In the race to optimize ad campaigns with AI, chasing every new data signal can backfire—sometimes, less really is more.
As an AI advocate and practitioner in advertising, I’ve spent countless hours chasing the holy grail of campaign optimization: the perfect data signal. You know the one: that magical piece of information that could make your AI algorithm so precise it predicts exactly when your audience will crave a pumpkin spice latte.
But as I’ve incorporated more data sources into my models, I’ve begun wondering: are we reaching a point of diminishing returns?
Let me paint you a picture. Remember that episode of Friends, where Joey Tribbian is writing a recommendation letter for Monica and Chandler. He starts with a perfectly good sentence: “They are warm, nice people.” Then he thinks, What if I make this fancier? He grabs a thesaurus and swaps every word. The result? “They are humid, prepossessing Homo sapiens with full-sized aortic pumps.” It’s a disaster. The harder Joey tries to improve it, the worse it gets. [For Gen Z readers: if you don’t know Joey Tribbiani, think Ava Coleman from Abbott Elementary.]
This, my friends, perfectly illustrates the law of diminishing returns. And we in AI and campaign optimization fall into the same trap. We keep adding data signals—demographics, browsing behavior, weather patterns, even lunar phases (okay, maybe not lunar phases…yet)—convinced each will unlock better performance. But eventually, the marginal gains become negligible.
I’ve witnessed this firsthand. In countless client meetings, someone inevitably says, “But our other AI vendor uses 20+ signals!” Early in my career, we tested this in sandbox environments, cramming in every possible data point. The result? Performance improved so marginally it was like adding a single grain of sand to a beach and expecting a mountain. At some point, you’re just overcomplicating things.
Signal Creep and the Illusion of Progress
Here’s the truth about data: while certain signals like weather patterns or real-time engagement metrics can significantly impact campaign performance, others merely add noise. It’s like fine-tuning a radio. At first the music gets clearer, but eventually you’re just cycling through static.
This isn’t to say data isn’t crucial. As an AI advocate, I champion data-driven insights. But I’ve learned that more isn’t always better. Often, the key to better performance lies not in adding more signals, but in refining existing ones.
Philosophers have grappled with this concept for centuries. Aristotle’s “golden mean,”the idea that virtue lies between excess and deficiency, applies perfectly to AI. The sweet spot isn’t throwing in everything but the kitchen sink, but finding the right balance of signals.
Consider Occam’s Razor: “The simplest solution is usually best.” In AI terms, a model with fewer, more relevant signals often outperforms one bloated with extraneous data. Complexity doesn’t equal effectiveness.
For marketers and AI practitioners, my recommendation is to embrace diminishing returns. Instead of chasing every new signal, focus on those that truly move the needle. Ask: does this add meaningful value, or is it just another grain of sand?
Get rigorous about measurement. Test fewer signals, but test them deeply. For instance, instead of layering in both weather and time-of-day data, isolate one to measure its real impact. For example, does local humidity actually affect click-through rates on skincare ads, or is it just a red herring?
Look for causal impact, not just correlation. Maybe users who interact with a product page twice are more likely to convert but is that behavior driving sales, or just a side effect of other intent signals like prior purchases?
And most importantly, build in space to audit and prune your models regularly. Cutting input signals can improve performance because removing low-quality signals reduces noise, allowing the model to better-weigh what matters.
Less can be more
AI in advertising shouldn’t aim for maximum complexity, but maximum effectiveness. And that means knowing when to stop adding grains to the beach.
Now if you’ll excuse me, I’m off to optimize a campaign. Or maybe I’ll just rewatch Friends. After all, even AI storytellers need a break sometimes.