Adform today published the first usage figures from the live rollout of the server that lets outside AI agents work inside its advertising platform, alongside comments from agency staff at Omnicom Media Türkiye, Publicis Norway and a third agency, Avante. According to the Copenhagen company, initial users gave the server a recommendation score of 9.5 out of 10 and a usefulness rating of 4 out of 5, and campaign reporting accounts for almost a third of the queries sent through it. The statement, dated October 28, 2026, does not say how many users answered, over what period, or how the queries were counted.

ELI5

Adform, a company that sells software for buying online ads, lets AI chatbots like Claude or ChatGPT log into its system and pull campaign data when someone types a question in plain words. Today it said early users like the tool and mostly use it to build reports, and staff at three agencies said it speeds up their analysis and spreadsheet work. The scores come with no sample size, and the statement does not say whether the AI can change live campaigns or only read them.

What Adform disclosed

The numbers in today's statement are few, and each arrives without the detail needed to weigh it. According to Adform, the 9.5 recommendation score and the 4 out of 5 usefulness rating came from "initial users". The company does not say whether those users were agencies, advertisers or its own staff, how many responded, or whether the 9.5 figure is an average, a median or something closer to a Net Promoter calculation. Nor is there a date range.

A gap between the two scores is visible even so. Converted to the same 10-point scale, a 4 out of 5 usefulness rating equals 8 out of 10 - a PPC Land calculation, not an Adform figure. Users, in other words, scored the server higher on whether they would recommend it than on how useful they found it. The statement does not comment on the difference.

The usage breakdown is thinner still. Campaign reporting is "the biggest use case, at almost a third of the queries," according to Adform. The remaining two-thirds or so of queries go unassigned. The statement mentions troubleshooting and cross-campaign comparison elsewhere, including one example of diagnosing problems in CTV campaigns, but gives no share for either.

The capability count is unchanged from earlier statements. Adform says it "has opened its entire platform to external AI agents through its MCP Server, making more than 800 capabilities accessible through a common connectivity layer." The company still does not define what one capability is, or list them.

What the agency staff said

Three agency employees are quoted, and their claims are of different kinds. One concerns spending. One concerns access. One concerns accuracy.

Adrián Falque, Digital Manager at Avante, gave the most specific commercial signal in the release. "This is a complete game-changer for our agency. Adform's MCP Server is already a vital part of our daily work," Falque said. "We have gone from spending hours analysing an underperforming campaign to uncovering deep, actionable insights in just minutes. It gives us conversational access to campaign data and surfaces insights we never would have had time to find. The results have been so great that we are tripling our seats to get this in the hands of more colleagues."

Tripling from what base? The statement does not say how many seats Avante held, what a seat costs, or whether a seat is a licence for the Adform server or for the AI assistant that connects to it. The release also does not say whether Avante belongs to either of the holding companies named in its headline.

Gökberk Ertunç, Head of Data Analytics & Programmatic at Omnicom Media Türkiye, framed the server as a way of widening access to data inside the agency. "Adform's approach to opening its infrastructure to AI agents offers new opportunities for agency workflows," Ertunç said. "Using natural language to explore campaign insights makes data more accessible to our teams, helping us work more efficiently and make informed decisions when building and optimising campaigns." Adform's own text makes the same point, saying agencies praised the tool's ability to let team members "pull advanced insights using simple, written requests."

Silje Sandsbakk, Performance Manager at Publicis Norway, made the accuracy claim. "It drastically cuts down our Excel preparation time, but the value goes beyond just speed. Because the numbers we pull through the AI agent match Adform FLOW perfectly, we have total confidence to act instantly without second-guessing the data," Sandsbakk said. "We also love the flexibility it provides, allowing us to choose exactly where we want AI assistance to handle the heavy lifting, and where we prefer to stay in the driver's seat."

The point about matching numbers carries weight because large language models can misread or invent figures when they summarise tabular data. Sandsbakk's statement is a user's observation from inside one agency office. Adform does not describe any audit or reconciliation test behind it, and no time saving is quantified.

All three quotes come from national operations - Türkiye, Norway and an agency whose country is not given - rather than from global leadership at Omnicom or Publicis. Neither holding company is described as having signed a group-wide agreement.

Jochen Schlosser, Chief Technology Officer at Adform, closed the release with the company's position. "AI agents are only as powerful as the technology they connect to. That's why we opened Adform's integrated platform to external AI agents, exposing over 800 capabilities through our MCP Server," Schlosser said. "We've spent years connecting data, technology, and intelligence with one goal: to make advertising better. Now our clients can put it all to work with AI, gaining faster insights, simpler workflows, and more freedom to innovate. Their feedback validates our integrated approach. And we're just getting started."

What the statement leaves open

The largest open question is the same one that hung over Adform's previous statements. Can an AI agent connected through the server change a live campaign, or only read from it?

Today's text points both ways. Adform says the server is transforming "how advertisers manage and optimise programmatic campaigns," and Sandsbakk talks about choosing where AI handles "the heavy lifting." Yet every concrete example in the release - reporting, analysing an underperforming campaign, exploring insights, troubleshooting, comparing campaigns - is a read task. None describes an agent pausing a line item, changing a bid or moving budget.

Several further points go unaddressed:

  • Which AI assistants the agencies quoted were using, and whether the figures differ by assistant.
  • How permissions, approvals and audit logs work when an agent acts on an account.
  • Whether the server is generally available to all Adform clients, and in which markets.
  • How seats are priced, given Avante's plan to triple its count.

None of this means the server is read-only. It means the statement does not say.

