Uncovering the Why: LLMs in the Next Era of Marketing Analytics

In this article, Richard Hachar explores how large language models (LLMs) help marketers understand WHY campaigns work, not just what works. By combining behavioral science, creative classification, and machine learning, brands can turn creative intuition into measurable, repeatable performance gains.

By: Richard Hachar

Modern marketing runs at a scale that older measurement methods were never designed for. With dozens of channels, thousands of creative variations, and the push for personalization, it's harder than ever to determine what drives performance. The challenge isn’t measuring outcomes; it’s explaining them. Only after you understand why something works can you repeat it. That is where new approaches like large language models (LLMs) come in.

This isn’t hypothetical. We hold two patents in this space: one on psychology-based segmentation of creative and another on systematically scoring content. We’ve already seen that classifying creative this way leads to stronger performance without additional spend.

The Cost of Optimization

For many marketers, the cost of optimizing a campaign has long been too high. LLMs like ChatGPT and Claude are reshaping that reality, but most firms haven’t figured out how to translate potential into results. On LinkedIn, it can feel like everyone claims to have “mastered” something new, but real proof is much harder to find.

While they’re often celebrated for their text generation abilities, their potential as classification engines might be just as transformative. You see, LLMs aren’t just great for generating content… they excel at recognizing patterns, interpreting context, and assigning nuanced labels to complex inputs, which are skills that traditional machine learning alone often struggles to match.

The key to successfully leveraging LLMs isn’t just adopting the technology; it's also about understanding how to utilize it effectively. Unguided, they often generate inconsistent or shallow insights. The opportunity lies in turning their interpretations into structured outputs that connect directly to decision-making. For marketers, this means moving beyond experiments with text generation and instead using LLMs to classify, organize, and interpret the creative elements that shape performance.

Forward-thinking marketers are beginning to pair LLM-driven classification with machine learning pipelines. This creates a feedback loop that not only identifies what makes an ad successful, but turns those insights into actionable, data-backed creative strategies. The result is a faster, more informed, and more adaptable approach to marketing performance.

Why Features Matter

Knowing which ads work is easy; knowing why they work has been nearly impossible. This is where feature classification comes in. Without it, performance analytics can feel like checking the score after a game but never knowing how the points were scored. You might see the outcome, but you can’t explain how it happened. With the addition of LLMs and image/text analysis, marketers now have the ability to unpack the layers of what makes their advertising work.

Traditional classification methods, whether rule-based systems or narrow feature engineering, only scratch the surface. Ads are complex: the same words can mean different things in different contexts, and images can stir reactions even without text. When applied well, LLMs can pick up on these nuances and surface the features that separate one ad from another.

An ad might lean on urgency, community values, or educational framing. Another might highlight discounts, lifestyle imagery, or aspirational messaging. Along with hard features like text placement, color, and tone, these categories form the bridge between creative expression and performance outcomes.

Our Creative Engineering platform, for example, is already analyzing thousands of ad images to surface patterns that traditional feature engineering would miss.

From Features to Forecasts

LLMs are good at surfacing insights. They’re less reliable when it comes to predicting outcomes. That’s where machine learning and statistics come in. The features identified by LLMs can be turned into data that traditional models can utilize. From there, those models use past performance to spot relationships between creative features and results, such as click-through, conversion, or retention.

Together, this pairing delivers something neither can on its own. LLMs surface the “soft” features that are tough to measure, while ML models handle the math of linking those features to real-world results. We’ve already tested this in client campaigns, and it has shown that classifications can feed directly into predictive models.

Here’s where it gets useful in practice. As new ads run, their performance feeds back into the training process, so the system evolves with audience behavior, seasonal shifts, and even platform changes. Ads that work today can be reclassified against tomorrow’s data, improving both accuracy and guidance over time.

It’s not about choosing between LLMs or machine learning. The real progress comes when you use them together: let the LLMs tell you what an ad is, and let machine learning tell you what it does.

Making Insights Actionable

For marketers, the value of this approach lies in turning complexity into clarity. Instead of treating creative success as a guessing game, testing and classification create a structured way to see what’s working and why. That moves us from simply knowing what worked to truly understanding why it worked.

Even small steps can create outsized insight. You don’t need massive infrastructure to start; simply classifying creative into a few meaningful buckets will make performance data far more actionable. Over time, those classifications become the foundation for predictive models that can shape creative strategy before an ad ever launches.

The lesson isn’t about choosing between human creativity and machine intelligence. The real value comes when you combine the creativity of one with the predictive strength of the other. The brands that succeed will be those that turn creative signals into measurable outcomes and use those outcomes to make each decision smarter than the last.

And if that still sounds theoretical: we’ve already seen it work. In live campaigns for healthcare and financial services clients, our classification approach has delivered measurable improvements. The next step is applying those lessons faster and more broadly, so every creative decision benefits from what we now know.