How brands are using custom AI to improve programmatic ad buying

Major brands are developing custom AI algorithms to improve their chances in programmatic ad markets as new tools make it easier—and cheaper—to tune large language models, according to ad tech and brand leaders. For instance, Lenovo and Sargento both have used custom algorithms, built on top of AI, to test digital campaigns, which they said helped identify new target audiences and higher-quality ad inventory. The brands are part of a growing push in the ad community to hand over more decision-making to AI, and ad tech startups are cropping up to serve these brands, including Chalice AI, Scibids, and Cognitiv. Advertisers see that AI technology is within their reach, especially after recent breakthroughs such as DeepSeek AI out of China. Last month, the team in China publicized the fact that it could train sophisticated AI models more cheaply than previously considered possible. “The advertisers are understanding how powerful these models are and how capital light they can be,” said Adam Heimlich, founder and CEO of Chalice AI. The ability to develop custom algorithms, frees brands from “the last mile” of the ad supply chain, which is typically controlled by platforms, such as Meta and Google, Heimlich said. The platforms all have their own AI algorithms that handle the customer matching and fulfill the goals of advertisers. But brands “can apply their own models that are built from their own data for their own outcomes to be deployed in any platform they want,” Heimlich said. The custom algorithms in programmatic markets often work with demand-side platforms, such as The Trade Desk, to discover ad inventory, and they are increasingly plugging in directly to the supply side, right into the publishers, to get a vast view of online ad impressions. NextGen Marketing Summit Join us for a crash course on Gen Z, DTC marketing and the creator economy Register today Lenovo’s AI race Lenovo has been testing campaigns running an algorithm developed through Chalice. The brand also has conducted overseas campaigns through Scibids, an AI ad tech platform that digital media measurement firm DoubleVerify bought in 2023. Lenovo also has worked with Seedtag, a contextual advertising firm that uses machine learning to identify relevant publishers for advertisers. Meanwhile, Sargento recently worked with Cognitiv, which uses custom-trained AI to target ads in programmatic markets. Programmatic has always used machine learning to help target ads, but the rise of generative AI and the sophistication of large language models increased brands’ ability to analyze web content more deeply, according to ad tech and brand leaders. For instance, Lenovo attributed a 270% increase in its return on investments in ad spending to Scibids’ AI. The cost per impressions were 34% higher, a Lenovo spokesperson told Ad Age, but it “generated solid revenue for the campaign, proving effective for conversion optimization.” Lenovo has been disseminating AI tools across its 500-plus-person marketing team, through enterprise deals with platforms such as OpenAI, developer of ChatGPT, according to Emily Ketchen, Lenovo’s chief marketing officer and VP for the Intelligent Devices Group and International Sales Organization. The embrace of AI has filtered into its programmatic strategy. Also read: 20 ad tech executives make 2025 predictions Lenovo has a major Formula 1 sponsorship as the racing league’s “official technology partner.” The league uses Lenovo’s tech to deliver deeper data and analyses to racing teams, and that helps generate more content for fans. For Lenovo’s marketing team, it wants to target more organizations that could adopt its technology through B2B advertising. Lenovo’s digital media strategy looks for audiences that overlap with online F1 fandom, according to Rick Corteville, Lenovo’s executive lead, global media Center of Excellence. “We could work with [custom algorithms] to auto-optimize the delivery of creative to a core audience,” Corteville told Ad Age, adding that Lenovo wanted to hit “more of those business decision-makers, but also people who were actively reading F1 content.” Lenovo has an account list of about 500 companies it is targeting with digital ads but typically it might only hit 50 of those companies through traditional programmatic, Corteville said. Services such as Chalice AI help identify ad inventory that could reach the rest of that target audience, Corteville said. Lenovo is a heavy digital media spender, according to MediaRadar. Lenovo Group’s most recent data, covering the first 3 quarters of 2024, showed it spent $150 million on digital media in the U.S., whereas its total measured media in the U.S. for that timeframe was $153 million, according to MediaRadar. Supply overfloweth Chalice has emerged as a high-profile player in this new programmatic AI sector, and it counts ad tech veterans among its investors—Brian O’Kelly, CEO of demand-side platform Scope3, and Jeff Green, CEO of The Trade Desk. Chalice has those demand-side relationships, but it is increasingly striking deals with publishers and building its own supply-side ties to access ad inventory. “We’re literally living in the supply side, next to the unthrottled stream of bid requests, something like a half trillion a day,” Heimlich said. Chalice’s website shows work with Hershey, HP and automaker Genesis. Chalice also recently hired Ethan Kuehl as a senior machine learning engineer; Kuehl previously worked in a similar role for Walmart’s global tech team. Kuehl works with open-source large language models, which are publicly available, to train the algorithms that identify new audience segments for brands. “Marketing teams can actually have a more in-depth view of their given audience for a particular campaign,” Kuehl told Ad Age. “[Large language models] can actually help us define new segments and audience members, as well as enable us to really monitor how campaigns are doing among groups that were maybe traditionally overlooked in a specific campaign,” Kuehl said. AI advances, especially since the release of OpenAI’s ChatGPT in late 2022, made it faster to ingest multimedia—text, images and videos—from websites and interpret that data for advertising purposes, Heimlich said. One example of an insight derived from these techniques: Chalice discovered that lenses customers of 1800Contacts highly overlapped with consumers who got their cars fixed at independent auto shops, Heimlich said. That type of insight could shape the media strategy, Heimlich said. Sargento, the packaged food brand, recently conducted tests with Cognitiv, which uses similar custom AI models to search for new customers. Sargento wanted to get new customers for its Balanced Breaks brand, according to a case study from Cognitiv. Sargento declined to comment for this story, but Cognitiv shared the case study for the campaign. Cognitiv created a custom deep learning model trained on Sargento sales data via Circana, the consumer research and analytics firm. “The model was designed to identify potential non-category buyers who exhibited similar purchasing behaviors to existing customers but had not yet made a purchase in the Balanced Breaks category,” Cognitiv said in its case study. Custom advantage Of course, it’s not new for brands to plumb data and use machine learning to find “lookalike” audiences that resemble their typical consumers. But AI enthusiasts in the marketing community have said the availability of these tools makes it easier for brands to create custom algorithms, bespoke for their needs. On the platforms, such as Google, Meta and Amazon, all the brands rely on the same algorithms, giving no specific advantage, media buyers and ad leaders told Ad Age. Custom algorithms are changing that equation. Sargento spent $20 million on advertising in the U.S., $6 million of which went toward digital channels, in the first three quarters of 2024, according to the latest data from MediaRadar. Lenovo’s use of AI in digital advertising will only increase as the technology evolves from pre-trained algorithms that plug into programmatic markets to agentic AI. The hope is that AI agents could eventually take over more of the media planning and buying strategies, Corteville said. “We could maybe tell, you know, a dancing paper clip within The Trade Desk exactly who we’re looking to go after, and the [key performance indicators] that we have, and for them to help build that campaign and execute it for us,” Corteville said, jokingly referring to Microsoft’s “Clippy” as an example of a virtual assistant. “So that the setup, and the time to get that campaign live, will get a lot shorter. That is more futuristic and down the line.”

How brands are using custom AI to improve programmatic ad buying

Featured Stories