machine testing new audience segments, or relearning ad relevance, before we see a surge in performance. And so, we must adapt, and adopt a new mindset – one that balances fiscal discipline with patience for the AI’s learning curve. 

The Apex Predator Formula

Every evolutionary leap in AI capability requires an adaptive leap in human strategy. The best marketers, the marketing apex predators, are those with elastic intelligence who meet change with a fearless determination to develop new tactics based on new insights. 

As we knuckle down to analyze Andromeda’s impact and effects, here’s what we’re testing and observing closely: 

• Healthy bleed tolerance: Instead of killing ads too quickly, we’re measuring the value of a controlled “slow bleed” to see if the system redistributes delivery to stronger audience pockets.
  
  
• Granular UTM tracking: Since a single ad may now splinter into multiple AI-generated variations, deeper UTM-level tracking is essential.
  
  
• Cross-platform attribution: We’re leaning on GA4 and Shopify conversion paths to identify delayed conversions that don’t show up in-platform but still originate from those “paused-too-soon” ads. Accessing both MER and Shopify's real-time sales attributed to ad channels and ad platform spends are key areas to assess performance more holistically.
  
  
• Creative diversification: We’ve reduced ad creative frequency to preserve learning phases, but are exploring AI capabilities outside of the platform to maintain control and quality while leveraging dynamic copy variation in the platform, where we still have more control, to meet the diversity that Andromeda seems to reward.

• Not giving in to the hype and making drastic sweeping changes. The digital marketing landscape has always been turbulent, and we have found holding to foundational methods of testing with small changes and measurement of impact is the steadiest way to move forward. At the end of the day, every business is unique and requires a deep understanding of what the target KPIs need to be to bring in a profit, and we have yet to see an AI feature cover this work. 

• What we do know for sure; now is not the time to relax on expertise and leave it up to the machines. 

“What excites me most about Meta’s Andromeda update is that it reinforces the importance of testing. Too often, companies hyper-fixate on finding that one ‘perfect’ copy and creative combination. That mindset comes from traditional marketing, where every change is expensive and slow. In digital, success comes from using data to adapt in real time.

By consolidating campaigns and spend while diversifying your creative, you give the algorithm one clear data source to build on. With stronger, more centralized data, you can better understand what your audiences are responding to and why. The Andromeda update may reveal layers of audience behaviour we weren’t able to see before, but we’ll have to test and see.” - Kasidi Belanger, Social Media Ads Manager

What This Means for Your Businesses

So, what does this mean for your marketing dollars? Here are what we believe to be the key takeaways for business owners today:

• The machines may be getting smarter, but they’re not infallible. It is still up to us to interpret context and ultimately make the call about whether an ad is “learning”, generating real revenue, or “losing.”
  
  
• Instinct and expertise matter; now more than ever. No automation can replace an experienced strategist who knows your brand, your audience, and your growth capacity; or an ads manager who deeply knows the ad platforms and your core business metrics outside of the ads platform.
  
  
• AI optimization and ad creative are not one-size-fits-all. Especially in smaller or regional markets — like Canada’s prairie provinces — where a lack of audience density and data volume can skew machine learning. Human calibration and market understanding keep our clients’ results consistent and cost-effective.
  
  

We are the translators between your goals and the algorithms that determine your visibility.

The Role of the Human Marketer in the Age of Machines

AI isn’t new to us. We’ve been working alongside machine learning for nearly a decade now. Each time Meta changes how it learns or delivers, or new regulations change what data we can see, we adjust in tandem — because that’s how we stay at the top of the food chain.

The Andromeda era will bring new insights, new best practices, and yes, new frustrations. But for lifelong learners, this is the exciting part! It’s rarely business as usual for social ad buyers. The platform just published a new module for us to cut our teeth on.

So dive in. Test boldly. Analyze deeply. Stay curious. The marketers who don’t fear the machines but rather evolve with them will continue to lead the pack — the true apex predators of performance marketing! Comment end  

‍