Meet Us at Affiliate World Europe 2019

Most people running Meta ads are still optimizing for a system that no longer exists. They’re splitting budgets across six ad sets, testing one variable at a time, and capping frequency because they’re scared of “ad fatigue.” Meanwhile, Meta’s infrastructure quietly rebuilt itself from the ground up. If you don’t understand what changed, you’re fighting the algorithm instead of working with it. The engine at the center of this shift is called Andromeda. It’s Meta’s internal ad matching and ranking architecture, and understanding even the basics of how it works will change how you structure campaigns, how you think about creative, and how you interpret performance data. The Meta Andromeda algorithm explained simply: it’s the system that decides which of your ads even gets a chance to compete before a human ever sees it. What Andromeda Actually Is Meta published the full technical breakdown of Andromeda in a December 2024 post on the Engineering at Meta blog. The headline numbers got passed around: 100x faster ad matching 10,000x increase in model capacity for the matching stage, +6% recall improvement, +8% ads quality improvement on selected segments. Most people read those numbers and moved on. But the implications are well worth digging in deeper. Before Andromeda, Meta’s system had real constraints on how many ads it could evaluate against any given impression opportunity. The matching step, where the system pulls candidate ads from the full inventory to rank against a user, was the bottleneck. You could have a phenomenal ad that never found its audience simply because the system didn’t have the computational budget to evaluate it. Andromeda changed that ceiling. It uses a two-stage architecture: a fast approximate matching layer that casts a wide net across candidates, then a more expensive deep-ranking model that scores the final shortlist. The system runs on NVIDIA Grace Hopper Superchips and Meta’s own MTIA silicon, co-designed hardware and software that enables far more complex neural networks to evaluate ads in near real time. The result is that the system can now meaningfully evaluate far more ads per auction, which directly affects how your creative gets distributed. The Number That Actually Matters: 10,000x More Variants When people say “10,000x more variants,” it sounds like an abstraction, so let’s make it easy to understand Say you’re running a campaign for a DTC skincare brand. You have 8 active ad creatives. Under the old system, many of those ads were effectively competing for evaluation slots before they even reached the ranking stage. Your best ad got found. Your fourth-best ad might have rarely been pulled into consideration at all. Under Andromeda, all 8 are genuinely in play, matched to the right user at the right moment. The system can explore the full creative space you’ve given it. That changes the logic of how many ads you need, how different they should be from each other, and how you interpret which ones are “winning.” We ran a test on this dynamic for a supplement brand spending around €850/day. We went from 4 creatives per ad set to 12, but made sure each one had a distinctly different hook, angle, and format. CTR on the campaign improved, but more importantly, our cost per purchase dropped from €38 down to €26 over a 21-day window. The reach into cold audiences improved significantly. We had more genuinely different creatives driving traffc. Not just 12 versions of the same UGC testimonial with a different color grade. Why Creative Diversity Beats Creative Volume This is the part nobody talks about enough. Most media buyers hear “more variants” and go produce 20 slightly different versions of the same ad. Same hook, same offer, same format. Just different faces or different opening lines. But that is not creative diversification. Meta has been explicit about this. In their official Creative Advantage post on Meta for Business, they describe the shift directly: the focus has moved from niche targeting to creative diversification as the primary lever for finding relevant audiences. And their follow-up three-step creative diversification guide makes it even clearer. They’re not asking for volume. They’re asking for conceptually distinct creative signals. Andromeda’s matching system is trying to match ads to users based on predicted relevance and engagement. If all your variants are the same conceptual ad with minor surface changes, you’re not actually expanding the candidate pool in a meaningful way. You’re just giving the system more of the same signal. What actually works is what I’d call conceptual diversity: ads that represent genuinely different creative theses. One ad that leads with social proof, another that leads with a transformation story, another that’s educational, another that’s founder-led. Different formats: static image, short-form