Publicis Groupe debuts new CoreAI platform and €300 million AI investment

Publicis Groupe plans to invest another €300 million into AI efforts over the next three years while also putting a new AI platform front and center.

As part of a major AI overhaul, Publicis has debuted CoreAI, which will be a new “intelligent system” for the Parisian giant and provide new AI capabilities across the entire company. Of the planned €100 million spend for 2024, 50% will be directed toward people — including training and recruitment — while the other half will be invested in technology including licenses, software and cloud infrastructure. 

The AI updates, announced today, have already boded well. Publicis Groupe’s stock price closed today at a record high of €92 per share, up from €88.76 a day earlier. The company, which will report its annual results next month, also noted overall organic growth in 2023 was 6.3%, up from previous guidance range of 5.5% and 6% shared in October. (The Parisian giant’s stock price has increased 55% since the beginning of 2023.)

“We’ve definitely been in the AI game for a long time,” Publicis Groupe chief strategy officer Carla Serrano told Digiday on Thursday during a briefing about the AI investments. “And we were mocked by it six or seven years ago when we launched Marcel. So it’s really fascinating to us that everyone’s getting on to their AI game.”

At the core of the AI efforts is a new platform named CoreAI, which is comprised of multiple layers. Using structured and unstructured data, CoreAI incorporates foundation AI models from across Publicis Groupe — including Publicis Sapient’s Bodhi machine learning platform — along with numerous AI agents and an API layer. CoreAI also has a dozen third-party AI services partners including OpenAI, Adobe, Amazon, Microsoft. Other third-party providers include Hugging Face, RunwayML, Midjourney, Pika.AI and Bria. 

Publicis execs see five key areas for CoreAI: Insights and strategy, media planning, creative and production, software development and operations. Publicis said it built the platform using “trillions of data points” from content, media and performance data along with a database of 2.3 billion consumer profiles. It also used nearly a petabyte of assets from Marcel, the AI platform Publicis originally debuted in 2018. CoreAI also leans heavily on Publicis subsidiaries like Publicis Sapient and data giant Epsilon.

“The training of the model is certainty important, but the integrity of the intelligence that is rooted in the differentiation of the data assets is the most important piece here,” said Sam Levine Archer, chief solutions architect for Publicis North America, who noted the company sees 650 billion midstream impressions daily across digital, commerce and social platforms.

Depending on the various applications, CoreAI will use different large and small LLMs, noted Sapient CEO Nigel Vaz. The architecture was built with a model library. Vaz also noted there are various efforts in place to sandbox data based on various privacy protocols and also to make sure confidential data from different clients isn’t used to train models in unintended ways.

“If you send me a link and you happen to be on one client, even if you wanted to share something, if I don’t have permissions access to those files — they are inaccessible,” Vaz said. “That’s even more so for enterprise or consumer-level data, which is specific to a client…And then the minute the use is completed, the data isn’t actually held or stored in any way.”

Engineering work on CoreAI began in the second half of 2023, and Publicis plans to roll out some capabilities in the first half of this year. It’ll also present in May at the annual VivaTech conference in Paris. Publicis declined to disclose how many clients have tested CoreAI so far during its Alpha phase, but execs offered a few client examples during the investor presentation including an unnamed EV car brand and a “heritage cookie company.” 

The kinds of data owned by Publicis Groupe will also be key when it comes to training its AI models on historical and real-time data, as well as any future data to improve CoreAI over time. Another wild card will be how any future AI laws and privacy laws in Europe or elsewhere might impact the Parisian giant’s abilities with CoreAI. To navigate this, consumer data is handled on a geographic basis to make sure various identity graphs comply with national and international laws in each region.

“We’ve talked about some of the new jobs that are coming to the fore,” Serrano said. “Prompt engineers didn’t exist a year ago. I think an ethical task force — that’s looking at both the training that’s happening and some of limitations and pivots that we need to make that we see with every project before we go public — will become a bigger role in our organization moving forward.”

Longtime analyst Brian Wieser said noted that Publicis’s longtime ownership of Sapient gives it more credibility than other agency holding companies. However, that doesn’t necessity give it a silver bullet. 

“I don’t think Publicis’ historical advantages via ownership of Sapient and an earlier focus on AI necessarily positions them to beat their competitors,” he wrote in his newsletter today about the updates. “Agencies can compete on many dimensions, as long as they manage their operations efficiently to drive down costs and look for new products that clients want for which agencies can drive growth.”

https://digiday.com/?p=532885

More in Media

Media Briefing: How Dow Jones is developing an AI model to help its planning team respond to advertisers’ RFPs

This week’s Media Briefing looks at how Dow Jones is incorporating generative AI tools into its ad ops workflows.

Retail media strategies

Snap survey: How brands’ retail media strategies are maturing

That was the overarching theme at Digiday Media’s Retail Media Strategies event held in New York City last month.

Google ad tech antitrust trial

What happens in the Google ad tech antitrust trial now that testimony is done?

While some observers felt like there were no “bombshells” during the testimony, there was still a sense of vindication that some evidence has come to light.