When demand-side platform Viant looked to demonstrate how its new AI assistant can craft media strategies for programmatic buyers, it initially appeared to make a glaring mistake. One of the first prompts asked the AI for a hypothetical retargeting strategy. 

Viant CEO Tim Vanderhook had been putting the company’s AI tool through a pre-launch stress test with Ari Paparo, ex-CEO at DSP Beeswax and current CEO of consulting firm Marketecture Media, who asked for the retargeting strategy.

“When he said a retargeting campaign, in my head, I was like, ‘Oh, no,’” Vanderhook recalled. He had not run that scenario, yet, and Viant AI seemed to get confused, suggesting that this hypothetical marketer, a sneaker brand, should buy digital out-of-home billboards as part of the media plan. The problem was that digital out-of-home is typically not part of a retargeting plan, because a billboard doesn’t know whether a nearby consumer previously interacted with the brand.

“I think you just broke it,” Vanderhook told Paparo.

But since this was generative AI, Viant’s new GPT—generative pre-trained transformer—could explain the method behind its ad math. “It said, ‘You are correct, you can’t do traditional retargeting of tracking users to your website and later showing them ads,’” Vanderhook recalled, and, the AI assistant explained, “we bought digital-out-of-home billboards directly outside of their parking lots [of hundreds of stores] to retarget the consumers that have just left their stores.”

It appears you can do retargeting to digital out-of-home, Vanderhook said.

For Viant, the AI is just in its infancy, assisting in basic media strategy sessions, and these are the types of challenges when putting GPTs into the ad assembly line. Some ad technologists, Vanderhook among them, see a future of fully AI-enabled automation. It starts as a chatbot that can help brainstorm a media plan and, as it gets more intelligent, trained on more data and plugged into more platforms, AI could do ad buying, campaign optimization, personalization and measurement. Also, generative AI is working on ad creation. 

‘There’s probably a lot of confusion’

Many marketers remain undecided on the usefulness of large language models in advertising, debating whether AI enhances performance, but the technology will ultimately be essential, according to Tylynn Pettrey, VP of data science at Chalice.ai. Chalice works with advertisers to develop algorithms that power programmatic ad campaigns based on specialized intelligence, such as first-party data from the advertiser that helps build audiences for campaigns.

“A lot of people are skeptical that [generative AI] builds on current methods, that it provides a lot of value, and it can be expensive to run computationally,” Pettrey said. “So, there’s probably a lot of confusion around, ‘why would I use LLMs when what I’m doing isn’t performing poorly, or even if it is, could LLMs actually turn the performance around?’”

Pettrey added, however, that these will eventually prove out: “I don’t think there’s any way that you can compete without having some version of an LLM in the future.”

Also read: How AT&T trained its AI assistant

In September, Viant launched its GPT, becoming the latest ad tech company to adopt generative AI with plans to develop it further to handle more than media planning and ultimately execute programmatic campaigns, too, Vanderhook said. AI has a long-time role in digital advertising, which has used machine learning for years to power real-time bidding decisions and other operations within online auctions. Machine learning is a subset of AI that takes large data sets and conducts complex calculations, while generative AI can spit out new knowledge and communicate back to the end user. With the proliferation of generative AI, these tools are becoming more functional.

Advertising autonomy

Generative AI and GPTs have grown in popularity since the 2022 release of OpenAI’s ChatGPT, the text-based platform. Brands and ad agencies have been adopting these tools to build chat-based assistants that aid in marketing and other activities. The major platforms, including Google, Meta, Amazon, TikTok and Microsoft, have ad products that rely on machine learning and generative AI tools that help create ads. There also are AI startups, such as generative search platform Perplexity, developing their own ad businesses.

Ad tech companies, including Viant, are incorporating AI technology into their systems. In programmatic advertising, which is the automated buying of internet ads in real-time on the open web, AI could make it easier to create and execute campaigns.

Also read: How AI search is affecting marketing strategies

Viant put the GPT into the DSP, training it on knowledge such as data from the bid stream, which is the trail of digital advertising information from programmatic auctions. Last year, Cognitiv, a marketing tech company, launched ContextGPT, which seeks to identify valuable inventory online based on signals about the websites available to buy through DSPs. Chalice.ai uses open-source large language models and GPTs.

There are important distinctions between the capabilities of machine learning and those of large language models and the kinds of work they perform in advertising, according to Chalice’s Pettrey. “When you run [data] through an LLM, you get more granular insights than you would from a normal data provider,” Pettrey said. “Think about audience segments, you can see what audience segment is performing well in a typical programmatic buy, maybe it’s ‘book enthusiasts’ or ‘fitness enthusiasts,’ but you can’t see anything deeper than that. When you run an LLM, you can actually see individual descriptions of these people that could be your target audience.”

While some people consider “GPT” a catchy marketing term with uncertain value, Vanderhook sees AI as the future of programmatic advertising.

“We want to automate that whole thing, make it autonomous,” Vanderhook said. With human supervision, of course, Vanderhook added.

