How Google’s AI ads take control of brands’ strategies

Google’s AI-powered ad product, known as Performance Max, is still mysterious to many marketers and brands experimenting with it are finding some surprising results. Sometimes the AI delivers bad leads and sometimes it reveals hidden search terms relevant for brands’ marketing. Take Brainlabs, a digital ad agency that works with more than 120 brands, which has been testing PMax with advertisers related to e-commerce, tech, auto parts, beauty and other categories. “Almost all of our clients are running PMax in some fashion,” said Jeremy Hull, chief product officer at Brainlabs, “some just a little bit, on top of [manual] search campaigns, and some are leaning in a lot.” Performance Max campaigns are similar to search ads—or a revved up version of search campaigns that extend to other reaches of the web—only Google decides how to target the ads. Instead of advertisers tinkering with keywords and managing where ads run, the tech does the work. The ads also run outside of search, on YouTube, Gmail, Maps and Discover, which is a mobile feed of Google content. Recent news: Google puts more search control in its AI ads At first, Google’s AI ads were a “scary” proposition, Hull said. “What their story was, is that you don’t need to worry about these details anymore,” Hull said of Google. “Search marketers did not like to hear that because we like some ability to point the system in directions and have control.” Check out the latest Ad Age events and award programs Last year, PMax was initially a dud for one brand, an online cybersecurity firm, Hull said, without revealing the name of the client. PMax was ideal for e-commerce and retail, but it was not good at lead generation for this B2B advertiser. When the brand turned on PMax, it did receive leads, which seemed like a goldmine for new business. But six weeks later, the client checked if those leads translated to sales. “The rate of unqualified and fraudulent inquiries went up,” Hull said. It seems PMax optimized for the “cheapest leads it could get, which in retrospect, to no one’s surprise over-indexed on low-quality leads.” To fix it, the agency taught Google’s AI, using anonymous internal customer data to train the model, Hull said. “You show it what good looks like, what a good customer looks like, and then it uses that to make better decisions,” Hull said. Google launched Performance Max in 2020 and the product has seen several upgrades this year that addressed some advertiser concerns. Google gave more controls to advertisers a month ago, with the inclusion of “search themes,” which let marketers specify interests they want to target. Google's AI is only as good as the data marketers feed it, and adding search themes to target, an advertiser can steer campaigns to subjects it might otherwise overlook. This year, Google also added the ability to exclude branded keywords so advertisers could tell the system not to run against particular search terms related directly to their brands. And Google recently unleashed generative AI to create the ads—writing the copy and building images with different products and backgrounds. “PMax is the best example of the manifestation of AI in our systems,” said Brendon Kraham, Google’s VP of search and commerce for ads, in a recent interview. “[It’s] the only campaign we have across the surfaces and across Google inventory that uses the power of AI to find those untapped and incremental conversions.” Automation everywhere Google does not reveal how much revenue PMax generates, but it is a core part of its future. Google made $44 billion in search and other ad revenue in the third quarter, attributing an 11% year-over-year rise in revenue to “growth in retail,” according to its October financial report. Retail and commerce are strong sectors for PMax adoption, according to ad experts. PMax is a sign of the growing reliance on AI in advertising, with platforms such as Google, Meta, Amazon and TikTok turning to such automation. The new mechanics require advertisers to relinquish power over their campaigns, which run in settings chosen by algorithms and target audiences based on machine learning decisions. The machines also handle the bidding in ad auctions. Marketers need new methods of measuring success, too. “We certainly did have to rethink how we did [campaign] reporting to align with the goals that the customer is looking to achieve,” Kraham said. Google has said that advertisers that used PMax saw 18% higher conversion rates, which measures outcomes such as sales and leads from ads, compared to advertisers outside PMax. “Tractor Supply saw a 53% increase in omnichannel revenue as a result of using PMax,” Kraham said. “Volkswagen [EMEA], for more of a lead-generating perspective, was able to increase cost-per-lead by 48%.” “The ad ecosystem is complicated, even here at Google we have 11 campaign types,” Kraham said, “and it takes at least four to maximize reach and even then, still there could be gaps.” “Cross-channel campaigns have become a new concept,” Kraham said, “this does require a little bit of new thinking and so there is that kind of discussion that we have with customers. There is goal orientation and optimizing for what is the exact specific thing that they want and then leveraging the AI, the predictive and the generative, to go drive performance. It is very different than manual campaigns in how you optimize.” Not every type of brand or large