Agentic AI Workflows and the Future of Streaming Ad Tech
AI is bringing a new coordination layer to ad tech that is impacting every area of the streaming monetization workflow, including executing transactions, preparing inventory, understanding audience, and improving campaigns and measurement. Media planning is moving from static spreadsheets to living systems. The best teams are using AI to ask better questions, such as “What if supply shifts midflight?” or “What if this audience over-indexes on live?”
AI makes it possible to model delivery across live, VOD, and FAST together and to update forecasts as signals come in, instead of planning channel by channel.
The Current State of AI Adoption
How many people on the publisher side are currently using AI in ad tech? “I would say 35%, or maybe 40%,” according to Luciano Marcos Escudero, VP of media engineering at Globant.
Even so, it sounds like there are some very promising use cases that Globant is working on for customers.
Some publishers are still testing the AI waters or just beginning to wade in. FreeWheel has encountered several publishers that are still evaluating things, while others are further along their adoption curve. Jeff Ellin, VP of product architecture at FreeWheel, says, “I think buyers are doing their homework earlier, and they already have an LLM of choice that they’re ready to use so they can connect to it and put their data through it a lot faster. I met with 18 European broadcasters,” he reports, “and there really were a number of them saying, ‘I am very interested in this. Can we start tomorrow?’ ”
Other publishers are expressing skepticism or doubt, Ellin says, and few, if any, are fully informed about the technology or the opportunities. “Some know a lot about it; some know nothing about it. This year, it’s more of a learning world. But at some point, [AI is] going to make the buying and the selling happen.”
Improving Efficiency
Shailley Singh, COO and EVP of product at IAB Tech Lab, says the real initial gains are to be found on the efficiency side: “AI for the media ecosystem is creating this ability for advertising companies from brands to agencies to ad tech companies and publishers to create this super efficiency in how they value impressions, how they negotiate, and how they work and manage campaigns.”
The industry is moving from basic machine learning into agentic workflows. From a standards perspective, IAB Tech Lab sees AI being integrated into three primary areas:
- Autonomous transacting and execution—AI agents representing buyers and sellers negotiating and executing deals dynamically
- Outcomes and measurement optimization—Using AI to connect upper-funnel engagement with lower-funnel conversions without relying on brittle legacy-tracking pixels
- Content and rights management—LLMs and AI systems negotiating in real time with publishers to crawl, index, and monetize publisher content
“IAB recently unified all of our AI standards under AAMP [Agentic Advertising Management Protocols],” Singh says. “Instead of ripping out the plumbing the industry relies on, we are extending it. We are integrating modern AI protocols like Model Context Protocol [MCP] and gRPC into existing standards like OpenRTB, AdCOM, and OpenDirect.”
IAB Tech Lab’s AAMP Framework (Source: IAB Tech Lab)
(This wouldn’t be ad tech without more acronyms, but we’ll unpack some of these later in the article.)
Automation and Agentic AI Workflows
“Predictive AI has been around for a very long time—especially in the ad tech world—around forecasting and using machine learning to predict what is going to happen in the future to help pick the right advertising at the right time,” contends FreeWheel’s Ellin. This may be so, but a lot has happened recently, and now, “AI in ad tech” has a number of new meanings.
“Two years ago, if you tried to talk to any media companies, they were a little bit resistant to introducing AI as part of the workflows. Now that is a much easier conversation,” says Globant’s Escudero. “There is not a workflow that we have identified so far that’s completely automated. Most of the workflows that we’re using with our customers are fully speeding up the process, but they’re still having humans in the loop.”
Research from IBM Institute for Business Value estimates that spending for agentic AI will triple in the next year, and 46% of those interviewed in Microsoft’s 2025 Work Trend Index Annual Report have said they are already using agents to fully automate workflows or processes.