More information: Ad Age’s AI marketing glossary

Viant AI looks like a version of ChatGPT, a blank screen with a search bar waiting for the media buyer to ask for advice. The large language model is 65% based on ChatGPT 4o, the latest version of the large language model, and it also has access to Anthropic’s Claude 3.5 Sonnet large language model and open-source AI, Vanderhook said. The platform so far is designed to give buyers advice on their media plans—which properties to buy, how they should allocate budgets, what CPMs (cost per thousand impressions) they should expect on various channels.

Soon, Viant plans to roll out measurement capabilities and media buying to act on those media plans across the DSP. Viant recently licensed Comscore data, which gives buyers audience information about digital publishers.

“With measurement and analysis coming on board, it starts to become a daily use activity [for DSP clients],” Vanderhook said, “because you go to there and say, ‘How many impressions did I deliver yesterday or over the weekend? How many households did I reach? What was my click-through rate? What was my conversion rate?’”

More news: Inside advertising’s identity-based targeting crisis

AI at buyers’ fingertips

Ad tech platform Cognitiv has developed a GPT, called ContextGPT, that can assist in creating programmatic deal IDs, codes used to target internet campaigns, based on contextual signals. ContextGPT can handle tasks such as determining the appropriate websites for brands based on their media buying principles. For instance, its “inclusivity and diversity” feature analyzes websites to ensure they align with a brand’s DEI goals. This type of tool also helps brands avoid appearing on websites that could put them alongside negative content, or content that puts their brand in a bad light, said Marc Hudacsko, Cognitiv’s co-founder and chief technology officer.

ContextGPT is integrated with supply-side platforms, including PubMatic and Microsoft’s Xandr with more SSP partners coming soon, Hudaksco said. Through the SSPs, Cognitiv can curate inventory from multiple publishers.

“We’ve scraped all the pages on the internet that have advertising, we’ve fed them through Open AI’s large language model service, and we understand then the kinds of content on the page,” Hudaksco said. “Our tool that we’ve built is kind of like a Google search for brand relevance. You paste in your brand, your media brief, in that ‘Google’ search box, and what it spits out is the most relevant content on the internet that matches your brand according to this large language model.”

ChatGPT itself has some powers useful to media buyers, but creating custom advertising tools with this technology is even more effective, Hudacsko said.

“If you go into ChatGPT, you could say, ‘Write me a media brief for Nike,’ and then you could follow up with, ‘Hey, what are the websites I should be advertising on,’” Hudacsko said. “It actually has that kind of knowledge trapped inside of it.”

“Our job is to figure out how to extract that out and put it at the fingertips of media planners and buyers in a simple way,” Hudacsko said.

The term “AI” is now plastered on many ad tech products coming to market. Advertisers have seen Meta’s push with Advantage+, an automated campaign that uses machine learning to help set the goals, manage the bids and target audiences. Meta also has an open-source large language model called Llama, and generative AI that can construct ads in various formats. Google has similar ad campaigns, with generative AI tools, called Performance Max and DemandGen. The Trade Desk, a DSP, has also built AI tools in a media buying platform called Kokai. The Trade Desk also has an AI assistant that delivers media recommendations.

Promising the future

However, last year, The Trade Desk CEO cautioned advertisers about relying too much on models such as ChatGPT, which have been known to “hallucinate,” or generate bad information. “Regurgitating bad data, bad opinions or fake news, where AI-generated deepfake, for example, will be a problem that all generative AI will likely be dealing with for decades to come,” The Trade Desk CEO Jeff Green told Wall Street analysts during an earnings call last year. “We believe many of the novel AI use cases in market today will face challenges with monetization and copyright and data integrity or truth and scale.”

As for The Trade Desk, the platform looks at 10 million ad requests a second, giving it a “unique dataset” with which to train generative AI, which is “a promising part of our future,” Green also said at the time.

For ad tech company Moloco, which helps app developers and performance marketers with programmatic buys, machine learning is still the most common form of AI, while generative AI is still experimental, according to Beth Berger, Moloco’s VP and general manager, Americas. “Machine learning creates what we think is a multi-billion dollar opportunity to help with performance advertising in spaces like mobile apps and retail media and streaming media,” Berger said.

Over the next 18 months, Berger predicted, generative AI will become more embedded into programmatic platforms, and it could be beneficial in helping marketers develop audiences to target. With GPTs, media buyers can create personas that act like virtual versions of their consumers, and then use those personas to hone their marketing messages, Berger said.

“Ask what kinds of stories resonate with them, what types of problems they have … generate ideas about what types of creative might resonate,” Berger said. “And then, separately, use some of the generative AI tools to do things like creative production.”

Most platforms haven’t cracked these use cases to make them widely available, according to Berger.

“We are in such a hype cycle,” Berger said. “I have yet to find a truly transformational point solution, leveraging generative AI. What I have found is that as the industry experiments with these different point solutions, over time, I think we will start to see them integrate into the core platforms.”

For Viant, as the DSP continues to develop its GPT, Vanderhook sees generative AI playing a larger and larger role in programmatic ad markets.

“When we say autonomous advertising is going to do it all for you, with human oversight, because it’s not fully self-driving yet,” Vanderhook said. “It does hallucinate here and there. It does make mistakes, but we think we can do it all. That’s our North Star on the product.”