It’s now using the findings to tighten up its creative production — and alter its commercial relationships with creative agencies.

The project is the latest example of a major advertiser using AI to operationalize the creative side of its marketing efforts. While Unilever and Kimberly-Clark have pioneered time and cost-saving initiatives, Kellanova’s executives are using the tech to pursue a solution to the age-old Wanamaker problem.

“AI is not just an efficiency unlock, it’s an effectiveness unlock,” said Nicole Vinson, vp, global digital, media and omni-shopper experience at Kellanova.

The promise of greater effectiveness is a tempting one for CMOs and marketers dealing with long-term declines in marketing and media budgets. According to Gartner, marketing budgets account for 7.7% of average U.S. company revenues in 2025, down from 11% in 2020; Kellanova’s global marketing spend ran to $611 million in 2024, down from $766 million in 2023, according to its full-year results.

What has Kellanova done?

The CPG giant spent 12 months testing the performance of 443 creative assets across 10 brands, including Rice Krispies Treats, Pringles, Cheez-It, Pop-Tarts and Kellogg’s, in the U.S. using Vidmob’s Aperture solution, which relies on machine-learning, historical data and LLMs like Google’s Gemini.

The study established 19 scoring criteria relating to copy, logo placement, tone, narrative, pacing and product placement — common to assets that achieved at least a three-second view-through-rate on Meta ad units — and built two predictive models (one for salty products and one for sweet) that can be used to anticipate the creative effectiveness of new assets, greenlighting them if they pass a predetermined score.

“If you can get somebody to pay attention for at least three seconds, you’ve started to catch their attention,” explained Vinson. The predictive models forecast the VTR of a given asset with 83% accuracy; deploying that system to creative assets has since led to performance against VTR doubling, while Vinson said that profit ROI (also called net sales volume) for scored assets has risen 11%, based on media mix modelling. 

Vinson said it wasn’t yet clear how those results would influence its paid social spending, but following the project’s completion Kellanova is in the process of rolling out its effectiveness frameworks to more brands, more channels and markets beyond the U.S., beginning with TikTok, Pinterest and Reddit. 

“This is a huge opportunity for us to be unlocking profit ROI and then reinvesting that back into our business,” she said.

There’s parallels between the Pringles owner’s initiative and Unilever’s in-house gen AI production suite, which also included pre-flight effectiveness checks. In Unilever’s case, that was provided by Brandtech Group’s Pencil Pro platform, also used by Danone. The solutions offer a speedy replacement for a formerly labor-intensive task and allow future campaigns to be better informed by past performance, “closing the loop between the data and what’s being made,” said Will Hanschell, Pencil’s CEO and co-founder.

CreativeX, meanwhile, built a similar pre-flight creative effectiveness system for Mars, Diageo and Nivea owner Beiersdorf. These developments are a sign of the growing maturity of generative AI applications, noted Dale Um, CreativeX’s COO. “There was no way you could have done this 10 years ago,” he said.

But this isn’t just about AI development. In the past year we’ve also observed rising investment in creative effectiveness measurement firms like System1 and EDO. Behind the software, it’s clear there’s demand among CMOs for data that helps justify their marketing spend to the rest of the C-suite.

“We’re finally seeing [the] ability for AI to help teams develop creative and test it before they’re spending an exponentially large amount of money to put it into flight,” said Gartner analyst Nicole Greene.

What does this mean for agencies?

Kellanova’s system inevitably holds consequences for its agency partners, which include FCB, The Martin Agency, MullenLowe, Tombras, Momentum and several Publicis agencies. Marketers employ agencies for the experience and good taste of their creative directors and copywriters. Now, execs like Vinson can use Vidmob’s predictive models to back up their own opinions about the creative work they’re presented with.

For the time being, Kellanova’s effectiveness approach is only being applied to paid social, an advertising channel defined by a high quantity of assets, each bearing minor variations in service of personalization. Alex Collmer, the founder and CEO of Vidmob, argued that the solution wasn’t intended to restrict agencies’ work, but provide another layer of quality control to a production process which can prove mechanical, repetitive and time-consuming.

“This is really not intended to limit the creative canvas,” he said. “It’s a tactical thing.”

Kellanova is using the predictive solution as an excuse to look again at agency remuneration, however. She said the company was moving toward an output based pricing system with some of its agencies. “We want our agency partners to have skin in the game,” she said, explaining that a portion of each agencies’ bonus fee was tied to creative performance — now based in part upon the effectiveness measurement project undertaken with Vidmob. 

“Now we have a quantitative metric versus a subjective metric,” Vinson explained, but she declined to share the precise financial details of the new arrangements. “We’re not trying to grade anyone’s homework … We’re trying to move ourselves in a direction where we’re not leaving any money on the table.”

She’s not alone in that ambition. 87% of marketers believe agencies are resistant to more transparent fee models, according to a WFA study published at the start of the year.

Industry analysts already believe that more large advertisers might use AI creative techniques as an excuse to review agency pricing agreements. “It’s going to become a major talking point,” Bruce Biegel, senior managing partner at Winterberry Group, told Digiday recently.

“The current agency model is just not sustainable with the future impact of artificial intelligence and particularly agentic AI,” cautioned Greene. “They’re going to need to better align their work to specific business outcomes as a result.”