Meta’s Andromeda update has been live for over a year. TikTok’s Smart+ has matured. And the industry is still asking: What does AI-driven ad delivery mean for creative strategy?
The data is clear: platforms have become far better at finding buyers based on creative signals. What hasn’t changed is what makes creative testing strategy effective. The shift has simply made the gap between strategic testing and executional testing impossible to ignore.
What Andromeda Actually Is
Andromeda is Meta’s retrieval engine. It’s the first step of ad delivery where the system narrows millions of ads down to a few thousand candidates. It picks the shortlist.
Why did Meta need this? Scale. What used to be one ad is now potentially thousands of variations: multiple text options, AI-generated headlines, Advantage+ enhancements, dynamic formats. Andromeda was built to handle that complexity, matching ads to people based on creative signals like visuals, copy, and tone.
The meaningful change? The system now recognizes creative similarity. Five product shots with slightly different headlines get treated as one ad. The algorithm rewards genuine creative diversity and collapses duplicates.
TikTok’s Smart+ operates on similar logic. In our recent Q4 testing for a global enterprise retail partner, Smart+ delivered 5x higher ROAS versus manual campaigns, but only when fed genuinely varied creative built around distinct buyer motivations. This underscores why scaling creator-led growth requires more than just volume; it requires distinct narrative angles.
The Testing Gap Andromeda Exposed
Most creative testing was built around executional variables. Hook A vs. hook B. UGC vs. polished. 15-second vs. 30-second. Different CTAs.
That’s legitimate testing. But it’s testing how to say something, not what to say, or who you’re saying it to.
When you test executional variations of the same underlying message, the algorithm recognizes them as similar and serves them to similar people. You think you’re expanding reach but you’re actually refining within a narrow band.
When you test genuinely different motivations, the algorithm finds genuinely different audiences. That’s the unlock most brands miss.
Testing Still Requires Discipline
The platforms want you all-in on automation. Meta removed the old recommendation of six ads per ad set. Some advertisers are running 50 creatives. But volume without intention is just expensive randomness.
Don’t dump creatives without a hypothesis. Every ad should represent a deliberate angle. If you can’t articulate why two creatives are different beyond execution, the algorithm won’t treat them as different either.
Meta now selects winners quickly, sometimes too quickly. An ad can be deprioritized after just 500 impressions, before other variants have had a fair chance to perform. Monitoring distribution and using rules to control spend early in-flight helps ensure stronger signals surface before budgets are concentrated.
Match testing volume to budget. Every creative asset requires sufficient spend to generate statistically significant signal. Spreading a budget too thin across 50 variations results in ‘expensive randomness,’ a critical risk for CMOs defending efficiency metrics to the CFO.
Market Intelligence Is the Layer Most Brands Are Missing
“Creative is targeting” has become a talking point. But the common response is “make more stuff and see what sticks.”
The smarter approach starts upstream.
Before you brief a single piece of creative, you need to understand who you’re talking to, what motivates them, and how they actually talk about the problem you solve. That’s what enables adaptive thinking, responding to real consumer signals instead of assumptions.
At The Shelf, we deploy Interest-Driven Market Intelligence as the engine that powers every campaign. We move beyond the social graph (who they follow) to the interest graph (what they care about) to ensure creative resonates before spend is committed:
That intelligence shows up in four ways:
- Social listening to capture the actual words and phrases your audience uses when they’re not being sold to. That language becomes the foundation for hooks and copy that feel native rather than forced.
- Persona development based on motivations, not demographics. The busy parent who wants convenience. The enthusiast who wants quality details. The gift-giver who wants emotional impact. These aren’t segments you target in Ads Manager. They’re segments you reach through creative.
- Funnel-stage insights that inform not just what to say, but when. The message for someone who’s never heard of you is different from someone who’s been to your site three times.
- Cultural context that makes creative feel relevant. What trends are shaping how people think about the problem? What language feels fresh vs. played out?
This is the difference between testing random creative and testing strategic hypotheses.
Why Influencer Content Wins Here
In this environment, creator content becomes a structural advantage. Creators naturally introduce motivational variation, ten influencers briefed on the same product will produce ten genuinely different angles. Those angles also arrive in native formats the algorithm is built to reward. Creator content belongs in the feed because it’s made for the platform, not adapted to it.
Just as importantly, social-first language comes naturally. When creators are briefed with real audience insight, they translate it into content that resonates. They’re not performing marketing copy, they’re speaking in a way audiences already recognize.
The key lies in the brief. Standard “product-first” briefs fail in an algorithmic world. Our Creator-Led Production Engine ensures briefs are rooted in data: “Here are four buyer personas, the specific syntax they use, and their decision-stage triggers.” This specificity enables creators to produce assets that function as targeted performance levers, not just “brand awareness.”
That specificity only comes from doing the intelligence work first.
What This Means
Executional testing has diminishing returns. The bigger unlock is testing whether different messages find different buyers.
Insights come before creative. The brands winning aren’t just making more content. They’re doing the intelligence work first.
Influencer content is a structural advantage. Creator content briefed with real insights is how you scale diversity without sacrificing relevance.
Platform updates will keep coming. The fundamentals won’t change. Start with intelligence. Build creative that’s diverse in ways that matter. Test with discipline. Let automation find the people who respond.
That’s the playbook. It worked before Andromeda. It works now. It’ll work for whatever comes next.
