‘It’s not sexy, but it is magic’: why media agencies are infatuated with AI
Forget the pope’s fake puffer jacket and bad ChatGPT poems, AI is pulling off feats of “magic” at media agencies according to the people who work in them. Here are just a few of them.
Why media agencies are where the AI ‘magic’ is really happening / Adobe Stock
The internet is huge... really huge... hosting about a billion websites, with around four billion of us using smartphones and social media, while Google racks up some 8.5bn searches a day.
The advertising that happens around all of this, meanwhile, is growing to a scale beyond the comprehension of any human mind, consuming as much energy as a small nation – and that’s not counting digital advertising’s recent expansions into out-of-home, audio, gaming and TV.
With the help of AI, however, media agencies are beginning to find some clarity. These businesses have long leaned on digital tools for competitive advantages, with programmatic buying, for example, having accounted for more than 50% of display advertising in the UK since 2015. Many of its teething problems came from media agencies not having the time and labor to monitor where campaigns are running. As of today, AIs (or, to be more accurate, machine learning algorithms) can do that heavy lifting. Amid the buzz, AI is genuinely changing the business.
Kate Scott-Dawkins, global president of business intelligence at GroupM, credits the surge of consumer-facing AI tools such as ChatGPT and Dall-E for inspiring new ways of working. While people are creating prose, imagery, video and more with generative AI, agencies are using it to solve some of their biggest riddles, from targeting to measurement to planning and delivering campaigns. Advances in machine learning models, cloud computing and the supercharged data capabilities in agencies leave them in a strong position to find real efficiencies.
How AI will change media planning
Rich Astley, chief product officer at GroupM’s Nexus, says human media planners will remain, but acknowledges that the role is changing: “The days of being taken out for lunches are long, long gone.”
There will always be a place for people who can balance “a little bit of art and science, good judgment and experience,” he says, and AI should free them up to do more of what they do best. “Humans might be able to optimize 10 or 12 line items, but the AI can optimize hundreds of micro line items simultaneously.”
Agencies have been claiming to have ‘smart’ tech and ‘AI-driven’ products for years, but Astley promises that this time “it’s not BS” and that there have been “dramatic and astonishing” improvements in campaign performance. Partly, this is due to the fact they can be trained on what a successful campaign looks like and then emulate these good decisions. And it can make these decisions faster than any human.
So what’s the role of a media exec in all of this? It is still very early days with this tech, says Astley. “You don’t want to train the performance algorithm on the wrong data set... one of the biggest challenges is getting the source of truth that’s driving the store, sales return or downloads of an app.”
Many a human has run a bad campaign in pursuit of a KPI. A machine could be trained to pursue these too at the cost of everything else. Planners will need to interrogate the decisions made by the machine, which is where the latest innovations come in. The development of natural language processing, as seen with ChatGPT, means that what was once a black box piece of tech that could only be interrogated by the most skilled individuals can now be probed by almost anyone.
Diageo is using AI to optimize creative every three minutes, making sure it is fully suitable for each platform it will run on using a Creative Quality Score. It says this halved its CPMs. Meanwhile, all of Nestle’s creative is going through a similar process to ensure that everything going out meets a minimum level of consistency and doesn’t deviate from historic norms. At times, the application of AI can be more creative than the creative itself.
Suggested newsletters for you
AI is performing analytics magic
The third-party cookie is crumbling, raising targeting and attribution concerns. Agencies believe AIs will be able to once again tie the loop using highly accurate probabilistic guesswork.
Scott-Dawkins explains: “We’re doing more with synthetic audiences where we don’t need any data. You just need it to look like the population, it’ll be close enough.” A machine trained on enough ad campaigns, and with a steady enough flow of analytics signal can make it work.
Alex Campbell, from Publicis Groupe’s Performics technology and intelligence team, jokes that it has spent a “truckload” of money on making sure AI can “fill the gaps”. Campbell admits: “It’s not the sexy, ‘hey, let’s, you know, create images with Dall-E’, but it is media magic. You can use them in most situations where you historically would have used statistical models or sort of complex mathematical regression models.”
The likes of Meta or Google have been implementing machine learning into their ad businesses for years. The AI creep at these businesses is so advanced that Google arguably botched its launch of Bard because it was already an AI business.
“Danger signals will flash at some of the traditional agencies and holding companies now that AI enables clients to plan media directly with the platforms,” jokes Campbell. Someone needs to mark the platforms’ homework, however, and he thinks agencies will be able to interrogate these results better than anyone. “We definitely can’t say to a client, ‘just trust us’. We’ll need to work out increasingly sophisticated ways to validate and calibrate the modeling.”
Agencies have already run hundreds of thousands of campaigns through the tech giants’ ad networks. They have enough performance reports to form a probabilistic benchmark on anything they run going forward. “You can have a strong, accurate, responsive attribution answer without having to raise any privacy issues,” says Campbell.
These AI projects sound too good to be true. But the very fact networks have been so secretive about their innovations perhaps indicates there’s a competitive advantage to be gained. Campbell opens up on one such project with an unnamed client – a big one with a series of non-transactional websites that is dependent on in-store sales.
He says the sites serve to give product detail and can direct people to the closest venues, and that a great deal of ad spend goes into doing that, and that his team is working to correlate online analytics to in-store sales to better understand how it should plan its advertising campaigns.
It records analytics events across the sites and then collates offline sales. An AI then finds correlations between the two “inherently disconnected” tables before a second AI assigns a score or “strong business objective metric” to each event to predict what actions on what pages were most likely to contribute to the offline sale. This data is then fed into the ad bidding to ensure it is sending people to the right landing pages at the right times – and at the right price. It works really well, he boasts.
On the other hand, Astley warns: “If you’re using some kind of proxy metric to start informing models, you could find that you are driving the wrong kind of optimization.”
Dawkins, meanwhile, adds: “They have to be constantly updated. And ideally, they’re often also being constantly looked at to make sure that, you know, they’re not including bias.”
AIs, just like people, can be just as misinformed by bad data and bad objectives. Someone is going to have to keep a close eye on that.
The AI talent question
Scott-Dawkins wants to make it very clear that there’s still a need for human involvement. “I don’t think any of us are here of the opinion that media is going to be entirely by drones and bots.”
Seemingly every other week a report comes out citing a skills shortage in media. The experts we spoke to for this article all share the opinion that AI has the potential to lower the barrier to entry in the industry.
Campbell sees coding entering a new “strata” where machine learning systems can write code. “You can go from having an understanding of the core concepts of coding language right into coding if you can express in natural language what you want it to do,” he says. He calls this a “hell of a democratization… that massively speeds things up”.
And it won’t stop at media campaigns. AI has the potential to reform the agencies themselves. Astley wants to see WPP “being more like an operating system than a string of companies trying to work together”.
A media agency worker will never be shy in explaining how busy and stressful their roles are. At this juncture, the tech could make their lives easier and free them up for higher-level decisions. Hopefully, agencies will be smart enough to stop there and just see AI as another tool in the kit that requires skill to use properly.