BCN, the joint advertising sales house of Hubert Burda Media, FUNKE Mediengruppe and Mediengruppe KLAMBT, today detailed a "Signals Agent" that matches a plain-language description of a target group against more than 1,500 audience and context signals, then hands the selection to PubMatic's AI agents, which build and switch on a private marketplace deal without anyone configuring it by hand.

In Short

BCN sells advertising for hundreds of German magazines and websites, and it has built an AI tool that lets an advertiser type out who it wants to reach and then picks matching groups of readers from BCN's data. PubMatic, a company whose technology sells ad space automatically, has connected its own AI agents so that those groups turn into a ready-to-buy ad deal, which the two companies say takes minutes rather than manual setup and negotiation. The ads people see stay the same, but the packaging and sale of German publisher data moves from human traders to software talking to software, while prices and which buying platforms can use the deals remain undisclosed.

A narrow first, claimed before Cologne

The two companies describe the arrangement as the first direct agent-to-agent implementation for activating publisher signals in the German market. According to BCN, the tool is being presented at DMEXCO, the digital marketing fair that opens in Cologne on Wednesday. The "first" rests on the companies' own assessment. The release names no earlier German deployment it was measured against, and it does not say whether any campaign has yet run through the connection.

BCN's argument is that audience selection, historically a sequence of steps spread across several systems, collapses into a single written request. PubMatic supplies the other half: once segments are chosen, a deal exists on its exchange. Between those two points, according to BCN, nobody forwards data from one system to the next.

What the Signals Agent does

Instead of assembling target groups across different systems and process steps, advertisers describe in natural language whom they want to reach, according to BCN, which offers one example. Translated from the German, it reads: "I want to reach people with a high income who are currently interested in electromobility and in buying a premium vehicle."

From a brief like that, the agent identifies semantically matching segments, ranks them by relevance and returns reach estimates and prices, according to BCN. Semantic matching generally means comparing the meaning of a request with the descriptions attached to each segment, rather than looking for shared keywords. The release does not say which language model sits underneath, where it runs, how relevance is scored or which taxonomy the segment descriptions follow.

The data underneath

According to BCN, the Signals Agent draws on more than 80 million consented profiles from over 50 premium domains in its publisher network. Buyers can address more than 900 segments describing people and more than 600 contextual categories describing content, which together make up the 1,500-plus signals BCN cites.

Two of those figures invite scrutiny. The first is scale. BCN says its brands reach 63 million people across every channel it sells, print included. A pool of 80 million consented profiles drawn from roughly 50 websites is therefore unlikely to correspond to 80 million distinct individuals. What exactly is a profile? The release does not define the term, and the number most plausibly counts identifiers, browsers or devices. The 50-plus domains are also a fraction of the 500 media offerings BCN markets. When BCN adopted The Trade Desk's European Unified ID together with Kleinanzeigen in February 2026, the reach given for its publisher network was 61 million people.

The second is identity. More than half of the available signals are based on privacy-compliant deterministic IDs, according to BCN, which presents the set as an alternative to modelled audiences, pairing the speed of AI-supported processes with the precision of real first-party signals. Deterministic identifiers rest on a direct input such as a login or a verified email address, as opposed to links inferred from shared IP addresses or behaviour. BCN does not name which identifiers qualify. European Unified ID is built on hashed and encrypted email addresses and phone numbers, but the release does not connect the Signals Agent to it or to any other named framework.

A little arithmetic sharpens the point. On BCN's approximate figures, contextual categories account for roughly 40% of the 1,500-plus signals and need no user identifier at all. If more than half of all signals are deterministic, nearly all of the 900-plus people-based segments must be deterministic too. The release does not break the split down, so that inference cannot be confirmed.

From request to deal on PubMatic

Once the Signals Agent has identified a suitable segment, PubMatic automatically creates a pre-targeted PMP deal and activates it on its supply-side platform, according to BCN. A private marketplace is an auction restricted to invited buyers, usually above a negotiated floor price. Each one is identified by a deal ID that travels inside the bid request, so the buyer's platform can recognise the arrangement. Pre-targeted means the audience or contextual targeting is attached to the deal on the sell side before any bid arrives, rather than applied by the buyer inside its own platform. That is the packaging logic the German digital trade body BVDW set out in its February 2026 whitepaper on programmatic curation, written by a 13-person team from 11 companies that included BCN Group.

A workflow that previously required manual setup and negotiation now completes within a few minutes, according to BCN. The release puts no number on the minutes, and none on how long the manual route took. The speed, BCN says, comes from direct communication: rather than passing data between systems, the BCN and PubMatic agents coordinate the entire deal lifecycle automatically. The release does not define that lifecycle beyond selection, creation and activation. PubMatic has built agents for deal management and deal troubleshooting, as it showed European ad teams at an IAB Europe session in April, but whether those functions sit inside the BCN connection is not stated.