video, carousel. Different lengths: 7-second hook-and-close versus 60-second narrative. When your creative pool has real variety, Andromeda can do what it was built to do: find which thesis resonates with which user segment, without you having to segment manually. What “Conceptual Diversity” Looks Like in Practice When building a creative strategy now, the three dimensions I think make sense to focus are: angle (the core emotional or rational appeal), format (static, video, carousel, collection), and length (short grab vs. longer story). You want good coverage across all three, not just variations within one. A campaign with one 15-second video and six slightly different thumbnails is not a diverse creative pool. A campaign with a 15-second video, a 45-second narrative, a static proof-based image, and a carousel showing before/after, us what the algorithm can actually work with. Jon Loomer, who has one of the more grounded practitioner-level takes on this, breaks down creative diversification across seven specific examples if you want to go deeper on the tactical side. Worth the read. Meta also published a companion piece, Demystifying Creative Diversification, that’s worth bookmarking as a reference for what they actually mean when they use that phrase. Meta Andromeda Algorithm Explained: What It Means for Campaign Structure When you fragment your budget across many ad sets, you’re starving the algorithm’s learning phase in each one. Fewer conversions per ad set result in slower signal accumulation, which means worse audience matching, which means you never see what the creative could actually do with proper data behind it. Advantage+ Campaign Budget, what used to be called Campaign Budget Optimization (aka. CBO), exists to solve this. Let the budget flow to where conversions are cheapest at the campaign level, and stop manually allocating between ad sets. Meta’s own page on this feature cites an average 4.6% decrease in CPA when it’s enabled, which seems to be conservative. The gains are usually bigger when you’re coming from a heavily fragmented structure. But there’s an additional effect that Andromeda emhhasizes. A nore consolidated structure means that the matching system has a bigger, unified creative pool to evaluate per auction. You shouldn’t split your creatives across multiple ad sets and limit learnings. You should have one campaign, broad targeting, multiple strong creatives. That’s the structure that lets Andromeda work at full capacity. It’s worth noting that this doesn’t mean you should never segment. Brand versus prospecting, for example, often need separate campaigns for better budget control. Just make sure not to create separate ad sets for every audience, placement, or demographic, that is what is working against you now. Why Your “Best Practices” Are Outdated There’s a common idea in the Meta ads community that you need to “control variables” the way you would in a lab experiment, one change at a time, isolated testing, clean attribution. But this approach assumes that the algorithm is a passive pipe that delivers your ad to whoever you tell it to. This doesn’t work anymore. Andromeda is actively matching. It’s finding the sub-audiences where each creative will perform best, and that process takes time and data. When you isolate variables too aggressively, pausing ads after 48 hours, testing hooks in isolation from the offer and CTA, killing anything that doesn’t hit your CPA target in three days, you’re interrupting a matching process that hasn’t had time to complete. You’re drawing conclusions before the experiment has actually run. Most Meta ad buyerss’ obsession with fast, clean testing loops made more sense when the algorithm was less sophisticated. Now it can cost you your whole campaign. For a more measured counterpoint, because not everyone agrees Andromeda changes as much as the hype suggests, the team at Motion put together a solid roundup of practitioner perspectives, including […]

When performance goes down, most marketers usually blame it on the creative. The truth is that the creative is rarely the problem, but the angle. Here’s what usually happens: You launch an campaign You find an angle that works Scale the working angle Angle burns out (performance drops) You start working on new creatives. The problem with this approach is that you’re promoting your campaign with a single angle (narrative).  And a single angle cannot carry long-term scale. If you want to add stability and scale up you need to run with multiple angles. Let’s break down ho to generate 10 strong angles for the same offer.   What is an Angle? An angle is the narrative or perspective you use to present your offer/product. It is not: A headline tweak A different image A rewritten CTA Think of the reason why someone cares to interact with your ads and convert on your offers. There are different motivators you can use to promote the same offer. That’s what actually helps you scale.   