From Vox to Hearst, How Publishers Are Navigating Agentic Buying

The summer season typically brings a slowdown in dealmaking and transactions, but for media publisher Vox, it’s when it announced its first agentic-powered media buy.

Vox leveraged an AI “seller agent” built by ad management platform Boostr to receive, negotiate and complete a media purchase. The ad campaign went from brief to live in under two minutes, with humans still in the loop for approvals, according to the announcement release.

Who was on the opposing side of the transaction? Boostr, which used its own marketing dollars and a “buyer agent” to run a marketing campaign across Vox’s titles, which include New York Magazine, Vox, and Eater.

This media buy might not be as exciting as one that takes place between two completely independent parties, but it is still one of the first agentic advertising transactions to be publicly announced by a media publisher, and more could soon follow suit.

“Every publisher I’ve personally spoken to, they all want to do this,” said Patrick O’Leary, founder and CEO of Boostr.

“The ones that have some understanding are like, ‘yeah, we want to be part of this,’” he said, noting he’d observed an awareness problem amongst some publishers at Cannes. “The ones where they didn’t, once they heard about it, they’re like, ‘oh, sign me up, like, how do I get involved here?’ Because the benefits are pretty compelling.”

Vox did not respond to a request for comment on the media buy, but that could be because it’s busy navigating the company being split into two independent corporate structures, with Lupa Systems acquiring New York Magazine, Vox, and the Vox Media Podcast Network, while Penske Media Corporation acquired the remaining titles to form a new subsidiary, PMX.

What Is AdCP?

Powering the transaction is the Ad Context Protocol (AdCP), which launched late last year and outlines how AI agent ad buyers and sellers will discover each other, negotiate terms, and transact. It’s an open standard governed by an independent nonprofit AgenticAdvertising.org (AAO), and has early backers including Yahoo and PubMatic. The protocol’s goal is to standardize the entire agentic advertising workflow.

Boostr is “a publisher expert” for the protocol, helping guide its development and contributing to the spec. When Boostr first started working with the protocol, it felt very incomplete, O’Leary said, noting that there wasn’t even a mechanism for billing.

As one of the largest sources for advertising on the web, Raptive, an adtech platform that helps creators and independent publishers monetize content, is also heavily involved in standards development and has also executed low-volume transactions with agentic advertising firm Scope3 and ad tech company PubMatic.

A transaction works by having a “buyer agent” and a “seller agent.” The buyer agent will have details regarding the campaign brief, such as budget, dates for campaign launch, audience requirements, and then this will be sent to publishers whose “seller agent” will leverage information like audience data, product catalog, and rate cards to negotiate a deal, O’Leary said.

The transactions that have taken place go back to the basics of what advertising was 20 years ago — direct sales between a buyer and seller, Ryan Maynard, SVP of programmatic sales operations for Raptive, told AMO. Direct sales would typically take four weeks at best, but usually nine to twelve, and now it’s being compressed into minutes, O’Leary said.

Benefits for Publishers

One of the other big benefits to publishers could be how it disrupts the demand-side (DSP) and supply-side (SSP) platforms, which are intermediaries in the ad buying process and take roughly 10-20% in fees. For half of ad impressions, the cumulative middlemen fee will range from 22% to 45%, according to research from digital media analytics firm Adalytics.

With AdCP, publishers can work directly with advertisers in a low-cost and fast way, which means more dollars going to media outlets rather than intermediaries, O’Leary said. Raptive’s Maynard notes that one of the benefits of this approach is ownership. He encourages publishers to build their own seller agents and establish direct partnerships and relationships with agencies rather than depend on intermediaries and vendors.

“If you depend too much on some of these vendors, SSPs or different data companies, you may have your inventory and your audiences bundled with others and sold as part of a one-to-many or a larger package,” he said.

Another big inefficiency that can be fixed is dealing with requests for proposal (RFPs). “One of the most time-consuming, annoying, and stressful actions of any sales organisation that I’ve been a part of has been responding to an RFP from an agency,” Maynard said. “I think agentic will, I hope, put an end to the RFP process.”

