Magenta wants AI agents to run retail media buying across every network

Adding retailers means adding consoles, settings and reporting formats. Every retail media buyer knows the problem: budgets don't necessarily grow as fast as the work needed to run them. Amazon lets an advertiser concentrate spend inside a single ecosystem. Across the rest of the market, fragmentation is a productivity handicap too.

Magenta, an agentic platform that opened commercially in September, is going after that overhead. Dali Ben Aleya, one of the founders of measurement company Datagram, is building agents that take on planning, creative, optimization and measurement.

Still in beta, the platform already counts about ten CPG advertisers as clients, including Heineken, Essity, Lustucru and Yoplait. A media agency also uses it and, during a test campaign, compared Magenta's performance with that of one of its own traders.

Magenta says it covers France, the U.K. and Germany, across sponsored products, display, video and audience extension.

"Every retail media network comes with its own operating model," said Dali Ben Aleya, founder of Magenta. A campaign designed for one retailer can't be copied over to the next. Exact or broad keyword matching, audience targeting, first-party data uploads, daily or weekly budget caps: the features differ from one network to the next.

Reporting adds its own gaps in granularity and attribution windows. Each additional network brings its own execution rules and its own way of reading results.

Magenta started by mapping those options. "The agents need to know the formats, targeting and settings available at each partner to build a campaign that can actually run," Ben Aleya said. That domain knowledge is what lets them translate one brief into settings adapted to each network.

On execution, Ben Aleya takes a pragmatic line: work with what exists rather than wait for every network to ship APIs, which let software interact with their tools, or MCP servers, which expose some of their features to AI agents.

Magenta plans three modes. The first relies on the technical connections platforms already offer, with the founder citing adtech vendors Criteo and Equativ. Where those are missing, the agents are built to navigate the web consoles buyers use: creating campaigns, changing settings, pulling reports. How much of that works today still has to be spelled out.

For networks that offer neither, there's email. "The agents send their requests to account managers, whether it's a mid-campaign change or a missing report," Ben Aleya said.

Automation stops at sending and tracking the request. The network still has to execute it, on its own timeline. "The point is to bring those exchanges into the same history as operations run through an API or a console," he said. Actions and retrieved data are centralized in a data warehouse to build what he calls the agent's "operational memory."

Ben Aleya puts automation at 85% of the operational chain on the advertiser side, without detailing the math. That doesn't mean the platform has full control of budgets. Increases and reallocations still go through human approval.

Magenta also leans on a handful of in-house consultants who run campaigns for clients and feed into product development. Their bidding, optimization and keyword tactics train the agents, which the founder describes as still learning. Buyers can use the platform directly, or have those consultants operate it.

In a sector where complexity is part of what pays some intermediaries, Magenta's arrival raises eyebrows. "There are those who see us as a potential partner and others who see us more as a threat to their business model," Ben Aleya said.

Which camp do media agencies and retailers' ad sales arms fall into? Partners, according to the founder. Agencies have every interest in tools that let them "multiply campaigns without growing teams at the same rate," he said. Their job doesn't stop at execution: the time saved could go into consulting. He makes the same argument to retail media networks, which he says he is talking to about doing "more with the same."

Magenta offers two packages. The first, self-service, is aimed at agencies, which use the platform to run their clients' campaigns themselves. The second serves advertisers that don't work with an agency, often smaller companies or in-house marketing teams, for which Magenta runs the campaigns directly.

On that managed offer, Ben Aleya points to a leaner cost structure thanks to automation, which makes media buying accessible to smaller budgets.

For self-service users, the founder gives a figure of 2% of media spend. Magenta is sticking with the standard media buying revenue model: "a proven model that agencies know well and that works," Ben Aleya said.

Magenta's other ambition sits on investment decisions. The platform wants to pair incremental ROAS, which measures additional sales against ad spend, with the margin on the products being promoted. The goal is to tell whether the sales a campaign actually generates produce enough margin to justify keeping it live.

"The same amount of additional sales doesn't have the same value depending on how profitable the product is," Ben Aleya said. The call can differ from what ROAS alone would dictate. "Let's not forget that in retail media, there's the word retail," he said, arguing for media decisions tied to the advertiser's P&L.

That approach requires access to the brand's economics and a reliable read on incrementality. Pulling reports together isn't the same as establishing what advertising added to sales, especially when networks use different attribution rules. The method behind that incremental ROAS still has to be spelled out: Magenta's own measurement, data supplied by the networks, or a modeled estimate.

The project also marks a shift from Datagram. For five years, the founder said, the company argued for measurement kept independent from execution. Magenta now brings both into the same infrastructure so results feed the agents' decisions directly. Datagram continues to serve its existing clients, but Ben Aleya is focused on the new business.

Bringing the two together should speed up optimization. It also makes the method worth checking: the measurement used to arbitrate buys can double as proof that the platform running them works.

Self-funded so far, Magenta is preparing an angel round before the end of the year. Ben Aleya is targeting Europe first, then the U.S. The roadmap includes connections to The Trade Desk, Meta, TikTok and Google DV360, plus an extension to CTV to follow the budgets CPG advertisers increasingly put into off-site.

That extension would move Magenta closer to general-purpose agentic media buying platforms. "We're all going to grow in concentric circles and we'll end up touching," the founder said. Its retail specialization will then have to show up in the calls it makes: which products to support, at which retailers, and with what contribution to profitability.

The first test is still execution cost. If every new retailer keeps demanding work out of proportion to the budget it gets, buyers will have little reason to spread spend further. Magenta wants to make that diversification cheaper. It's then up to retailers to show what it returns.