With apologies to the legendary David Ogilvy, advertising today is no longer simply “the business of words.” Though one might be forgiven for not realizing it given the genuine excitement and abundance of public discourse produced by the growth of generative AI tools built on Large Language Models (LLMs).
The transformative rise of LLMs is underpinned by advances in neural networks, a class of machine-learning algorithms that function like the human brain. LLMs demonstrate a remarkable ability to learn from text- and image-based data labeled with text, which allows products like ChatGPT to generate text at lightning speed, covering a massive range of topics in a multitude of languages, including programming code.
As the name suggests, these models work well with problems that are expressed in language. It’s no surprise, then, that according to the latest annual McKinsey Global Survey, marketers are mainly using generative AI for word-based tasks, such as crafting first drafts of text documents, personalizing messages and summarizing documents.
But while LLMs are increasingly demonstrating their usefulness for brainstorming messaging, generating images and other creative applications in marketing, they are less well suited to solving fundamental challenges in programmatic advertising – especially performance advertising – which require learning from and responding to numerical data, not words.
Limitations of generative AI
Advertising today encompasses a wide variety of digital formats, channels and devices with a supply chain that typically involves multiple tools. These solutions primarily use structured data from billions of online consumer interactions and transactions. It’s this structured data that enables the segmentation and targeting needed to deliver against a campaign’s goals.
Thus far, as McKinsey also notes, LLMs allow generative AI to successfully interpret unstructured data sources such as copy, pictures, social posts, news articles, etc. But when it comes to structured numerical data, even relatively simple mathematical problems can be a struggle for tools like ChatGPT, as Stanford University found.
Fortunately, LLMs aren’t the only game in town. Today, many applications of AI, from autonomous vehicles to AI-centric DSPs, use a range of machine learning approaches that organize and harness a variety of data signals to achieve predefined objectives.
As opposed to technological approaches that must be explicitly programmed and require extensive and continual human input, machine-learning algorithms learn from example. That makes them especially powerful for identifying nuanced and ever-changing patterns within multidimensional data. That’s precisely what’s required to create relevant ad experiences for billions of consumers.
In programmatic advertising, machine learning algorithms can analyze data to predict consumer interests, identifying patterns in real time. While the marketer adds their unique expertise, such as specifying business goals, establishing campaign priorities, formulating strategies, considering delivery constraints, allocating budget, and defining measurement approaches, AI automatically finds potential customers based on their online behavior and delivers the most appropriate ad, with the optimal bid, to maximize campaign performance. The result is more relevant and effective campaigns.
AI can strengthen campaign performance
As the overall volume of signals declines (due to third-party data deprecation), AI and machine learning will be essential to providing greater accuracy in understanding the available audience signals and in illuminating the best, most relevant advertising opportunities.
In fact, marketers can test this right now by looking at their campaign delivery on Safari. Programmatic vendors that use machine learning can find users where they see no third-party cookie-based signals. The marketers and technologies that are already experimenting with AI in cookieless environments can reap the competitive advantages immediately while future-proofing themselves for the signal loss that’s expected in Chrome in the coming year.
At the same time, new technological solutions that bridge the gap between unstructured and structured data will allow marketers to harness the respective strengths of LLMs and machine learning. This will enable users to fully realize AI’s potential for both brand and performance advertising. The resulting value to clients will be truly beyond words.
“Data-Driven Thinking” is written by members of the media community and contains fresh ideas on the digital revolution in media.
Follow Quantcast and AdExchanger on LinkedIn.
For more articles featuring Konrad Feldman, click here.
