Shirli Zelcer, head of data and analytics at Merkle, and the team began building GenCX, a custom set of capabilities driven by artificial intelligence (AI), earlier this year. The foundation brings together different types of data to build what she calls "large knowledge models" (LKM).
“We started by playing around with some of the AI capabilities being developed across the industry,” Zelcer said. “You can ask it live questions about the data and it will return insights.” In addition to insights, it returns well-thought-out calculations in a matter of minutes to improve ad-campaign performance.
Some of the calculations may focus on a better understanding of the life cycle of a campaign. For example, the questions could be as simple as “how does this campaign perform?” or “if I had a million dollars at the end of the quarter, what media should I invest it in and what channels should I use?"
“We are building large knowledge models,” she said, explaining that these LKMs combine data that was not previously possible. “Many companies talk about building large language models, but we are building large knowledge models, which take into account all the text data, but numeric, too, like video and creative.”
Merkle calls the large knowledge models “foundation models” that account for text and numerical data, structured and unstructured data, videos, images, and product reviews.
Merkle is working with a small handful of clients to build out different use cases, with the goal of growing it into supporting a major part of its clients.
When asked to cite the most exciting part of the project, she pointed to the results of building the predictive and the large language models to support the calculations. The models took about a week to build, but once built, the calculations took minutes and resulted in better performance.
One challenge has been understanding the different modeling. Other challenges include maintaining privacy and compliance and ensuring that everyone at the company as well as clients understand the possibilities.
“I want to ensure people are thinking about human intervention when using AI,” she said. “Their jobs won’t go away, because you need humans. The jobs will just become more efficient. We also need to ensure there’s no bias in the AI data.”
Merkle has a large practice around ensuring that bias does not seep into the data. Ongoing validation checks this continually. Those companies that do not have the checks and balance run the risk of unintended bias. Merkle calls it Ethical by Design.
Part of the future vision is to empower marketers with more insights.
Zelcer said GenCX could become a hybrid service -- part managed and part self-service.