Take your mind back to the old Meta setup. Campaigns are divided into twelve audience segments, each with several creative assets. The algorithm was hungry for signals. There is no clear study of which art worked for which audience, and no reliable way to find out. For campaign managers, it meant sitting in the dark, pulling levers and hoping something moves.
That world is mostly gone. The algorithm activates the audience now. What matters is what you actually show people, not who you decide to show it to. That sounds like progress, and it is, but it also puts all the pressure on creativity in a way that most brands haven’t fully accounted for.
The ceiling no one talks about
Most brands, when they get creative, do the logical thing: they iterate on it. They changed the copy. They change the background color. They cut the hook in two seconds. And for a long time, this method works. Then it stops. Assets that were returning returns begin to rise, and incremental changes stop producing incremental returns. This diminishing returns to artand it’s more common than most media groups want to admit.
The problem is not repetition. Repetition is good. The problem is when repetition becomes the whole strategy and the team stops finding new creative directions. If you only press the winner, you don’t have a pipe behind you. So when it finally dries, you start at the beginning.
The new goal is a system that keeps finding new winners, while protecting the benefits of what’s already working. Here’s how we build it.
First stage: Learning signals that cannot be detected by the platform
Meta provides a lot of signal data, and it’s useful. But platform data tells you what was done. It doesn’t always tell you why, and it doesn’t always tell you what to try next. For that, you need to dig deeper into what your audience is talking about, where those conversations are happening, and what’s driving them.
This is where Sentiment Analysis comes in. The tool we built crawls TikTok, Reddit, and the wider web content at scale, not analyzing what people are saying about a category or product, but how they feel about it: whether the conversations are positive, neutral, or negative, and most importantly, what is fueling them. With the spicy noodle product we are working on, something has emerged that has never been taken for granted: a significant amount of content surrounding people making the product with milk instead of hot water. That wasn’t a well-known use case, but it clearly resonated with a real audience segment, and it pointed to a new creative space, including creative partnerships built specifically around that behavior.
Trend data adds another layer. Identifying a trend when it’s starting to gain momentum, rather than after it’s peaked, is the difference between being part of a cultural moment and being a brand that reacts to it. For Adidas, spotting an emerging visual trend early enough to go live within three days means over a million views and a 16% engagement rate within 24 hours. That kind of effect doesn’t just come from good creativity. It comes from good plans implemented at the right time.
Phase two: Building with purpose
Once you have those characteristics, the question is how to translate them into a creative strategy that shows how people interact with your brand. The traditional funnel is still a useful planning device, but the buyer’s journey is not linear. People see a conversion ad before they see a product ad. They encounter the creator’s content before they search for the product. So the art in each category needs to tell a coherent story even if it is seen completely out of sequence.
The way we approach this with clients is by thinking in layers of context rather than sequential stages. The most honest brand art does the job of establishing who you are visually and emotionally. Creator content drives authenticity and trust through the funnel in a way that polished brand marketing cannot. Promotional messages and product catalog ads bridge the gap for an audience that needs a specific reason to act. Each layer does a different job, but everything should read as one sign regardless of the entry point.
Working with a global beauty brand, this meant creating a journey where creativity ranges from product content designed in motion produced by our in-house team to clothing produced by creators, and the goal throughout was the same: when someone sees one piece, they will immediately understand what the product is about. That consistency is what makes the methodology work even if people don’t follow it in a certain way.
Phase three: Structural testing
More testing is not the answer. A smart test. The distinction is important because unstructured testing produces data that you cannot act on. If you run multiple tests at the same time without clear differences, you end up with interesting but incomplete results, and the learning doesn’t continue.
A planned test means knowing what you are testing before the test begins. It means a wide audience with compressed overlap, so each asset gets a clean read without the distraction of close targeting. One variable at a time, consistently. It’s slow, but it produces findings that you can build a creative strategy around.
There is also the question of what you are testing. Many groups default to putting art directly into payment, which is an expensive way to find out if something is working. Organic is a better place to start. If the property is not connected there, it is less likely to pass payment. Testing organic first means you’re putting media budgets behind things that have already shown some traction rather than hoping they will.
One practical note here: check your promoter contracts before you paste a piece of creator content. If paid augmentation isn’t written into the contract, you can’t reuse a high-performing asset without going back to renegotiate. It’s a minor handling problem that makes sense when you’re trying to go fast.
Stage four: Scaling without losing speed
If something proves to work, the bottleneck is almost always speed. Analysis of what worked, brief to creation, production, shipping: traditionally this cycle took days, sometimes longer. The window between “this works” and “we’re doing more of this” was wide enough that momentum kept getting lost before anyone could implement it.
Our Creative Intelligence tool is designed to bridge that gap. Rather than asking the team to manually review performance data and translate it into a summary, the tool does that analysis automatically. You open an ad and instead of a blank page, you have a full breakdown: which hooks led to completion, where viewers dropped off, which viewing times are associated with engagement spikes. He pointed out, in one instance, that the spike at the four-second mark in the video may have been driven by a dog from the frame. That level of clarity about what works within the art itself is not something we’ve seen anywhere else, and it means that the briefs from the creative team are based on real evidence rather than instinct.
The other side of scaling is what you do with learning over time and across clients. Checking within a single account tells you something. Checking on hundreds of accounts, anonymized and aggregated, tells you something very useful. We track all of our tests in a proprietary system called Hippocampus, which currently hosts over 35,000 tests for our customer base. The patterns in that volume are things you won’t find alone.
The part that actually holds it together
This all boils down to one thing that no tool can fix: a creative team, a media team, and a data team talking to each other. Not periodically, not in a monthly review, but consistently, they treated it as one loop rather than three separate activities delivering handoffs. If these groups are closed, the system breaks down. Art repeats without knowing what the data shows. Media develops without understanding what the next creative direction is. Data generates information that never reaches the people who need to act on it.
You don’t need a complex workflow design to fix this. You need the right conversations to happen at the right times. This is where most creative systems fail silently, and it’s an easy problem to solve.