Ellin expects to see companies on the ad tech side “understand and manage their inventory by using an LLM to find and package inventory to potential buyers who are looking for very specific audiences or types of inventory. Lately, we’ve been providing this new infrastructure AI layer for our clients to build on top of, connecting us with their own AI.”
Because it leverages the MCP, Ellin explains, “this AI infrastructure allows people to use natural language to interact with a third-party system—in this case, our streaming hub, our buyer cloud platforms, and all the APIs that we’ve historically exposed.”
A Model Context Protocol (MCP) schematic (Source: modelcontextprotocol.io)
“Most of the companies that are implementing AI have a much better understanding of their content, which allows them to do much better planning in terms of the ad insertion,” says Escudero. Earlier ad workflows required a planner to differentiate assets like episodes, movies, and so forth and to reach out to talk about their inventory to agencies. “The process could take weeks just to identify the proper assets, targeting the different potential agencies, and then come back with a proper plan,” Escudero explains. “With AI in place, there is a lot of advancement in terms of understanding the content before getting to inventory creation.”
All new assets (and also old content) now come with the potential for viewing the metadata as part of the media ingestion workflow. “There’s an extraction of metadata and understanding of the videos at the scene level,” Escudero says. This includes what type of artifacts are in the scene and what type of activities are happening. This sort of context, in principle, provides value to advertisers.
Nuts and Bolts of Metadata
Last year, Bitmovin launched a product called AI Scene Analysis that provides contextual metadata enrichment using multimodal AI models to extract metadata from content. These models can map content to IAB taxonomies, according to Jacob Arends, senior product manager for Bitmovin’s AI Scene Analysis and Playback.
Bitmovin’s AI Scene Analysis pipeline (Source: Bitmovin)
“If it’s a car chase scene, your content taxonomies might be car, road, and traffic, but your ad opportunities might be automotive and travel, which can then be passed to the ad server, and that can then influence the decision of which ad is then served to the customer,” Arends says. To accomplish this, Bitmovin uses an array of foundational models.
“The SCTE 35 markers send information to the ad server. It can be content ID for a show, or it could be about the scene or a list of taxonomy terms,” says Olivier Cortambert, head of solutions architecture for video streaming/ad insertion at Yospace. With previous workflows, he notes, “the ad server was just aware of what the content was.”
“I think the biggest problem is that there are so many components in an ad workflow right through the stitching, the servers, DSP, SP, the buyers, etc.,” Arends contends. “For contextual advertising to work properly, a lot of standardization across the board has to happen.”
Arends goes on to say that it’s difficult to measure the impact of contextual advertising at this point “because it’s not a small lift. For example, if a publisher decides to have the contextual information, and they then decide to pass it on to the ad server,” he explains, “the ad buyer needs to understand the value of why they’re paying more money for a contextual ad. Until that’s a bit more standardized and they can validate the signals coming from the publisher, I think it will still take a little while.”
AI-driven scene analysis typically is based on input minutes. Bitmovin will charge 9 cents per input minute on a pay-as-you-go basis, Arends says.
“If you process an asset, it could take between 60 to 90 minutes to do that flow, even running multiple times,” Escudero says. This puts the total cost per asset at “not more than $30–$35.” But costs begin to add up “when you multiply them by hundreds of assets. The return on investment is being analyzed right now. The process per se is not that expensive. The expensive part is doing it at least once for your entire archive,” he explains.
Creative Management
“People are using AI to create multiple versions of creative, even down to a personal level, and that creates a couple of challenges downstream,” reports Ellin. Publishers can optimize based on which versions are resonating, but the management of the sheer volume of ads is equally, if not more, important. “The publishers that use our FreeWheel platform [programmatically] need to review all of the creatives that come in.”
Publishers don’t want to show something that represents a brand in a bad light. As more and more versions are created, both assessment and management of the multitude of versions is going to be ever more important.
