NEW YORK — Campaign optimization can now operate at impression level through real-time agent decisioning, representing peak efficiency for advertising technology that traditionally relied on periodic human intervention across broader campaign segments.
“Can you decision an impression with an agent as opposed to the way it was done in the past?” Joseph Hirsch, CEO of Swivel, told Beet.TV at the Beet.TV/Horizon Media AI Media Summit. “I think the depth is what I’m seeing as the future, all the way down to the impression level or all the way down to the millisecond level.”
This represents the logical endpoint of automation progression that scales from weekly human campaign management to continuous real-time optimization.
Natural language automation spans platforms
Swivel enables natural language interaction for creating business automations, communicating with data, optimizing campaigns, and generating publisher yield across multiple advertising platforms rather than singular systems.
“If you want to use natural language to create an automation in your business to do any task that humans have done in the past, you can do that in Swivel,” Hirsch said. “If you’re a seller using one, three, five, seven, 10 platforms, now you can use a singular platform to interact with these ad platforms via agent.”
Depth replaces width as AI focus
Industry development shifts from broad AI application across various functions toward deeper automation that handles complete workflows rather than partial task assistance.
“There’s a lot of width. Can AI touch everything? Now we’re starting to see more depth where instead of doing 50% or 75% of the workflow, maybe you’re doing 100% of the workflow,” Hirsch said.
This progression enables agents to exceed traditional human limitations by operating continuously rather than periodically.
Campaign tasks scale through frequency
Traditional processes like allow-list and block-list management that occurred once per campaign can now run continuously through agent automation, extracting incremental value through repeated execution of proven methodologies.
Hirsch cited examples of humans analyzing hundreds or thousands of app names and bundles to identify top-performing segments, then manually removing underperforming elements—processes that agents can perform continuously rather than sporadically.
“That allows you to extract incremental value. People are looking for an agent that not just has hypothetical value, but one that’s delivering a result,” Hirsch said.
Future depends on scale
Platform tools remain underutilized not due to feature limitations but because users haven’t discovered capabilities through prompting, similar to unexplored large language model functions.
Organizational penetration deepening from executives to entry-level employees requires accuracy improvements that build trust in automated decision-making systems.
“We have tools inside of our platform that have never been touched because there are things that AI platforms can do that people don’t know because they’ve never prompted them to do,” Hirsch said. “People will become more expert driven in how they prompt, and they will discover new things that they can do.”
You’re watching coverage from The AI Media Summit, A Beet.TV & Horizon Media Collaboration, presented by Seedtag & The Trade Desk. For more videos from this series, please visit this page.
