The first clue was a banana. During a graduate supply-chain class at MIT, Brian O'Kelley learned that an identical piece of fruit carries a different carbon footprint depending on where it lands. O'Kelley knew a messier supply chain. He had helped create the ad exchange, co-founded AppNexus and watched programmatic advertising turn webpages into high-speed bazaars. A banana crossed oceans. An ad impression crossed servers, auctions, vendors and devices in milliseconds. Both left exhaust.
Scope3, which O'Kelley started in late 2021 with Anne Coghlan and engineer Emma Mulitz, began by measuring that exhaust. Its model estimates emissions from the corporate operations, ad selection, data transfer and consumer device involved in delivering digital media. The pitch was deliberately practical: give a marketer a carbon number at campaign level, then let the marketer spend away from waste.
That sounds like a sustainability dashboard. It became a map of advertising's plumbing - and maps are useful to people who intend to reroute traffic. Scope3 now describes itself first as infrastructure for agentic advertising. Its Interchange connects buying agents carrying a brand's brief with sales agents representing publishers, streaming services, retailers and other “surface owners.” They discover products, negotiate terms, execute a buy and send results back through an open language called AdCP.
The carbon meter found bad media
Scope3's original insight was less sentimental than it appeared. High-emissions advertising often meant too many auctions, duplicated requests, long chains of resellers or pages manufactured to harvest ad money. Those are climate problems, but they are also lousy-media problems. In a joint analysis with the Association of National Advertisers, Scope3 found that just 2 percent of sites accounted for half of measured emissions. Brands in the study cut emissions by as much as 36 percent while redistributing spend toward higher-quality inventory and, in some cases, reaching performance goals sooner.
The product made the number actionable. A Green Media Product is a private-marketplace package that excludes “climate risk” inventory. Buyers could purchase one through existing ad platforms without installing a pixel or calling a new API. If a seller chose to compensate the remaining emissions, Scope3 set a price of $100 per metric ton of carbon dioxide equivalent - roughly a ten-cent CPM for each 1,000 grams emitted.
The best case study is almost comically unglamorous. Scope3 worked with an unnamed global publisher's ad-operations team and reduced average US emissions from 366 to 113 grams per thousand impressions. The publisher streamlined partners, cleaned up its ad stack and reported no negative effect on revenue. The reader can copy that move without buying a climate slogan: map every intermediary, compare emissions and quality together, remove redundant paths, then watch revenue rather than assuming more pipes mean more money.
“Could you bring sustainability-based optimization to digital advertising and make the ecosystem price carbon into key decisions on a per-impression basis?”Brian O'Kelley on Scope3's first question
Then the measurement layer learned to negotiate
The turn toward AI did not arrive from nowhere. In 2024, Scope3 raised $25 million in a strategic round led by GV, after a $20 million GV-led Series B the previous year. It also released a methodology for estimating the carbon and water costs of generative AI. The company said usage revenue had grown twelvefold from the second quarter of 2023 to the same quarter in 2024. It had evidence that buyers wanted decisions, not simply reports.
What failed first: the blocklist
In March 2025, it introduced an agentic media platform and Brand Standards. The latter attacks one of advertising's oldest blunt instruments: the keyword blocklist. A generic system may keep a brand away from any article containing “shooting,” whether the story concerns a crime, a basketball game or a film set. Scope3 trains a bespoke agent on a brand's own labeled examples, standards and tolerances. The intended payoff is finer judgment - safer placements without automatically starving serious journalism of ad revenue.
Interchange takes that idea from classification to commerce. A marketer can express an objective, budget and rules in ordinary language. A buying agent queries seller storefronts containing inventory, formats, audiences, pricing and terms. Agents negotiate inside boundaries set by their human principals. Approval thresholds can keep a person in the loop; decisions are logged for compliance and reporting.
This is Scope3's clearest difference from carbon-accounting rivals such as Cedara or IMPACT+, verification firms such as DoubleVerify and Integral Ad Science, and conventional demand-side platforms. It combines sustainability data, bespoke suitability rules and transaction infrastructure. Instead of adding an AI assistant on top of familiar menus, it wants agents representing buyer and seller to converse directly.
Open protocol, closed beta
Every would-be new market has a chicken, an egg and a PowerPoint deck. Scope3 is trying to reduce that problem by making the underlying protocol open. AdCP - the Advertising Context Protocol - is published through AgenticAdvertising.org and co-developed with companies across the ecosystem. In October 2025, collaborators used it for a first real-money agent-to-agent buy on LG Ads inventory. By 2026, Scope3 said Interchange had 26 live storefronts and an average campaign budget of $511,000.
Its business model favors supply acquisition. Opening and operating a seller storefront is free. Buyers pay for Interchange access, though the price requires a sales conversation. Scope3 can also handle consolidated invoicing. The bet is familiar: make it easy for publishers and other sellers to list differentiated products, then charge brands for the intelligence, access and governance needed to buy them.
The customer list explains why the company can attempt it. Its sustainability products have touched Mastercard, Sanofi, Coca-Cola and General Motors; agency relationships include GroupM and IPG; platform partners range from Microsoft’s MSN and Snap to Magnite and Adform. Those are different seats around the same table. A brand wants performance and control. An agency wants a workable planning layer. A publisher wants its good inventory recognized. A platform wants data that fits its existing pipes. Scope3's job is to make one set of instructions legible to all four without pretending their incentives are identical.
The proof is still young. Buyer access is waitlisted, public customer references for the agentic product are selective, and 26 storefronts are not the open internet. Incumbent platforms already possess enormous pools of demand and supply. Agencies will not evaporate; Scope3 says they can run agents for clients or alongside brand-owned agents. That accommodation is sensible, and also an admission that changing a market is harder than improving a workflow.
The thing to steal is not the agent
Scope3's most portable lesson is about sequencing. It started with an externality that almost everyone agreed was undesirable. It measured the problem granularly. That work revealed operational waste. It then turned the data from an annual reporting artifact into a campaign control, and the campaign control into a transaction system. Carbon was not discarded; it became one criterion among performance, safety, suitability and cost.
That sequence also explains when the approach will not work. It falters where delivery data is too coarse, where the emissions model lacks regional inputs, where a buyer cannot integrate agents into procurement and compliance, or where the available storefronts do not cover the audience. Small advertisers may not justify enterprise onboarding. Highly regulated brands may require human review so strict that the speed advantage shrinks. And a seller with undifferentiated inventory cannot train an agent into having an interesting product.
O'Kelley once helped make each impression biddable. Scope3 now argues that the better goal is to make each media product discoverable - understood for its context, audience and value before price flattens it into another rectangle. The company has raised enough money to test the claim and has enough ad-tech scar tissue to know what can go wrong. The next question is no longer whether two agents can buy an ad. They can. It is whether thousands of brands and publishers will trust them to decide why.