Yield Optimization
In 2026, it has become common for every publisher, SSP, and DSP to have a chatbot powered by various agents. As the ad tech industry becomes increasingly populated with a lot of agents, where are we putting these agents to work? What are some of the solutions these agents work on? How should you allocate your budget? Where do you buy what? How do you create a deal and then propagate that deal through the systems necessary to have that deal ID set up?
Another use case FreeWheel sees is finding and identifying the campaigns that aren’t delivering yet. “If it’s a programmatic deal, tell me which programmatic deals are bidding or not bidding and help find and tweak those things a lot faster,” says Ellin. “We’ve seen clients lean in from an ad ops perspective pretty strongly to manage campaigns and understand how they’re performing.”
What FreeWheel has found so far is that clients are more comfortable getting those results and delivering them manually—that is, not requiring the agent to go off and actually make the change for them and trusting the results.
Kantar has worked with Microsoft to take historic data and develop agents that turn decades of intelligence into real-time, conversational insights. Media planners can ask detailed questions about reach in different kinds of programming, how different creatives impacted outcomes, or how to modify buys to gain a better result. Being able to do this means they are going to be capable of scaling their advertising in a way that did not exist before.
Globant is observing the same thing. Escudero suggests a hypothetical scenario to illustrate what he sees happening. “The planner is interacting with a chatbot or conversational UI that understands what happened in the movies or what happened with inventory that is coming, and they start asking questions about the best way to sell ad spaces to the different agencies,” he explains. “The planner will go to this chatbot and start saying, ‘OK, give me different examples on the new inventory that is coming to our platform that will be able to make ad markers [for selling cars] across the movies.’ The chatbot responds with a list of movies and the exact times where they’re talking about cars.”
Escudero goes on to say that “most of our customers are not only using the context of the assets; they are also adding historical data from sales.” This means applying not just specific historical data to a piece of content, but also similar types of content to a knowledgebase of all of the sales they have done in the past with their inventory.
Dentsu built a forecasting and optimization agent to let media planners test assumptions and try out ideas. The company went from proof of concept to production in less than 12 weeks. The results: 80% less analysis time and 90% faster time to insight.
Ethical Sourcing
IAB Tech Lab’s Singh reminds us that in most respects, it’s still early days for AI in ad tech. “Nothing is mature in this space yet,” he says. “We are all testing, we are all learning, and we are all evolving pretty fast. But there are two or three areas where” usage has far outpaced the availability of reasonable guardrails. Compensation, he notes, is one of them, although his organization hopes to remedy that.
How do we prevent AI systems, LLMs, or RAG systems like Perplexity from scraping publishers’ content without compensation? IAB Tech Lab has come up with the Content Monetization Protocols (CoMP) initiative, which standardizes commercial agreements between publishers and LLMs before crawling occurs. It has developed APIs for communication between an AI system and a content owner to ensure that a bot has acquired a license and negotiated terms to access the content it wants.
IAB Tech Lab’s Content Monetization Protocol (CoMP) workflow (Source: IAB Tech Lab)
“The other thing we launched is the agent registry, where you can register your agents with us, and we can verify that it’s a legitimate company,” says Singh. This means that “the agent belongs to the company, and the agent has these capabilities. That’ll help the industry to be able to discover agents, understand their capabilities, and have confidence [that] because it’s listed at Tech Lab, it’s a legitimate agent [they] can work with.”
Agent to Agent
Does all of this agentic AI adoption mean we’re going to have to rebuild our entire tech stack? This is a question several of the experts interviewed for this article say they hear often these days. The answer, they say, is no.
“Our core message is that we are ‘agentifying’ existing standards,” explains Singh. “If you are fluent in OpenRTB and OpenDirect, you have the foundation for the agentic future.”