ad agency is sold on PMax, however. There are drawbacks to having Google place all the ads, including on YouTube. The brands don’t want to overlap on YouTube, when they’re already managing campaigns there, according to one ad executive, who spoke with Ad Age on condition of anonymity. There also are issues of controls, where brands want to guide their ads to safe websites. PMax lacks tools Google offers in other ad products to avoid unsuitable content, or it can be harder to opt-out of certain online settings. The point of the product is to generate clicks, optimized to appear wherever Google’s AI detects a high-value impression. Just yesterday, research firm Adalytics reported that search ads were appearing on unsavory sites that used Google’s third-party tools for their search experiences. When it conducted searches on the unsuitable sites, Adalytics said it saw paid search results from major brands and other entities. The report also referenced PMax campaigns, saying Google does not share details with advertisers about where ads appear. Google has pushed back on previous Adalytics reports that found ads running on low-quality websites, saying the group has mischaracterized its ad platform and controls. To counter the latest report, Google said that Adalytics found sites that represent a miniscule portion of its third-party search program. However, there are still concerns among advertisers that more automation and AI in internet advertising mean less visibility into the ecosystem. Meanwhile, not all brands, for instance, consumer goods brands, are set up for performance marketing, and they aren’t looking for immediate sales or actions from their ads. For those types of brands, Google has launched Demand Gen, which is for social media-style marketers with “consideration” goals in their campaigns. These brands want to farm new customers who could later convert to sales, and Mondēlez, Samsung and Toyota have been interested in Demand Gen. For Google, all these products work together, with traditional search, PMax and Demand Gen feeding into one overarching internet ad strategy. Keyword keys For other brands though, the automation of complex ad systems is too easy to pass up. “PMax is an extremely efficient way of running a paid acquisition program, just from an effort and personnel standpoint,” said Sib Mahapatra, co-founder and chief product officer at office furniture retailer Branch. The company generates annual revenue in the “high eight figures,” Mahapatra said, and the bulk of its marketing strategy goes to Google search. Branch still needs to manage some of its more niche search campaigns manually, but with its “core DTC acquisition program on Google, most of our spend is going on PMax at this point,” Mahapatra said. There are drawbacks to letting PMax drive all the decisions for search campaigns. Search advertising is all about testing the best keywords to drive customers to a site, and then retargeting those consumers with follow-up ads. Branch wants to run ads when consumers search specifically for its brand and when they search for generic terms related to furniture. In manual search campaigns, brands bid differently for branded and non-branded keywords depending on how valuable they are to the business. If a consumer is already searching for the brand specifically, then that ad is not as valuable because the customer was predisposed to buying anyway, Mahapatra said. “The one downside from a measurement or incrementality perspective for PMax is that it includes branded keywords,” Mahapatra said. “Branded keywords are critical to run but we want to break out, to some extent, the impact of branded versus non-branded. That’s all rolled together in PMax.” To account for that, Branch sets a higher “return on ad spend” goal for PMax campaigns, Mahapatra said, so the machine potentially avoids lower-value searches. Branch can see that PMax is outperforming manual campaigns, and the campaigns are leading to sales, Mahapatra said. “You can run some pretty simple tests,” Mahapatra said, “you basically have to just figure out does the revenue actually relate to the return on ad spend that’s being reported by Google.” The advice for brands is to run PMax alongside manual search campaigns, Brainlabs’ Hull said. With the exclusion tool Google launched this year, advertisers can tell PMax to avoid branded keywords, and then marketers can bid on those manually. Brainlabs manages advertising for an e-commerce brand that wasn’t ready to try PMax last year over the holiday sales period, Hull said, but this year will spend $4.5 million in November. Brainlabs declined to name the brand, but said it learned new marketing tips by studying PMax. The AI found that keywords related to “pumpkin spice” were effective for the e-commerce retailer. “Of course, pumpkin spice is a trend, but they never went after it before,” Hull said. More from Ad Age: Why pumpkin spice still resonates Google’s AI picked up on the fact that people interested in pumpkin spice content were also interested in this brand, even though none of its products are related to pumpkin spice. “Where PMax sees a spike in sales it responds to that shift quickly when it sees success,” Hull said. “It isn’t as transparent, and it doesn’t ask permission,” Hull said, “but if you look at conversion data you have access to, you can find something that PMax found and acted on.”

How Google’s AI ads take control of brands’ strategies

Featured Stories