Agencies: AI-powered advertising is efficient but ‘mediocre’

Acadia chief executive Jared Belsky says brands were suspicious about Performance Max when it first rolled out, but have let go of their inhibitions following the proliferation of AI tools like ChatGPT.

The tech company said in its first quarter earnings in April that daily revenue from Advantage+ Shopping campaigns increased seven-fold in six months. 

“Our faith in technology to drive more of those decisions about which ad is going to be most relevant and rewarding to people and how best to spend dollars is absolutely going to increase,” says Mark D’Arcy, partner and chief creative officer of The Brandtech Group.

Ads powered by AI often outperform on performance KPIs because they are highly personalised and adaptive. According to Google, advertisers who shifted from Standard Shopping campaigns to Performance Max between October 2022 and March 2023 drove an average 25% increase in conversion value at a similar return on ad spend (ROAS). Meta said advertisers who use Advantage+ shopping campaigns recorded a 32% increase in ROAS in recent tests.

“They're more relevant because they are based on a previous visit or other insights. Personalisation usually powers performance,” says Will Phipps, media lead at Allen & Gerritsen (A&G).

“An algorithm that can bid while you're sleeping is inherently better than you,” adds Belsky.

But for advertisers and agencies, surrendering control to technology comes with risks. 

Sacrificing insights and quality

While tools like Advantage+ and Performance Max can help advertisers efficiently achieve their performance goals, such as hitting sales targets or acquiring leads, it can be difficult to draw insights on how that goal was achieved.

These tools are most useful to smaller advertisers who may not have the budget or resources to craft a complex media plan on their own, says D’Arcy.

“It's this trade off,” adds Phipps. “We find Performance Max can work really well for some clients, but the problem is, we don't know why it's working well. I can tell a client that I sold their product, but I can't tell them if it was using display, or video or Gmail. So how do you learn from it? I feel like we're outsourcing our analytics and intelligence.”

Google provides stats on how Performance Max ads perform via “asset groups” reporting, which shows how certain campaigns drive actions like clicks, impressions and conversions. But it does not break down how the tool allocates budgets to different channels and audiences, which is easier to track in its more manual Standard Shopping campaigns.

In a statement, a Google spokesperson said: “We’ve continued to build new Performance Max features directly based on advertiser feedback, including new insights like asset group reporting and audience insights which help advertisers understand how to improve performance.” 

Meta similarly provides performance results by audience type in Advantage+.

The quality of leads and the creative generated by these AI-powered tools can also be questionable, according to Phipps.

Optimising on cost can mean appearing on obscure websites or serving ads in the middle of the night when the cost per action is lower.

“Google will do its very best to generate more leads, but it doesn't know if they're junk,” says Phipps.

Belsky adds: “Performance Max is making the choices in ratios that you may not have wanted, and to get the good placements, you have to get some of the bad placements.”

Advantage+ doesn’t allow advertisers to control targeting beyond country, so they may inadvertently end up serving ads to irrelevant users. 

Creative quality can also suffer. For instance, Meta offers dynamic creative optimisation (DCO) within Advantage+ shopping campaigns, which reformats ads and descriptions to fit different ad types and automates up to 150 creative combinations at once. But the result can sometimes end up looking like a badly cropped image, according to Phipps. 

“We've switched off most auto placements on social because within 10 minutes a client's calling you saying, ‘why does my ad look so bad?’ and it’s because it was auto formatted incorrectly.”

“Meta’s automated placements are on the naughty step, and we'll welcome them off the naughty step once they prove themselves better able to do it,” he adds.

New generative AI-powered image and video editing tools, including those offered by Adobe, Google and Meta, promise to fix many of the bugs in DCO, but in early demos they produce simplistic and templated work that is inferior to what humans can create, the agency leaders believe.

“Mediocrity is free,” says D’Arcy. “I don't mean that pejoratively. AI will absolutely generate something that will be a five out of 10, something totally fine. Our job as humans is to know that, accept that and really focus on excellence and not volume.”

As more assets are created by AI, quality control becomes harder. While Google has said it will offer advertisers the ability to check AI-generated copy and images before their campaigns goes live, Phipps expects these checks to fall by the wayside as volume ramps up.

“Do you want to review 1,000 different variations for a piece of work?” he asks.

Meta said advertisers are able to see a preview of assets created by Advantage+ but not the final ads that go live.

“We're moving from AI assist to AI control,” Phipps adds. “It's like removing the edit room from a newspaper. The cognitive check is being given away — and I don't think the trade off is worth it yet.”

Advertisers risk damaging their brands if their experience is not cohesive, says D’Arcy. “If you do something in order to create an action on a digital surface that's tasteless and poorly designed, you damage your brand, and if you do that over millions of impressions, that's expensive.”

Humans and machines

Every agency that is investing in AI products and partnerships is also arguing for it to be used as a “co-pilot” or “assistant” to humans.

AI chatbots like OpenAI’s ChatGPT are powerful tools that can be used to conduct research, generate ideas and produce copy, but the quality of their responses depends on the human steering the queries.

“You could ask ChatGPT to write you a strategy and it would be vanilla, the summation of the average, number 50 in the Billboard charts. It needs someone to have a vision for it,” says Phipps. 

Belsky echoes: “You still need an idea, you still need a hook, you still need a strategy — then the machine can do the iterations.” 

If everyone has access to the same tools, it is the “quirkiness of our humanity and our curiosity” that will set agencies and brands apart from one another, adds D’Arcy.

This narrative is critical to agencies as AI tools can be used for many of the ad services that they offer to clients. Automation is predicted to shrink the U.S. agency workforce by 7.5% within the next seven years, according to Forrester.

But every new technological advancement also presents an opportunity to create new services. For example, the Brandtech Group is focused on developing tools and insights that work across multiple platforms.

Another way to provide value is by utilising first party data to make smarter decisions.

“If I have more data on my audience and I know how to handle it skillfully, I can win, because I'm going to be feeding the algorithm smarter data upfront to find those customers, and I'm going to be making better choices of how much a click is worth,” says Belsky. Acadia gained SOC 2 compliance so it can handle personal data.