Why Most Marketers Stop at Just One Angle Most of them think the offer defines the message, but in reality it doesn.t The offer defines the outcome, while the angle defines the story. If you only see one way of positioning or promoting an offer, you’re not thinking deeply enough. Strong offers/products can support multiple narratives, you just have find them.   The 10 Angle Framework Here’s a simple framework that works. Take the offer or product you want to promote and run this through these categories. Problem Agitation Angle Focus on the pain point. Example: “What Most Homeowners Don’t Realize About Their Current Insurance Coverage” This angle highlights the existing problem. Fear Angle Highlight risks or loss. Example: “This Simple Insurance Oversight Could Cost You Thousands” Fear drives action when used responsibly. Savings Angle Focus on cost reduction Example: “Homeowners Are Saving an Average of $X With This Insurance Adjustment” Savings angles perform well in uncertain economic times or price oriented markets. Opportunity Angle Frame it as something beneficial. Example: “Why Now Might Be the Best Time to Upgrade Your Home Insurance Coverage” Opportunity appeals to ambition and curiosity. Curiosity Angle Create intrigue without overselling. Example: “Why Experts Are Quietly Talking About The Latest Insurance Changes” Curiosity works well in discovery campaigns. Data-Driven Angle Lead with statistics Example: “7 Out of 10 People Miss This When Signing For a New Insurance Policy” Numbers build credibility. Authority Angle Leverage the expertise. Example: “Insurance Experts Recommend Reviewing This Before Year-End” Authority builds trust. Story-based Angle Tell a relatable narrative (test multiple) Example: “How This Family of Four Reduced their Home Insurance Cost by 38%” Stories humanize the offer. Make it relate to them. Localized Angle Make it geographically relevant. Example: “[City] Homeowners May Qualify for a New Insurance Benefit This Month” When used right, localization increases relevance. Timing or Urgency Angle Tie it to a season or deadlines. Example: “Experts Warn New Insurance Rule Could Raise Prices by April 2026”   As you can see, for a single product like “home insurance,” we were able to generate 10 different angles you can build creatives around. Hook Supporting Copy Landing Page Visual direction. If the angle changes, everything else changes. That’s how you should test it. Why Angles Protect Campaigns from Fatigue Most campaigns die because they are heavily relying on a single angle. If the angle dies, the campaign dies with it too. But, if you pushing have 8-10 angles then you can: You can rotate different narratives You can test adjacent motivations You can expand without ruining what’s already working. Angle diversity supports longevity. How to Systemize Angle Creation Instead of brainstorming randomly, follow this process. Define the core outcome of the offer. List all emotional drivers connected to that outcome. Match each emotional driver to a narrative category. Build one creative per angle. Test angles before optimizing creative variations. Refrain yourself from launching 12 versions of one angle. Instead launch 5 distinct angles first, then refine winners. Advanced Angle Combination Once you have tested and validated individual angles, you can easily combine them for a stronger impact. Example: Data + Fear “New Report Warns Many Homeowners May Be Underprepared for Major Damage” “Data Suggests Millions of Homeowners Could Be Underinsured” “Insurance Study Highlights Risks of Outdated Home Coverage” Authority + Urgency “Experts Urge Homeowners to Review Their Insurance Now” “Regulators Advise Homeowners to Review Insurance Before the Next Storm” “Experts Urge Homeowners to Review Insurance by the End of This Month” Story + Savings “How One Homeowner Discovered They Were Paying Too Much for Insurance” “Why One Family Decided to Revisit Their Home Insurance Policy” “How a Simple Insurance Check Helped One Homeowner Cut Costs”   Proper combination creates better resonating angles, but only after you know which one works individually.   Why Angles Matter in Scaling Scaling isn’t just about spending more. Scaling is about expanding to a broader audience. To do that you need to expand on different narratives. When you have multiple validated angles: You don’t rely on a single creative angle You expand to new audience segments You increase volume without risk. That is how seasoned performance marketers scale consistently across different offers.   Final thoughts If you feel stuck with your creatives, you don’t need a better design, color change or a variation of your headline. You need a different and better narrative. Every strong offer or product supports multiple narratives. Your job is to uncover them and build angles that convert around them.