What happens after activation is left open. A deal on PubMatic's exchange only produces spend when a buyer's demand-side platform bids on it. The release does not say how the deal ID reaches the advertiser's buying platform, which DSPs are enabled, whether an agency must accept the deal on the buy side, or what floor prices apply. Nor does it explain what the prices returned by the agent represent: a CPM floor, a data fee added to media cost, or a fixed rate. Who sets that price, BCN or the agent?

MCP and AdCP

Technically, the BCN Signals Agent communicates directly with PubMatic's agents through the Model Context Protocol(MCP) and the Ad Context Protocol (AdCP), which BCN describes as an open standard for agent-to-agent communication in advertising. MCP, which Anthropic published in November 2024, standardises how an AI model discovers and calls tools it does not own. AdCP sits above it as an advertising-specific layer. The protocol arrived on October 15, 2025 with signals activation among its supported workflows, a design in which audience data platforms expose targeting capabilities through standardised interfaces. BCN's product name echoes that part of the specification, although the release does not say which AdCP tasks the two companies implemented.

PubMatic is not a neutral party to that standard. It was one of six founding members, alongside Scope3, Yahoo, Swivel, Triton Digital and Optable, when the protocol divided the industry within weeks of its debut over whether another specification was needed at all. Even the name varies: BCN writes Ad Context Protocol, while the specification's public GitHub repository calls it the Advertising Context Protocol.

AdCP also competes with the work of IAB Tech Lab. The standards body named its own umbrella initiative AAMP in February 2026, and on August 20 its chief operating officer, Shailley Singh, counted 13 functions on which the two frameworks overlap, audience discovery and activation and deal creation among them. Those are precisely the two steps BCN and PubMatic have now joined. Singh's document says AAMP supports private marketplace deals and deal creation as well. A publisher exposing supply to both frameworks would, on that reading, maintain two descriptions of the same inventory.

Governance inside the automation

Control and governance remain part of the automated process, according to BCN. Before activation, the selected signals are checked against the relevant policies and brand safety requirements, and every step is documented and therefore traceable. Whose policies apply, BCN's, PubMatic's or the advertiser's, is not specified. Neither is whether a person approves the deal before it goes live.

PubMatic has already built a product for this layer. On August 5, 2026 it added a five-step governance framework to AgenticOS, restricting what agents can transact and recording each decision they take. The BCN release does not say whether the new connection runs through those guardrails.

What the two executives said

Emma Newman, Chief Revenue Officer EMEA at PubMatic, framed the partnership around standards rather than speed. Translated from the German, she said: "This partnership shows how open standards in programmatic advertising unlock real economic benefits. We are not only enabling faster processing of deals, but are also fundamentally changing the way publishers and the buy side interact with each other." When two independent systems can communicate and transact autonomously, she argued, the friction that has hampered programmatic advertising for years disappears, and for the German market in particular the arrangement positions both BCN's publishers and PubMatic as pioneers of a new operating model. "Speed is the proof, but the strategic gain lies in interoperability," she said.

BCN's release spells her surname Newmann. The same executive, with the same EMEA revenue title, appeared as Emma Newman when PubMatic brought Channel 4's VOD inventory to buyers through Activate in June.

Carsten Sander, Managing Director at BCN, described the product as a reduction. Translated from the German: "With the Signals Agent, we reduce an otherwise laborious process to a simple question: which target group do I want to reach? From this description, an activatable deal can be created within a few minutes." The potential of AI agents in programmatic advertising, he said, lies in making access to high-quality publisher data simpler and faster without compromising on quality or control. He also set out who does what: "BCN brings product, data and the Signals Agent, PubMatic enables the direct activation."

BCN's route to an agent

BCN began operating on January 1, 2024, as a realignment of Hubert Burda Media's advertising sales combined with publisher partnerships with FUNKE and KLAMBT. In today's release, the company says it markets more than 300 brands and 500 media offerings reaching 63 million people across print, display, mobile, podcast, social media and events. The February identity move put its headcount at more than 200 people, across locations including Munich, Offenburg and Hamburg, serving over 3,000 advertisers and agencies.

The sales house has been pointing its products at AI for some months. In June, Digiday reported on GEO Brand Impact, a BCN service meant to get brands accurately surfaced inside ChatGPT, Gemini and other assistants. The Signals Agent runs the other way: rather than making brands visible to AI systems, it makes BCN's own data legible to software acting for buyers.