Agencies send out RFPs to many different publishers, and publishers only have a short amount of time to respond with a media plan. Hearst has already used AI to optimize its RFP process. It’s built an agentic AI layer on top of its audience customization tool, known as Aura IQ, to enable custom audiences to be built from scratch based on RFPs.

“It was built to really optimize for the people inside of the Hearst walls to really allow them to be more strategic and focus on the things that they need to and commoditize the things that took a lot of time but were the easiest to do with AI,” Mike Nuzzo, head of data solutions at Hearst, told AMO.

The next phase for Nuzzo and his team is to connect this into Hearst’s pitch-to-pay process so the whole process, including pricing, flows freely from start to finish, but right now it’s still a managed service platform via the operations team, who pull the data and put it on a media plan. The agentic AI layer leverages a model context protocol (MPC) based architecture, which means the platform’s coordinator agent could eventually be used in an agentic advertising setting.

“I don’t think that the AdCP protocol is there yet, but we are thinking about how that would work in the next year or two years to have it almost fully automated,” Nuzzo said. “I don’t see a fully agentic platform without any human intervention in the next couple of years.”

How Does Pricing Work?

The key to successful negotiations, with minimal intervention, is having access to the right data. Boostr’s seller agent can access the data in a publisher’s order management system (OMS), which can be used to inform the agent on what rates or discounts to offer. Similarly, for standard direct deals, Raptive already uses internal agents, which leverage its rate card and internal logic regarding discounts, and this information could also be used by the seller agent for negotiations, said Paul Bannister, chief strategy officer at Raptive

“I think our approach has been to just become discoverable and eligible for agentic orders, and if there’s a need to get more specific with our products or our prices are too high or whatever, like, let’s go into this with the flexibility to adjust what we have,” Maynard said.

Publishers are going to need to know their pricing, units, placements, targeting, audience, and be in a position to hand this information over to agents, Bannister said. “Data’s going to be a big part of what feeds these queries from the buyers — they’re gonna be looking for audiences, and you need to be able to define your audiences and have them ready to surface as a package, as a response to some of those agentic buyer prompts that come your way,” he added.

Paradium.ai, formerly known as the Arena Group, is one publisher that has been preparing for this future. It built a first-party data platform Encore AI, which pulls together all the information the company collects directly about its audience and customers and uses AI and large language models to turn that data into detailed and personalized audience segments for advertisers.

“We have a very small direct sales team; agentic becomes a really great place for us to win because, one, our data story is incredible,” Stephanie Mazzamaro, who oversees digital revenue for the Arena Group, previously told AMO. “We have a really clean [data] foundation, so we’re able to surface that to the agentic buyers, and they’re able to discover Arena Group in a much easier fashion.

What’s Next?

Paradium.ai already has its first agentic direct insertion order (IO). For the most part, publishers are in this experimentation and pilot phase.

“They’re still, I think, in the learning phase, and I don’t know if too many of them have taken the plunge and actually served any agentic ads yet,” Maynard said.

Boostr’s O’Leary expects publishers to stay in this phase for at least the next three to six months. Then next year, he expects to see buyer agents talking to multiple publishers and aggregating buys, but there’s not yet the infrastructure for this; it needs to mature, and there needs to be critical mass on the supply side, he said. When this is stable, he expects budgets will move dramatically out of programmatic and towards this new channel due to the economics for publishers, he said.

The Internet Advertising Bureau (IAB) in the UK estimates that AI-driven advertising spend in the country is forecast to hit approximately £18 billion ($24 billion) by 2030, representing around 32% of total digital ad spend. In the US, consulting firm Madison and Wall estimate that AI-powered ad revenue will grow 63% this year to reach $57 billion by the end of the year.

However, only 4% of IAB UK members consider themselves fully “agent-first”, with concerns around transparency, governance, data security and accountability, while 67% say they do not trust AI agents in advertising because of a lack of transparency in decision-making.

“Ultimately, like a lot of things in our industry, the agencies and the big buyers and the advertisers are gonna be the ones [to decide] where the dollars want to flow,” Maynard said. “And if they lean into agentic buying then the industry will follow that, and we’ll see this really rapid adoption, but typically, in my experience, it does take agencies a little bit of time to adopt new technologies.”

The post appeared first on A Media Operator.