Scope3 CEO and Co-founder Brian O’Kelley described it as a “universal custom algorithm platform” for custom algorithms across direct buys and via programmatic ads through SSP curation or deployment within a DSP. Although more significantly, it also shot across the bow of both the Walled Gardens and the brand safety giants.

The idea stems from wanting to find ways of using LLMs to add nuance to how brands express their content preferences for ad placements in more complex ways than traditional keyword blocking. On a flight to Japan last September, he built a prototype using a locally installed LLM, analyzing a major brand’s website to create a framework for AI-driven content classification. Testing it across AI models, he found it blocked 80% fewer articles than traditional keyword filters — avoiding false positives like “champagne lipstick” (which contains no alcohol) or mildly off-color celebrity jokes that wouldn’t actually harm a brand.

In an interview with Digiday, O’Kelley (a.k.a. one of ’the ‘godfathers of ad tech’) explained the idea behind Scope3’s new Brand Standards platform – which “delivers an approach that reimagines brand safety” – and lessons he’s applying from his previous company, AppNexus.

“I look at this as if every single ad should have an AI making a decision,” he said. 

“First of all, that’s computationally impossible today, but it won’t be at some point. So how do we bridge that gap? How do we take this sort of flow of buying some stuff, seeing how it does, learning from it, and pushing those learnings back into systems. That’s kind of the closed loop flow that an agent would do.”

How Scope3 uses the AI models

As a Google-focused platform, Scope3 has primarily used Google models – GV, formerly Google Ventures, has participated in two of Scope3’s investment rounds – including natural language processing tools, Gemini models and the cloud AI platform Vertex AI. However, O’Kelley said Scope3 isn’t tied to any particular model and continually tests and integrates multiple AI models, including Claude and Llama. However, he said AI model pricing and capabilities change all the time and are often disrupting each other, such as the recent introduction of DeepSeek.

“If I were building a model for my brand, I would want to make sure that we call exemplars [of] these particular kinds of content,” O’Kelley said. “We actually have a bunch of specific pieces of content where I personally have said I’m okay with that or I’m not okay with that. I can look at any new model against that.”

How the Brand Standards platform works

The first phase for a publisher is a brand safety analysis that checks for illegal content, offensive content, and other things that O’Kelley said are “clearly unsuitable for advertising at all.” Internally, the company calls it the “common sense brand safety process” because “anybody on inspection would say, ‘Yeah, that shouldn’t be by ads, this is not about nuance.’” For a publisher, this phase includes allowing Scope3 to analyze all its content, which can be shared through a stream of URLs with an API or uploaded in other ways. Scope3 also has a crawler to scrape content if needed.

The second phase involves the industry-wide vertical models — which also includes other categories like beauty and luxury — based on what the largest 50 or 100 brands or agencies might want to consider regarding brand suitability.

Scope3 also can create brand-specific custom models tailored to individual advertisers and agencies. The process then generates brand suitability scores that help publishers and advertisers understand the content and ad suitability. For example, if 20 of the brand models don’t like a piece of content, a publisher can look to see why that might be the case. They can also press a button to report something that seems potentially wrongly categorized.

“Broadly speaking, the idea is to be in this continuous learning loop so that all of these models get smarter over time as they see different kinds of content… One really cool part of that is there are no keywords; there are no categories. You don’t have to keep maintaining these complicated lists,” he said. 

When asked about competing with the brand safety giants of DoubleVerify and Integral Ad Science, O’Kelley quips that Scope3 has “full brand safety capability that is equivalent to what they’ve done.”

Another part of the plan to compete is through transparency, including page-level URL reporting for every piece of content where ads are served. “You can see those classifications, you know, in the UI, and we share those with the publishers as well so they can see how their content looks,” he said. 

Who are the initial partners?

For the Brand Standards platform, Scope3 is already integrated with The Trade Desk, Google’s DV360, and other demand-side platforms – “the DSP layer is really important,” said O’Kelley. It’s also integrating supply-side providers like Index Exchange and “almost every other major SSP.” Another launch partner for Brand Standards is Meta while Amazon Ads is the first DSP to integrate with Scope3’s Agentic Media platform.

Initial publisher partners include DotDash Meredith, although the plan is to roll out Brand Standards to the more than two dozen publisher partners it’s already partnered with on sustainability. “From our perspective, this is a way to help these publishers to actually understand and improve the way these ratings work,” he said.

How Scope3 plans to use AI agents

The way O’Kelley described it, Scope3 is “building an operating system for agents” that will handle tasks ranging from competitive research and audience analysis to data science and media execution and ad-checking. The goal is to create a platform that’ll be a place where agencies, advertisers and publishers can bring their own agents or build new ones depending on the task. 
“I sort of think of us as the AppNexus of agentic, which is maybe of maybe not the right way to think about it,” O’Kelley said. “AppNexus empowered an ecosystem of hundreds of innovators and ad tech to build incredible businesses. That’s kind of how I see this too.”

Scope3’s competitive moat? 

Scope3’s technology is complex and expensive, handling 2 million queries per second in the U.S. from Index Exchange alone — more than most DSPs. Each new partner, whether an SSP, DSP or publisher increases this volume “significantly.” The company processes high-fidelity signals, including URL and audio data, making replication difficult. Beyond the technical challenges, securing partnerships across the industry is complex. 

“It’s technically hard. I think about all the computation we’re doing at every request, plus bringing in all these URL content signals, audio signals, like we’re doing a huge amount of technical work,” O’Kelley said. “I also feel like we’re kind of inventing this, so it’d be hard to compete with us unless someone’s randomly inventing the same thing we are.”

Lessons from O’Kelley‘s AppNexus days? 

The key lesson from AppNexus is to choose partners and competitors wisely. O’Kelley highlighted Microsoft as a valuable ally, helping AppNexus gain a strategic foothold in digital advertising while generating significant revenue for both companies. Conversely, Yahoo’s lack of investment in Right Media allowed AppNexus to outpace them in RTB, capturing a major share of their business with superior technology. 

“And then we chose a bad enemy. We went after Google, and we did not win against Google. We competed valiantly and lost,” he said. “And by the way, it was a good loss. Like, we still sold the company for a lot of money, but I thought we could build a better ad server. I still think we built a better ad server. But it didn’t matter because the market conditions were not such that a superior product could actually win. We got our ass kicked.”