How Deep Learning is Reshaping Performance Marketing

At eTail Boston, Jaysen Gillespie didn’t come to talk about AI in the abstract. He came to strip it of hype and anchor it in what works for performance marketers.

Gillespie, Head of Analytics and Data Science at RTB House, opened with a simple observation: most AI conversations are dominated by what he calls “the AI you see” , tools like ChatGPT, image generators, and customer-facing bots. But the most transformative AI for performance marketing is often invisible, operating quietly in the background. It’s the “AI you don’t see”, decision-making models that evaluate millions of signals in milliseconds to decide which ad to serve, at what bid, and to whom.

Tracing the technology arc, Gillespie explained how early programmatic relied on static business rules. That evolved into machine learning, where models adapted based on observed results. Today, deep learning represents the next leap, able to process vastly more inputs simultaneously, not just clicks and conversions, but richer contextual data, cross-device behavior, and even patterns in browsing velocity.

The result isn’t just faster optimisation. It’s a shift from reactive to predictive, models that can infer purchase intent earlier, and with more confidence, than rules-based or shallow learning approaches ever could.

For marketers, the lure of deep learning is scale. But Gillespie warned against wholesale change for the sake of novelty. “You don’t want to rip out the engine while the car is moving,” he noted. The smarter approach is layering using deep learning models to augment existing strategies, starting where they can have the most immediate impact.

In practice, that often means beginning with the lowest-funnel activity, conversion-driving retargeting, before moving upstream into acquisition. Once conversion models are tuned and stable, acquisition models can be trained with confidence that they’re optimising toward the right outcomes.

In an environment where third-party signals are vanishing, Gillespie stressed that first-party data is now the most reliable input for deep learning systems. But data alone isn’t enough - it needs to be clean, structured, and fed into models in real time to capture the subtle shifts in consumer behaviour that can make or break campaign performance.

A recurring theme in Gillespie’s talk was measurement discipline. Models are only as good as the outcomes they’re trained to optimise. If a team’s KPIs are too narrow, focusing solely on last-click conversions, for example, the AI will faithfully optimise for that, even if it undermines long-term growth.

The better play is a measurement framework that blends short-term conversion signals with longer-term indicators of customer value. That ensures the model doesn’t just drive transactions, but finds customers worth keeping.

To illustrate, Gillespie shared how RTB House uses stacked retargeting, layering multiple deep learning models, each tuned for a different subset of the audience. The approach delivered a 57% increase in scale without eroding return on ad spend. The key, he said, was in letting each model specialise, rather than trying to force one model to do everything.

Spots are limited, grab yours now.

Independently Created. Not affiliated with eTail Boston.

Unofficially eTail is a ClickZ Media publication in the Events division