German peers are moving in the same direction, with varying reach. OS Data Solutions, the data subsidiary of Ströer, shipped an MCP server with its OSDX platform on September 2, 2026, although nothing in that release described an agent able to buy anything. Kleinanzeigen, BCN's partner in the identity rollout, is one of ten founders of the European Media Marketplace, which was announced on July 7, runs on Equativ's infrastructure, supports MCP connections and scheduled its pilot for September. The BCN and PubMatic connection differs from both in one respect: it ends in a live deal.

PubMatic's agentic build-out

PubMatic (Nasdaq: PUBM), independent since its founding in 2006, has spent 2026 turning agentic AI into its central argument to investors and clients. It released AgenticOS on January 5, with a transaction layer running through its Activate platform across Programmatic Guaranteed and private marketplace deals. In March, Optable's Audience Agent was embedded in AgenticOS, letting buyers discover and activate publisher first-party audience data without the underlying records changing hands. That arrangement is structurally close to what BCN describes, though Optable is a data collaboration vendor rather than a publisher sales house. In April, AdRoll and PubMatic linked buy-side and sell-side agents over MCP to diagnose underdelivering deals.

The volumes are growing but remain modest. By its second-quarter report on August 6, PubMatic counted more than 80 fully autonomous campaigns and more than 4,000 AI-powered deals since AgenticOS went live, both cumulative figures rather than quarterly run rates. Its speed claims have shifted along the way. The August guardrails release cited up to 87% faster campaign setup, after a company blog post in May had put the figure at up to 98%, and neither came with a methodology. BCN's "few minutes" arrives in the same form: a vendor figure with no stated baseline.

Firsts multiply across Europe

Claims of a market first have become routine in European agentic advertising. R2B2 presented a fully autonomous AI ad buy as the Czech Republic's first in June. PubMatic and the trading desk Amnet ran what PubMatic called France's first agentic campaign, publicised around July 20, and on September 8 Magnite set out its first EMEA agentic campaign, also with Amnet France, crediting it with a roughly 70% reduction in setup time. Each claim is scoped by country, channel or mechanism, BCN's to the activation of publisher signals in Germany. None of these firsts has been independently verified.

Why it matters for buyers and sellers

The first consequence concerns where agentic money is actually flowing. Independent data suggests agents behave less like auction bidders than like television buyers. DataBeat's June report, drawn from May 2026 data, found agentic demand appearing in 86% fewer auctions than conventional buyers and clearing at $6.13 CPM against $6.95. A tool whose output is a private marketplace deal fits that pattern. It also moves targeting to the sell side, where BCN, not the buyer's platform, decides which consented data defines an audience and at what price.

The second concerns interoperability, the benefit Newman singled out. The implementation described today is bilateral: one German sales house, one exchange, one pair of agents. In an interview published on July 16, Magnite's chief executive Michael Barrett described current agentic activity as one buyer working directly with one seller, with the one-to-many stage, where a buyer agent transacts across many sellers at once, not yet reached. BCN's release fits that description. It does not say whether other buyer agents that speak AdCP can query the Signals Agent directly, or whether BCN's signals can be activated through supply-side platforms other than PubMatic. Only if both became possible would the interoperability argument be tested in practice.

The third concerns control of data. For German publishers, whose consented audiences have long been packaged by sales teams, an agent that answers buyers' queries and returns prices makes the catalogue itself the product interface. That could shift some bargaining power toward whoever operates the agent, and some scrutiny toward the consent behind 80 million profiles, a basis the release asserts but does not document. Advertisers are promised speed; publishers keep the data on the sell side; PubMatic gets another stream of deals through its exchange. How the revenue from each deal divides between the three, the release does not say.

Timeline

Summary

Who: BCN, the joint advertising sales house of Hubert Burda Media, FUNKE Mediengruppe and Mediengruppe KLAMBT, represented by Managing Director Carsten Sander, and PubMatic (Nasdaq: PUBM), the independent supply-side platform, represented by Emma Newman, Chief Revenue Officer EMEA.

What: BCN's Signals Agent matches plain-language audience briefs against more than 900 people-based segments and more than 600 contextual categories drawn from over 80 million consented profiles on 50-plus domains, then communicates through MCP and AdCP with PubMatic's agents, which create a pre-targeted private marketplace deal and activate it on PubMatic's platform after policy and brand safety checks.

When: The release is dated today, September 21, 2026; BCN presents the tool at DMEXCO on September 23 and 24, 2026.

Where: The German market, with the presentation taking place in Cologne.

Why: The companies say the connection replaces manual deal setup and negotiation with a process that finishes within minutes and describe it as the first direct agent-to-agent implementation for publisher signals in Germany, while the release leaves pricing, DSP access, human approval and the underlying identifiers undisclosed.