By Helen Myers, Advertising Week Correspondent
In a marketing landscape saturated with data, the challenge isn’t access — it’s connection.
When data lives in silos — behavioral here, contextual there — performance suffers, and managing it becomes a full-time job.
At Advertising Week, Melinda Han Williams, Chief Data Scientist at Dstillery, unveiled how multimodal AI is reshaping audience targeting. Her core argument: the next wave of performance won’t come from bigger datasets or faster models, but from systems that can learn across data modalities — unifying behavioral, contextual, and environmental signals to deliver both performance and flexibility.
The insight was clear: AI’s future isn’t just predictive — it’s fully connected.
What Multimodal Means in Practice
Just as self-driving cars merge radar, vision, and GPS data to create a coherent view of the road, Dstillery’s multimodal AI integrates information from across the digital ecosystem — including website journeys, search terms, shopping behavior, CTV viewership, mobile app activity, and even podcasts.
The result is a high-dimensional behavioral map that captures why people act, not just what they do.
“In the same way a language model learns what words mean by predicting the next word in a sentence,” Melinda explained, “our model learns what websites mean by predicting the next site in a user’s journey.”
This embedding — built across 128 dimensions — allows Dstillery’s AI to represent the intent behind every digital behavior. The model doesn’t rely on arbitrary category labels; it learns meaning directly from how people move through the digital world.
Performance and Flexibility
Because Dstillery’s models are fluent across data types, advertisers can begin with any seed behavior — such as first party data, or a list of search terms or product purchases related to the campaign — and allow the system to translate that intent seamlessly across channels.
Melinda illustrated this with a performance CTV activation example: “A ski apparel could start with first-party purchase data and generate predictive segments to reach the CTV devices most likely to drive outcomes for the campaign. And the same behavioral signal that identifies those likely buyers can also power contextual CTV targeting on specific shows, networks, and genres.”
That cross-channel consistency produces a level of precision previously out of reach. “We’re seeing a level of performance on CTV that advertisers aren’t used to expecting — the ability to apply behavioral predictions not just to audiences, but to individual programs,” Melinda said.
From Intelligence to Action: DS-1 and Agentic AI
The culmination of this system is DS-1, Dstillery’s agentic AI interface — an intelligent layer that connects directly to all of the company’s multimodal AI audience capabilities.
Unlike typical AI assistants that simply recommend, DS-1 acts. It can create deal IDs, activate campaigns, or build custom models through a simple conversational interface — even inside tools like Slack.
Marketers can ask, “Find me audiences that will perform for my brand,” and DS-1 instantly shows simulated performance across more than 20,000 audiences to identify top-performing segments before any media spend occurs.
“The agent doesn’t replace predictive AI,” Melinda emphasized. “When you activate audiences with DS-1, the targeting intelligence comes from multimodal AI predictions, grounded in behavioral data, not generative AI synthesis. DS-1’s role is to make that proven predictive intelligence accessible. The accessibility enables faster iteration, richer customization, and ultimately, better results.”
Why Multimodality Matters Now
Early marketing AI focused on automation and content generation — producing more, faster. But the next competitive edge lies in understanding: mapping intent, emotion, and behavior across channels to act with precision and empathy.
Multimodal AI makes that possible. It allows brands to move beyond one-dimensional targeting and see customers as complex, evolving decision-makers. By translating behavioral signals into actionable predictions across every format, it restores a sense of coherence to digital marketing — and raises expectations for what “relevance” truly means.
The Lesson for Leaders
AI’s challenge is no longer capability — it’s connectivity.
Organizations that treat multimodal learning as a foundational skill, pairing it with agentic interfaces like DS-1, will unlock faster workflows, better results, and more human-centered marketing.
The next era of targeting belongs to AI systems that see the whole picture — and to the leaders who know how to orchestrate them.
