“Where should I spend my next dollar?” is now a question that marketers can pop into an AI chat box instead of a media dashboard.
On Thursday, media measurement provider Measured launched a Model Context Protocol (MCP) server that allows brands to ask ChatGPT, Claude, Gemini and other AI platforms how their media is performing.
The answers are based on aggregated and anonymized data from over 30,000 incrementality tests across more than 200 of Measured’s brand clients, some of which spend hundreds of millions on paid media.
An MCP server is a bridge that lets AI platforms connect to external systems, data and tools through a standard protocol. In this case, chatbots like ChatGPT or Claude can query Measured’s system directly and return an answer in the chat window.
Measured didn’t build this bridge on a whim, said CEO and Co-Founder Trevor Testwuide. Enterprise clients have been asking for ways to bring incrementality insights into the AI tools they already use every day, he said, rather than having to log into yet another platform.
“AI is becoming the primary interface for a lot of knowledge workers,” Testwuide said. “That’s where they’re spending their time, so that’s where incrementality intelligence has to live.”
Behind the interface
But there’s a lot of math happening behind the chat interface.
Measured runs thousands of cross-channel experiments every quarter to see how (and whether) campaigns move the needle on sales and other outcomes. The results feed into what Measured refers to as its intelligence database.
All of this is obscured from marketers, however. They just type in questions – for example, “Did my Meta prospecting move Amazon sales?” or “If I spend one more dollar, what does that do for new customers?” – and get plain‑English answers back in the chat.
From there, they can ask follow‑ups, cross-reference results across channels, see how their lift compares with competitors and surface channels or tactics they might want to test next.
“Clients want to know how they’re doing incrementally, but they also want to know how they stack up against their peer group,” Testwuide said. “Are we over‑performing, are we under‑performing and what else should we try?”
That curiosity reflects a broader shift in how marketers want to measure performance, favoring tests that show what a campaign really contributed as opposed to reports that just tally clicks and impressions.
And it’s been a long time coming.
Testwuide has been in the measurement space for around 15 years — including stints at Visual IQ and as co‑founder of Conversion Logic (acquired by VideoAmp) before launching Measured in 2017 — and says it’s taken most of that time for brands and platforms to move away from correlation‑based tools like MTA and MMM and toward experiments that focus on causality.
Getting on the same page
This framing isn’t just for marketers, though, Testwuide said. It gives marketing and finance a way to communicate.
“Incremental return is the language the CFO already speaks,” he said. “Applying that to marketing gives the CMO and CFO a common way to talk about what the media really did.”
The large ad platforms are starting to speak that language, too.
Rather than grading their own homework on last‑click performance, as they’ve been wont to do, biggies like Google, Meta, Pinterest and Snap are leaning more on third-party measurement and incrementality tests, especially for upper‑funnel and video formats that can look weak in platform reports but stronger in controlled experiments.
“They’re all getting religion around incrementality measurement as their North Star currency,” Testwuide said. “The brands are there, the third‑party measurement is there and now the platforms are there too.”
But it’s still TBD on whether marketers want their data flowing through AI tools. Some are all in; others are wary.
According to Testwuide, roughly 15% of Measured’s clients don’t want their data exposed to AI in any form, while others are pushing in the opposite direction and want to consume as much as possible through large language models.
“You see the full spectrum,” he said. “And there’s also the bigger middle that’s still figuring out how far they want to go.”
On a short leash
Part of their hesitation comes down to trust. Even with tightly scoped prompts and a fixed underlying data set, LLMs are still capable of inventing numbers, trends or narratives that look plausible but don’t actually reflect reality.
Testwuide said the solution isn’t to ignore those issues but to box that behavior in as much as possible.
Rather than letting a general-purpose model roam freely over raw event data, Measured is constraining what the model can see and do. Its agents work off of structured experiment results, Measured’s own summaries of how campaigns performed and selected learnings from client work, all tied to specific workflows.
There are roughly 20 of these task-focused agents, each nudging the model toward answering concrete questions, like lift for a given campaign or the shape of a diminishing-returns curve, instead of free-associating across the entire data set.
The idea is that if the model is only allowed to operate within those predefined contexts and only on top of incrementality reads that have already been validated, it’s less likely to produce the kind of overconfident but incorrect answers that make many marketers skittish about trusting AI with budget decisions.
“The magic of AI doesn’t happen when you dump a massive data set into an LLM and say, ‘tell me the insight’ – that’s actually when you run into a lot of these issues,” Testwuide said. “The contextual layer is incredibly important here.”