How Adform got here

Adform first described the server publicly in a release that ExchangeWire carried on May 13, 2026. According to Adform, clients could connect directly to Adform FLOW - its user interface for the integrated demand-side platform, ad server, data management platform and ID Fusion identity product - through Claude, ChatGPT, Microsoft Copilot or their own AI tools. That release grouped the access into campaign planning and forecasting, in-flight optimisation and troubleshooting, and cross-channel reporting and insights, quoted early feedback from Omnicom, Publicis Groupe and GRIT Media, and described the capabilities as available immediately. Business Wire distributed the same text in June.

On July 11, 2026, Adform's Solutions Engineering team published 29 agentic skills in a public GitHub repository. Those skills run through the Adform GraphQL Model Context Protocol server and cover reporting, forecasting, audience discovery and campaign and creative audits. They are read-only by design. Each is a markdown file with example queries and constraints rather than executable code, and campaign edits stay with the trafficker.

Two months later, on September 10, Adform was named a technology partner for the European rollout of ChatGPT Ads, joining a partner group that had until then been entirely US-based. That deal concerns buying ads inside ChatGPT and is separate from the MCP server, which lets ChatGPT and other assistants operate Adform's tools.

Then, on September 21, two days before DMEXCO in Cologne, Adform said AI assistants could reach more than 800 of its platform capabilities across four stages: planning and forecasting, activation, in-flight optimisation and troubleshooting, and cross-channel reporting and insights. PPC Land noted at the time that the release did not say whether agents could change live campaigns, did not define a capability, and gave no breakdown by stage. Today's figures are the first usage data in any of Adform's statements on the server. They answer none of those three points.

Where other ad platforms stand

Adform is one of several ad platforms now offering MCP access, and the field divides along the read and write line that today's statement skirts.

Google's route was cautious. After saying on July 7, 2025, that it was exploring an MCP server for the Google Ads API, it released an open-source server on October 7, 2025 that runs only in read-only mode, with two tools: one for searching account data and one for listing accessible accounts. Microsoft took a similar line on June 17, 2026, when it expanded the Microsoft Advertising MCP server to an open pilot, again with read-only access, naming agencies including Mediaplus Performance and Stagwell among its pilot partners.

Others allow writes. Amazon put its Ads MCP Server into open beta on February 2, 2026, with pre-built workflows that can create a ready-to-review Sponsored Products campaign from a single prompt. Meta's Ads AI Connectors, opened in beta on April 29, 2026, support creating and editing campaigns, ad sets and ads from the start. X Ads MCP, launched August 24, 2026, exposes 23 tools, 13 of them read-only and 10 that write, and starts every agent-created campaign paused so nothing spends until a person activates it.

In the independent DSP segment, where Adform competes, Yahoo DSP added agentic capabilities on January 6, 2026 that let external agents connect through MCP for campaign activation. Its troubleshooting agent executes fixes only with human approval.

Against that list, X and Yahoo publish the controls that govern write access in some detail. Adform has published a capability count.

Why this matters for agencies and advertisers

Reporting as the top use case fits the pattern across the industry. Pulling numbers is the lowest-risk thing an agent can do in an ad account: a wrong report wastes time, while a wrong bid change spends money. For agency teams that rebuild client reports in spreadsheets every week, handing that one task to agentic AI tools may matter more than most of the 800 capabilities Adform counts.

The cost side is less visible. Large tool catalogues consume model context, and agencies that connect several platforms at once have hit usage limits. Lars Postmus, owner of the agency Draft Digital, said in July 2026 that his team had hit Claude usage limits several times in one working day, and described a stack that replaces 12 direct MCP calls with one buyer agent. Adform's statement does not say how its 800 capabilities are presented to a connecting assistant, or what that costs in tokens.

Security has been raised as well. In July 2025, an analysis shared by machine learning engineer Akshay Pachaar described tool poisoning attacks against MCP implementations, in which manipulated tool responses could push an agent into unauthorised actions. For a server connected to live media budgets, the absence of any description of permissions or logging in today's release is a gap.

There is also the question of how agencies will divide their agent work. Omnicom's Czech operation took part in what was described as the Czech Republic's first fully autonomous AI agent ad buy on June 24, 2026, using ChatGPT connected to R2B2's MCP gateway through the Ad Context Protocol, a standard that six ad tech companies launched on October 15, 2025. Adform did not appear in that transaction. Units within the same holding company are therefore working with more than one agent pathway at once: a general-purpose assistant connected straight to a DSP, as at Omnicom Media Türkiye, and agent-to-agent buying through a sell-side gateway, as in Prague.

For Adform, the commercial argument in today's release rests on one agency tripling its seats and two others describing time saved. Those are early signals from named people. They are not yet a performance result.

Timeline

Summary

Who: Adform, the Copenhagen-based advertising platform, with comments from Jochen Schlosser, its Chief Technology Officer, Gökberk Ertunç of Omnicom Media Türkiye, Silje Sandsbakk of Publicis Norway and Adrián Falque of Avante.

What: Adform published the first usage figures from the live rollout of its MCP server, which gives external AI agents access to more than 800 capabilities in Adform FLOW. Initial users gave a recommendation score of 9.5 out of 10 and a usefulness rating of 4 out of 5, and campaign reporting makes up almost a third of queries. Avante says it is tripling its seats, and Publicis Norway says the tool cuts its Excel preparation time.

When: Today, October 28, 2026, following Adform's first MCP announcement in May 2026, its read-only skills release on July 11 and its DMEXCO statement on September 21.

Where: Adform is based in Copenhagen, Denmark. The quoted agency staff work in Türkiye and Norway, with Avante's location not stated.

Why: Ad platforms are opening their systems to general-purpose AI assistants, and agencies are testing whether those assistants save time on reporting and analysis. Adform's figures are the first usage data it has published, but they come without a sample size, a survey period or any statement on whether agents can change live campaigns.