Agencies Need to Stop ‘Thinking Big’ About AI

‘Thinking big’ is the biggest mistake agency can make when thinking about artificial intelligence right now.

Instead, your agency should employ a tactical and focused approach which tests and learns with AI in small tribes and units, last Thursday's event heard.

Fergus Dyer-Smith, founder of video-production company Wooshi, described “analysis paralysis” as a blocker to companies making meaningful strides in AI. Three major challenges are typically to blame: a failure to obtain the right permissions, an inability to bring about collaboration, and difficulties with creating a common language when a company has multiple brands.

Wooshi’s client, the Canadian bedding products giant Sleep Country, fell victim to this when trying to plan for a system-wide approach to integrating AI, Dyer-Smith revealed.

“You’ve got too many people who are sold the promise, they have a little bit of a dibble and it doesn’t quite work, or they haven’t got the time to do it,” he said. “And then the word gets out that you should just forget it and go back to the old system.”

Instead, agencies should begin with small, discrete areas in which to integrate AI, while being mindful of how that smaller project could (rather than should) develop into a wider approach across their business.

‘Just get going’

The talk just happened to be hosted by LinkedIn (owned by Microsoft, a global AI leader that backs the wildly popular Chat-GPT service.

All the more reason, now that the AI hype cycle is quickly maturing, to look at practical ways in which the tech is meaningfully going to help agencies which generally faced a tough business climate in 2024 and are eager for quick wins.

Dyer-Smith highlighted the importance of identifying both obvious and hidden workflows, including those managed through tools like spreadsheets, by individuals and small teams within the agency.

This is what Publicis Groupe’s B2B specialist agency Publicis Pro did, according to its chief strategy officer Billy Hamilton-Stent. The agency was rebranded last year, three years after Publicis acquired Octopus Group in 2021.

“Just get going,” Hamilton-Stent advised. “It's not a bunch of money, not in the real scheme of things, [and] the goalposts are moving on a daily basis.”

But, he warned, agency leaders must be very mindful of how AI will impact on agency culture.

That culture must include a willingness to test and learn when a potentially transformative tool like AI comes along, as well as committing to transparency. This is particularly important for smaller or independent agencies, which are typically built on strong foundations of culture, whether they are company values, distinct methodologies, or founder-led ways of working.

And, over time, it will become important for the agency to develop a unified point of view on how it is thinking about AI.

‘Vision’ means treating AI as part of the agency’s fundamental offer, instead of simply estimating time-savings from outsourcing or financial ROI of software investment.

Being able to push limits in a ‘safe space’

Surely if any business can deliver this, it’s a creative industries business?

Dyer-Smith said getting an AI “vision” right is important, particularly for a creative industries business that employs creative and strategy specialists that can plan for the future more effectively than extrapolate from past events (which is what most human beings do).

“[Agencies] are asked to do by our customers is often create things that don't exist is to project out alternative realities…. Use all that creative force that you have within your building,” he added.

While Publicis Pro is now allowing the agency to use AI “at will” currently, Hamilton-Stent stressed the importance of installing “guardrails”.

Since there is a much higher potential for disruption and business transformation from AI that goes far beyond single-use applications, the risk-reward tradeoff may be huge. But this step-change only happens if people are allowed to push the limits of AI within a “safe space”.

“[AI] is not like marketing automation, it’s not like email,” he said. “It’s a bit like when mobile phones came in as a ‘tech wave’. You might have to go back to the combustion engine to figure something that has this kind of multidimensional impact on what we do.

“If you think about it like ‘we’re about to roll out something, you’ve missed a trick.”

AI is not a ‘pivot’

Here comes the paradox: it’s the agencies who do develop a strategic offer built on AI that will reap the biggest gains in the coming years.

And yet not many marketing-services business are doing this right now. according to Paul Winterflood, a senior media and marketing analyst from accountancy Moore Kingston-Smith,

While around half (51%) of agencies are using AI to increase operational efficiencies, research revealed by Winterflood also showed less than a third (29%) are using AI to improve client outcomes.

Another quarter are using AI for content creation (and, of course, we’ll get to that at the end).

"If you've invested in and developed an AI platform that's got potential to create scalable returns, then venture-capital investment is definitely possible, but there's a clear risk involved in making a pivot in that,” Winterflood said.

But, he then warned: “To attract VC investment, you need to demonstrate that you can scale it and deliver over 10 times return for an investor. For the vast majority of agencies, that's that type of pivot is probably not going to be possible. So it's more about how can you use these tools to support your capabilities.”

Anyone who plays cribbage will recognise the concept of ‘shooting the moon’. There will be certain businesses who will win big by betting heavily on AI (David Jones’ Brandtech Group immediately comes to mind.

But for most agencies, any AI strategy is likely to be optimised by the ‘evolution, not revolution’ mantra.

Excuses for inaction

Even if none of the above were happening, there’s another practical reason agencies need to be aware of how to integrate AI into their businesses: their clients may be on this journey, too.

That’s certainly the case at Mars, according to its global digital commerce director Swagat Choudhury, who emphasised the need for continuous learning and upskilling to adapt to the evolving role of AI in the workplace. Choudhury predicts that 90% of marketing will eventually be AI-driven and play huge roles in data analytics, video creation and execution.

That’s all very well, but how do we mitigate the biases which are inherent in the algorithms that create AI models, or the way all that data is collected to train the models?

Choudhury wonders whether a fear of biases leads to “excuses” with which to “write off” automated technology.

“You will get these distortions, or biases, but the good part is that you can correct them when you scale it up, because the distortion was always there in humans as well, right? If a machine is giving you news that you have to look at with a little scepticism, that was there with humans as well—there are editors and publishers who have different biases.”

Me? Biased?

Anyone who reads this should naturally consider the biases of this article’s author.

Journalists are trained to aim for objectivity in reporting, while ignoring the uncomfortable truth that our inelegant minds are held together like wrinkled Sellotape by myriad biases and prejudices.

Perhaps AI tools will get better at recording meeting transcripts and providing key quotes, or even summarising whole speeches. But, consider what would be lost had I not been in the room with my eyes and ears. The AI stenographer would be unable to sense emphasis in the speakers’ voices, the revealing eye movements and facial expressions, or the reactions of the audience.

Nor would the AI choose to stray into indulgence by commenting on his own limitations, just because he thought the reader might find it interesting. Another small-scale gamble.

And then we remember two key facts of life: people buy people, and around 70% of human communication is estimated to be nonverbal.

People are irrational. Can an AI be trained to do this by feeding it large volumes of text and numbers? Smaller and independent agencies, I would guess, are more sensitive to this than those part of conglomerates that have access to scaled automated systems.

So, keep AI adoption simple and discreet. It might have the power to change the world in countless ways, even if none of those ways matters all that much right now.