How this LLM designed for ad agencies surfaces insights from big campaign datasets

Campaign data is a useful thing, until you have so much that insights are obfuscated. But what if you could speak to the haystack, and simply ask for the needle? This is the idea behind Agency Data LLM, or AD LLM, a new large language model developed specifically for advertising agencies overwhelmed with data on clients’ marketing efforts. The generative AI system uses a text-to-text interface popularized by chatbots to offer a direct pathway to retrieve insights from this information. When someone asks the model to identify a pattern within a dataset, the model generates code to search the dataset and then delivers the pattern. AD LLM, which launches today from AI company Akkio, reflects how generative tools are offering more precise applications for marketing. Last month, Edelman released an LLM that helps brands comprehend their level of trust with consumers. And Apple this week showed how its all-new Siri can perform tasks based on personal context, which brands may soon be able to leverage in their apps. The primary benefit of this specificity is the unraveling of small but otherwise stubborn bottlenecks within the marketing workflow. In order to retrieve insights from large and intricate reticulations of campaign data today, many agencies require workers to file tickets with their analytics teams and wait days for the results, only to find that what came back wasn’t exactly the information for which they were looking, said Jon Reilly, co-founder and CEO of Akkio. “Every second and third question you have opens up the same process—these things take weeks,” he said. Check out the latest Ad Age events and award programs AD LLM effectively works by turning each agency worker into a data scientist, even if they have little to no background in the field. The chatbot interface allows users to ask questions in their own wording and quickly outputs results that are readable by the untrained eye. Previously, these results could only be acquired by workers with a high-level database skillset, who knew how to code in programming languages such as SQL, Reilly said. Users can request that the bot construct audience segments based on different variables, and it can build charts and graphs illustrating performance reports. Since the model is plugged into an agency’s entire data warehouse, as opposed to siloed data sources, users can connect different sets of data to zero in on specific insights. For example, the following video shows how a hypothetical agency worker might quickly determine BMW owners’ favorite streaming platforms. Shifting generative AI for marketers Akkio’s agency-focused model could help marketers at a time when many in the industry are struggling to advance their uses of generative AI. In a survey published this week, Boston Consulting Group identified AI applications that are high-yield—such as personalization, segmentation and customer insight generation—but found that CMOs have largely failed to scale their exploration of these areas in 2024. Instead, many are focusing their AI efforts on “low-hanging fruit” such as content generation for social media posts. Reilly has seen much of the same, he said. And because marketers are applying AI elsewhere, large datasets continue to cause problems. “The level of actual value extraction from data is quite low, because it’s hard and takes skilled resources,” he said. The rise of highly customized LLMs, which can plug directly into a company’s data environment, could begin to show marketers the value they may be overlooking, potentially leading to a shift away from general-purpose models such as GPT-4, Reilly said. While incredibly powerful, “they’re not optimized for the specific needs and challenges of marketers.” Application-specific LLMs benefit by learning a lot more about the task at hand. In the case of AD LLM, the model can be likened to an expert who has been working in the agency space, Reilly said. It can thus provide deeper and more specific insights for agencies, which can, in turn, offer better results for their clients, he said. Models that leverage more precise data sets may also be better suited for the shifting advertising landscape. The deprecation of third-party cookies continues to rankle advertisers, placing a higher priority on data with qualified parameters, such as first-party data. However, an LLM’s ability to seamlessly scan datasets and fill in gaps provided by other data sets obviates much of the need for identifiable information. “At the end of the day, what you really want to know is what makes [your] audience different and what they care about, so that [you] can reach them in a unique way. And you can still do that,” Reilly said.

How this LLM designed for ad agencies surfaces insights from big campaign datasets

Featured Stories