AI is exposing weak points in creative production, Williams Lea panel warns

AI can turn out creative work faster than ever, but organisations that fail to redesign the processes around it risk speeding up old problems and creating new ones.

That was a central theme of ‘The Role of AI in Creative Production’, a webinar hosted by Williams Lea.

The panel was made up of Mark Kuhillow, co-founder and chief executive of Trimontium ai; David Hamilton, director of business development at Williams Lea; and Amardeep Devadason, global head of marketing and CGO, creative and digital, at Williams Lea. They discussed how AI is being used in creative production and how the coming 18 to 24 months could affect for agencies, brands and production teams.

From experiment to production

The speakers said adoption had reached a turning point. Businesses can now build convincing prototypes quickly with widely available tools, but the harder task is turning those experiments into dependable systems that work at scale.

Kuhillow said the growing number of workable AI tools meant organisations now had a far clearer sense of what the technology could do.

“But what does that look like when it’s productionised now?” he asked. “How do we take those prototypes that people are creating within most organisations we’re talking to, and how do we pass them through scalability checks? How do we roll them out? How do we talk about internal adoption? How do we talk about governance?”

Making that move, he said, means looking past individual tools at the entire creative operating model.

Devadason said the sheer number of products on offer risked becoming a “problem of plenty” for creative teams. Teams should start by working out exactly where they need more speed, accuracy or capacity.

Tools built into platforms such as Adobe and Figma already help with day-to-day productivity, he said, but deeper change will come when AI is used to redesign workflows rather than just to speed up individual tasks.

“We are still in this place where it’s more about individual and team productivity,” he said. “How do you get a little more efficient? How do you do something fast? How do you bring something from concept to an asset in a reduced amount of time? That’s the larger question.”

Bottlenecks move downstream

Hamilton said much of the current debate centres on generative AI at the end of the production process. Producing assets faster brings limited benefit if briefing, approvals, compliance and quality assurance stay the same. Thousands of adaptations might be produced in days rather than weeks, only for clients and agencies to struggle to review them.

“If you turbocharge the engine at one end, is the rest of the process ready?” he said. “No, it’s not in my experience.”

The danger is a faster production engine feeding an operation without the capacity to cope, with the bottleneck shifting from making assets to checking, approving and distributing them.

Hamilton said AI could in time help with other parts of the workflow, for example by interrogating briefs, tagging assets or checking image licences, and these uses could turn out to matter as much as the generative tools that currently get most attention.

The panel said AI should ideally be part of a campaign from briefing and planning onwards. That becomes harder when a client is working with several agencies and production partners, especially if the production specialist is brought in only after the core creative has been signed off.

Devadason warned against bolting AI on late simply to cut the cost or time of a campaign already in progress. “You’ve got to be very, very specific about what it is that you want to do differently,” he said. “Or where it is that you need assistance.”

That also demands greater trust between brands and creative partners. Clients need clarity on where AI is used, how their data is treated and whether the technology will really make the work better.

Where it already helps

The panel singled out asset adaptation as one of the areas where the benefit is already clearest. After a campaign and its central idea have been signed off, AI can help adapt the work across formats, territories, channels and languages.

Campaigns that once needed dozens of assets may now need hundreds or thousands, and AI can help teams handle that volume as long as people keep firm control of creative direction and quality.

Kuhillow called AI “amazing at getting to a first pass”, allowing teams to move from idea to prototype at unprecedented speed. But its output can seem better than it is. Taking an illustrative 80/20 split, he said a piece of work could be mostly strong yet still contain weak or wrong material. The skill lies in spotting what is valuable and applying specialist expertise to improve it.

Devadason described AI as a “force multiplier” in creative development, letting teams go quickly from a rough sketch to a prototype or a digital mood board in short order.

That extra capacity puts more onus on creative directors, he said. Generating 150 possible ideas or executions achieves little unless someone uses their judgement to pick the best.