IAB Tech Lab has created a framework using AAMP to help companies integrate advertising into their existing systems. “We are using the well-known container technology to allow companies to build agents that can act on the bidstream and inform the bidstream, but they then live in the host environment,” says Singh. “Because the agent is embedded within the platform systems—like within the SSP or within the DSP—they can actually listen to and respond to the bidstream by either enriching the bidstream or providing DSPs with a better bid evaluation with their own AI. So it’s really low latency,” he notes, delivering as much as “80% improvements in latency when you do that versus server to server. The bidstream is already very high scale and very high speed. What we are seeing—especially with live TV and a lot of the CTV stuff that’s happening because of the spikes and the concurrency of audiences—is that you need to really act fast.”
Audience Targeting
AI is also being used in audience targeting, which helps when device IDs are unavailable.AI makes contextual and cohort-based targeting dramatically smarter, using data like content genre, time, and aggregated viewing patterns to understand intent, without needing to rely on first-party data.
“Today, the way a user is represented is that you give them an ID, and you give them some context,” says Singh. But an agentic AI system can assemble “a more comprehensive picture of the user. If you visited different finance sites, then it knows that it will have an embedding that will represent financial interest. Then it also has the intent, the propensity of this user to actually take action,” he continues. “You’re giving a very comprehensive 360-degree view of that user to an AI system that can easily match it to what is required for that campaign. [It’s a] very quick match because it’s basically a bunch of numbers, and the AI system only has to match the bunch of numbers and look at the similarity of one number to another.”
Ad Load
Another perk of adopting AI for ad tech is that it enables variable ad-load management. “We are moving away from static ad pods to predictive, dynamic ad loading,” Singh says. “Publishers are using AI agents to analyze user engagement in real time, determining if a specific
user will tolerate a heavier ad load or if they are a high churn risk requiring a lighter load.”
IAB Tech Lab’s priority, according to Singh, is ensuring that when an AI dynamically alters an ad pod, the metadata defining that pod is communicated cleanly through OpenRTB so buyers aren’t bidding blindly into a highly cluttered environment.
Most of these AI strategies promise to automate what is currently a very labor-intensive workflow. That should help to create a better ad experience overall, with better placement, better targeting, and better pricing.
Admittedly, this almost sounds too good to be true. Check in again soon to see how AI-powered ad tech is faring. AI’s implementation in this space is just beginning, and there should be any number of new stories to tell once these solutions have been in the market for a while.
Related Articles
How will the advent of agentic AI impact CTV advertising and streaming adtech in ways that previously predominant AI technologies (AI/ML, generative AI) haven't left their mark on OTT monetization technologies, workflows, management, and strategy? IAB Tech Lab's Shailley Singh offers a concise explanation of the current state of play, and what IAB Tech Lab is developing with its agent-to-agent communication framework, while FreeWheel's Jeff Ellin offers a candid take on where buyers, sellers, and publishers are with agentic and LLM implementation today in this discussion with Streaming Media's Nadine Krefetz at Streaming Media Connect 2026.
16 Jun 2026
How are publishers and advertisers leveraging AI in CTV ad ops in 2026? According to IAB Tech Lab COO and EVP of product Shailley Singh, managing creative and data coming out of campaigns are two key areas; others include prevalidating creative and generating contextual data, says Sargeway, LLC owner Sarge Sargent. Both explain how these processes work and the value AI adds in this discussion with Streaming Media contributing editor Nadine Krefetz from Streaming Media Connect 2026.
29 May 2026
The most important AI systems in sports may end up being the ones fans never see. Agentic AI is not flashy. Fans are unlikely to notice it directly. There will not be endless social posts about orchestration layers dynamically rerouting workflows or prioritizing distribution paths.
19 May 2026
How are streaming content companies doing? Publishers, creators, and aggregators have traditionally made the majority of their revenue on subscriptions, so why does everyone want to talk about advertising? Ad-supported content is more affordable. Because most streaming consumers have maxed out their subscriptions, leaving the SVOD market saturated, advertising is more likely to be the dominant incremental revenue driver for streaming over the next several years, based on how media companies are talking about themselves and guiding investors.
26 Mar 2026