AI is not a magical panacea … it’s a scalpel

Four years ago, WPP acquired Satalia, an AI-tech company co-founded by globally recognised AI expert Dr Daniel Hulme, who soon became WPP’s chief AI officer. His role, much like this magazine’s, is to demystify AI to boost growth - not just for agency leaders - but for clients, partners, governments and media outlets across the world. Stretching beyond WPP’s acquisition in 2021, Hulme has been involved with AI for 25 years, all the while building solutions that use algorithms across not just marketing, but entire supply chains. It is safe to assume then, that Hulme has vast amounts of experience in understanding where organisations get technological solutions right - and where they get it wrong.

Between ten and fifteen years ago, the shiny new tech was machine learning and data science. Data lakes were built, machine learning experts were hired to extract insights from data, and leaders therefore expected better decisions to follow. “I could’ve told you back then that wasn’t going to work,” Hulme adds.

Now, of course, there is a tremendous amount of excitement, not to mention investment, in AI as a magical gift to marketing. “It’s an incredibly powerful tool - but typically a lot of organisations try to apply it to solve the wrong problems. It’s not a panacea that’s going to solve everything. It’s important to understand what the right technologies are for the specific frictional challenges ahead.”

Satalia, which now has 300 AI experts working within WPP, prides itself on identifying and understanding the right problems to solve across retail and marketing supply chains, and then building scalable algorithmic solutions to drive growth.

For Hulme, there are several key areas in marketing where you can apply these algorithms. One is dynamic segment identification - identifying emerging segments - which requires machine learning and data science algorithms. Another is channel optimisation - allocating content across channels to maximise return - which requires optimisation algorithms. Generative AI algorithms come into play with content generation, using Large Language Models (LLMs) to create generic content.

But building LLMs that create brand-specific, production-grade, differentiated content is “extremely difficult,” Hulme says. As well as this, AI can help with audience perception. WPP has created synthetic “audience brains” that replicate how people think and feel about content, helping create more resonant work and predict activation such as clicks, likes, comments or sales.

The emphasis for WPP is on ensuring that it is differentiated in solving problems within these areas and applying the right combination of algorithms to each. The company’s AI marketing platform WPP Open Pro supports this by connecting professionals, data, tools and AI in one place to deliver better-informed creative ideas faster, at scale and at lower cost. 

WPP Open Pro, and AI more broadly, are also transforming how workforces operate across disciplines within WPP. “Software once took years to build in waterfall-style, siloed teams. Agile changed that by bringing designers, developers and testers together to move fast. Now AI takes it further - we can build augmented teams that include AI beings, like an AI philosopher or historian, working alongside humans.”

Hulme’s mission is to bring the power of AI to everyone - safely, responsibly and with purpose. That is where WPP Open comes in: a curated ecosystem that ensures the right models are used in the right places, avoiding the “Wild West” of unchecked AI use. “The more people use this end-to-end system, the more we learn from them and the more the whole experience of the system improves.”

WPP’s dual strategy, he explains, is to empower people to use AI independently - giving them “superpowers” - while also being strategic and selective about where to build scalable, expert-led solutions across the supply chain. As Hulme puts it, “Quick wins and low-hanging fruit don’t differentiate a business - true differentiation takes time and depth.”

In many ways, the job of marketing is to reinforce unconscious associations with brands. AI is getting better at creating these connections, Hulme says. It is also improving campaign generation.

However, Hulme recognises that coming up with content that connects a brand to what someone truly values is extremely difficult as it requires empathy, macro thinking and sometimes misdirection - “skills and capabilities that are outside the realm of AI”. He knows that AI cannot replace creatives, but he believes that it can help them create hundreds of effective and award winning campaigns.

“In a world where you can create content rapidly, test it, predict activation and push it across channels - the key differentiator is creativity. It’s really about pushing the boundaries of creativity.”

When thinking about the creative process, Hulme notes that creatives broadly use around seven story arcs and 40 templates or archetypes. AI struggles to invent new iterations of these. “Generative AI itself might be one - we’ve seen that in our J Lo and Shah Rukh Khan campaigns.” But by using AI, creatives can now multiply their own story arcs and archetypes with media types and audiences, and end up with millions of potential creative ideas. Many of them, Hulme says, turn out to be nonsense - but the challenge for humans is navigating this space to find the excellent ideas.

“With regards to myths, people often think AI is magic and can replace humans. In reality, it’s extremely hard to get models to that expert level - and even then, you need a lot of human input. And yes, there’s a lot of emphasis around AI being used for productivity improvements - but the real opportunity is providing digital solutions that stand you apart from your competitors.” 

To get real differentiation, agencies will have to tune and train models into genuine domain experts - “and that takes months of exploration with experts and the right data,” he says. “It’s still very much an art. If you do it right, you can create expert agents that augment and empower your workforce. You can have an AI that’s a creative genius or an AI that’s great at predicting activation.”  

 What excites Hulme now is that clients can test tools against each other and see which ones are best, for example, at performance prediction. “Historically, we took old bait, but now we can actually demonstrate that value.”