“Where I see it right now, the human expert has to be at the helm, and sort of orchestrating what’s being done,” he said.

Hamilton agreed, saying creative expertise would matter more as machines produced more of the work. “Put it in the hands of creative people, and it’s a hugely powerful tool,” he said.

The danger of sameness

Concern about weak, generic output ran through the discussion. The panel said careless adoption was already leading to content that looked, read and was structured in much the same way.

Kuhillow gave an example from his own business, which had received a long brief that had plainly been written by AI.

“We’ve had a number of briefs in this week, and one of them was very clearly an AI-generated brief, probably nine or ten pages deep,” he said.

“We’ve had to sit down as a team and actually try and deconstruct the brief, to rip out what the AI has put into the brief, to try and get to what a human being would have briefed us with in the first place.”

The case showed how apparent productivity gains can simply shift work from one organisation to another: the client produces a hefty brief quickly, while the recipient spends longer working out what it is for.

Devadason said the answer lay in the culture and standards of creative teams. “The use of AI has to be governed,” he said. “It has to be ethical. It has to be responsible, and at all points, it has to be used in context and not indiscriminately.”

Data and ‘shadow AI’

Data was identified as an important and often underestimated part of adoption. Kuhillow said access to the top large language models was becoming more like a utility, with organisations switching providers on performance and cost. That would make the quality and organisation of a company’s own information more important as a competitive edge.

He suggested agencies could find the change especially difficult. “Agencies don’t tend to be that great at process on the whole,” he said. “I think it’s a challenge for the industry because we’re asking it to do what it’s never done before. This genuinely is change management.”

Creative businesses may have built up years of knowledge about what works for their clients and audiences, but a lot of it is undocumented or sits with individual staff. AI cannot make consistent use of it until it is captured, organised and made accessible.

Hamilton said the businesses with the richest data and a willingness to spend time training models would be best placed to cut mistakes. That takes more than typing a brief into an off-the-shelf tool and expecting an instant result.

The panel said governance was struggling to keep up. Staff are already turning to public AI tools without approval, creating what Kuhillow called “shadow AI” inside organisations. He said leaders should look at how their teams use the technology, assess those uses and put suitable policies and technical safeguards in place.

The risks are greatest when agencies hold confidential client material. Material covered by a non-disclosure agreement could end up in an outside AI system, which raises questions about where it is stored and processed and whether it can be completely deleted.

The lack of a sovereign UK large language model complicates matters further. Kuhillow said that even when data was held securely in the UK, processing it could still involve technology from the US or China.

Hamilton said procurement teams were paying growing attention to the issue, especially in professional services and other regulated sectors, and AI policies were starting to appear in contracts and pitch processes.

Devadason recommended that experiments should begin in controlled environments, with proven uses then passing legal, ethical and operational checks before they are added to production workflows.

Will cheap AI last?

The panel also questioned whether today’s inexpensive access to powerful AI would continue. Kuhillow said the leading model providers were in effect subsidising adoption by offering large amounts of computing power for modest subscription fees.

“All of the major frontier models are loss leading in terms of what we’re paying for versus what we’re getting,” he said. “At some point, we’re going to be charged an appropriate amount for what we’re using, and that definitely will link into energy costs, and that will see a huge downward revision in terms of what people are actually using it for.”

If prices start to reflect the computing power and energy used, businesses may need to be choosier about what they generate and focus AI on areas where its value can be measured.

Looking ahead, the speakers expected AI to become embedded across creative and marketing operations, to the point where organisations may stop treating it as a separate category of technology.

Hamilton said the biggest efficiency gains would come from AI working across the whole creative process, from briefing and asset management to licensing, checks and approvals.

Kuhillow advised people to use the technology but remain critical of it. “Embrace change,” he said. “Embrace the tools, but just keep an eye open as to what they’re churning out.”

Williams Lea, part of RRD, provides technology-enabled business and marketing services. It employs 15,000 people, serves clients in 20 countries and has UK offices in Leeds and London.