Situation analysis: what problem were they trying to solve?

The Omni operating system uses a massive dataset: it's powered by more than 14 petabytes of data about different marketing audiences.

However, according to Paolo Yuvienco, executive vice president and chief technology officer of Omnicom, with huge amounts of data, it could be difficult for the average employee to make proper use of it.

"You'd be able to see a bunch of different demographic information about [an] audience," Yuvienco told Business Insider. For instance, Omni users could access information about websites visited by a specific demographic group. "Frankly, you needed to be somewhat of a subject matter expert to interpret that and then derive the right insight from that," he said.

In order to optimize the use of data-driven decision making across the company and to drive innovation, Omnicom set out to incorporate artificial intelligence capabilities into the Omni operating system.

Key staff and partners

Omnicom partnered with Microsoft in late January 2023. It was the second company, following Disney, to gain access to the latest OpenAI GPT models via Azure, Microsoft's cloud computing platform.

AI in action

Omnicom developed Omni Assist, a generative AI virtual assistant that sits within the Omni operating system. Yuvienco said that working with Microsoft and gaining access to the latest GPT models early on was crucial, as it allowed Omnicom to "build out some prototypes very quickly" and to "fail fast."

What makes Omni Assist different from, for instance, OpenAI's free-to-use AI chatbot, is that it draws on Omnicom's own data.

"We incorporated the large language models and pointed it to this vast ecosystem of data," Yuvienco said. He added that this created the ability to surface insights through "a very natural human-like interface."

Yuvienco gave an example of what the generative AI has allowed the company to do. "You can say, okay, for this campaign against this audience, identify five influencers that would be interesting to use for this campaign, and give me a preliminary budget split, if I had $100,000, to spend against those influencers."

Yuvienco added that while this kind of insight-surfacing already existed in Omni, the implementation of generative AI means that users can extrapolate and analyze data much more quickly. "What it does is that it gives our people a much further down the line starting point on the journey that they need to go through to drive those outcomes," he added.

Did it work, and how did they know?

Yuvienco said that the project went through extensive prototyping, which took eight months. After going through a client and user alpha stage earlier this year, Omni Assist is now in the "closed beta stage." It has been rolled out to a select group of Omnicom agencies and clients.

"We want to make sure that we get it right the first time out, and [we want to] cost control, because it's still quite expensive to utilize the type of processing that's necessary to get to the outcomes we're looking to drive," Yuvienco said. "Once it reaches a certain critical mass, it will take off and become the norm," he added.

One of the metrics that Omnicom is using to measure the success of the generative AI is time taken to complete specific tasks. "We are not at the point where we basically eliminate the time needed to perform a specific task, because we still need a significant amount of human intervention, but what we are doing is cutting that time down," Yuvienco said.

What's next?

In November, Omnicom announced a partnership with Getty Images, to incorporate a generative AI image tool into Omni Assist, to produce "commercially safe and legally indemnified" images. The company also uses Google's Vertex AI platform and image generation model, Imagen.

Yuvienco said: "I think we're at a point now where we can really start tweaking and finding the perfect balance between humans and machines that's going to be transformational for not just our industry, but for every industry."