Though difficult to assign a dollar value against, each of them represent an underwater rock that might hole an AI-assisted creative team below the waterline.

58% of marketers are using generative AI for content production, according to an October Gartner survey of 400 global practitioners. Semi-automated production systems like that pioneered by Unilever will be the goal of many advertisers — but setting up such an assembly line took over a year.

“AI [production] is the equivalent of building your own house instead of renting someone else’s,” said Craig Elimeliah, chief creative officer, Code & Theory.

That “house-building” work involved in that might involve legal consulting, deciding which LLMs are suitable for a brand’s needs, as well as the boring but necessary hours required to collate a brand’s identity and past output into a guide or brief digestible by a generative AI tool — as well as the trial and error, red herrings and dead ends involved in testing an automated system that handles something as commercially sensitive as a brand identity.

It all takes time — unfortunate news when marketers also judge the success of such investments on the time they’re able to shave off ordinary processes (81% of marketers use time saved as the main KPI, according to the Gartner study).

“The most costly thing is just the newness of the process,” said Dave Rolfe, global head of production at WPP’s production business Hogarth.

One unexpected cost? Finding the staff to design, implement and then use such a system in an era when tech giants and advertisers alike are competing for AI skills. ”AI talent is hard to come by,” said James Thoams, global chief technology officer at Dentsu Creative.

While everyday enterprise access to generative AI tools like Perplexity often comes in the form of a subscription, some AI companies have begun capping usage — instead selling credits back to companies that want more from their tech. OpenAI, for example, doesn’t publicly disclose pricing but offers credits as a kind of pay-as-you-go model for businesses using ChatGPT’s advanced features; it follows that intensive testing using a top of the line generative AI could prove pricey.

At Optise, a company that provides B2B advertisers with a generative AI tool designed to improve the organic search performance of their websites and digital presence, CEO and co-founder Ómar Thor Ómarsson notes that usage credits can mount up. “If they’re using it to create content on the fly, a lot, that’s going to cost the company money each time,” he cautioned.

Individually, those tokens a few dollars each and cover multiple uses, meaning each single prompt might cost just a fraction of a cent. But with some early campaigns involving tens of thousands of individual prompts — Coca Cola’s Christmas ad, for example, used 70,000 — those cents might turn into dollars.

“Lots of undisciplined testing with big prompts and premium models can quietly add up,” added Ómarsson.

With copyright battles between AI companies and authors ongoing, the legal implications of using one LLM over another remain in the back of some clients’ minds. It’s a worry that larger agencies have moved to take head-on, by offering indemnification to worried clients. WPP, for example, embedded a system of compliance checks into its WPP Open platform when it launched early last year. Brands building in-house won’t have the same air cover.

“It is a concern,” said Rolfe. “You’ve got to do it into a contained system that is connected by our compliance standards,” he added.

Farther down the line, it’s the humdrum human processes that add speed bumps. Approval processes within agencies, or on the client-side, become time sinks relative to the mere hours now required to actually make a film or still image.

“The real cost isn’t generating assets, it’s generating your assets,” said Elimeliah. He added: “A single prompt can give you 200 like, pretty good options. But someone has to sift through and judge and refine. That decision work used to be invisible; now it is the job.”

As a consequence, review meetings and approvals processes built for businesses that had to take a quarter to shoot and edit a brand film are costing advertisers time and, ultimately, money. “Client approvals are the new bottleneck. AI collapses the time it takes to make content but it does nothing to collapse the time it takes to approve the content,” said Elimeliah.

“The gap is becoming the most expensive part of the pipeline; AI moves in minutes and clients move in weeks and months,” he added.

Some organizations have overhauled the way they design their briefing processes in response. Other brands are using generative AI tools and specialized templates to refine briefs at an early stage. 

“Most of my clients are now using gen AI specifically to develop higher quality, strategic creative groups to pass to their agencies,” noted Gartner analyst Nicole Greene.

At Hogarth, Rolfe described how his production philosophy had been turned on its head. He said that, earlier this year, the company was able to reduce the time taken to produce promotional campaign materials for a telco client from seven to two weeks.

While the time saving comes from generative AI usage, he said realizing economies of scale depended on moving from a mentality that prioritized capturing as much material as possible in a short space of time, to a “component mindset”. Post-production, in other words, has taken over the entire production process.

Others look to automated systems that reduce creative quality control to a series of pre-flight checklists: Does a film use the right aspect ratio? Is the lighting consistent? Is the logo shown swiftly enough? How does it compare to the best performing 15-second film from your last campaign?

Few marketers are comfortable with the idea of hands-off automation in these matters, however. 

“AI content can be awesome and it can be awful. We’re all learning to trust… what’s coming out of the LLMs,” said Ómarsson.