Let’s talk about something that quietly destroys more campaigns than bad creatives ever will. Impatience! Most media buyers launch a campaign and start staring at statistics. Day 1: CPA is 40%-50% above the target or there are no conversions at all Day 2: It slightly improves but still not enough conversions Day 3: CPA fluctuates again not getting better. Day 4: nothing… They have already paused the campaign by midday on day 3, or sometimes halfway through day 2 (or even 1). The typical panic reaction! Then tree weeks later they see someone else scaling the same offer on the same traffic source, and potentially with their original (unique) creatives. Sounds familiar, right? Let’s break down why this happens and, more importantly, how to avoid shooting yourself in the foot.   Expecting Stability Too Early Performance marketers and affiliates love controlling their stuff. They want: Conversions within the first few hours of launching the campaign. Accurate and clean performance data. Predictable results, regardless of how hard they shake the algorithm. All while forgetting that most campaigns are quite messy in their early stage. The algorithm has to learn how the funnel and offer performs. It tests which creatives perform best. Also test what audience pools convert better. A normal process that generally lasts between 48-72 hours, but sometimes can extend up to 120 to 150 hours. Assumption is the mother of all screw-ups! As such stop judging your campaigns too early.   Why Early CPA Fluctuations Are Normal Here’s what happens when you launch a new campaign: The platform starts exploring difference audience pools. It tests delivery timing. It optimizes towards early signals while still testing new variables. This stage is commonly referred to as the exploration phase, and strong fluctuations are normal. You might hit your CPA within a few hours from launching a campaign, just like you might not even get any conversions at all on day one. Everything is unstable at this stage, so don’t panic and let it run. The Difference Between Bad Performance and Insufficient Data This one is critical, so let’s make sure both concepts are crystal clear. Bad performance looks like: Extremely low CTR. Terribly low Conversion Rates. No engagement signals. Spending multiples of the CPA without any improvement. On the other hand, insufficient data looks like: CPA is slightly above the target. Inconsistent early conversion rate. Mixed engagement signals. One needs to be cut quickly, while the other needs patience.   How Much Data is “Enough”? This is one of the most common questions, but there is no universal answer. A good rule of thumb you can use is this: Spend at least 2-3x if your target CPA per Angle before making a decision. For example: If your target CPA is $50, don’t kill an angle after spending $60 or $70. Give it at least $100 or better $150 before doing that. Your main goal is to see patterns from the data you’re collecting, not just conversions. Why Emotional Optimization is Dangerous Let’s be honest. When CPAs are too high, it feels personal. You start questioning: “Did I pick the wrong angle?” “Am I buying bad/fraudulent traffic?” “Is this offer saturated?” A typical emotional reaction. But performance marketing is about data, not feelings. The best media buyers follow strict rules and make optimization decisions based on patterns, KPIs, and thresholds. You need to remove emotions and gut feelings from your optimization process. That alone can improve your campaigns’ performance dramatically. The Right Way to Kill Campaigns If it is wasting money, you should definitely kill it! Here’s a simple framework. Kill immediately if: CTR is below baseline expectations Conversions are inexistent or random Metrics show no signs of recovery. Keep it running if: CTR is healthy Engagement rates (LP CTR) are decent. CPA is slightly above your break-even threshold. Once you collect enough data from campaigns with promising performance, you can easily turn it into a winner.   Why Killing too Early Hurts Long-Term Scaling Here’s what happens when you kill campaigns too fast: You never validate angles properly You don’t build a reliable data history (much needed for future tests) You stay stuck in perpetual testing mode. Instead of giving campaigns time to generate data for confident decisions, you end up constantly chasing new offers.   Change your Testing Mindset Instead of asking: “Is this profitable yet?” Ask: “Is this showing promising KPIs?” That means that: People are clicking They are engaging with your funnel There is intent. Profitability comes once you validation. Make sure you test properly, then scale and generate profits. Final Thoughts Most campaigns don’t need more optimization, or a wildly different optimization approach. They need enough time, so let your campaign mature. Sometimes the difference between a losing and winning campaign is discipline.