It’s tempting to think AI models can hack their way to the best solution possible, no matter how basic the input we give.
But feeding clean first-party data to AI models is an absolute prerequisite to ensuring that the output is on-brand and accurate. And there is little value in generating five times more content if you cannot determine what works and what doesn’t.
While the likes of OpenAI and Google have done a great job making their models increasingly capable, the rise of generative AI in day-to-day marketing workflows has put pressure on marketing technology providers and IT organizations to shake up their data strategy accordingly.
New use cases have emerged in the past 18 months that are revolutionizing how a modern marketing cloud should be designed, with more powerful applications that intertwine creative and media for more integrated campaigns.
Here are three areas that marketing technology professionals must handle successfully to get the most out of their AI applications.
1. Combine structured media performance data with unstructured creative data sets
In the past decade, marketing technology leaders have almost exclusively focused on creating strong cloud ecosystems around structured first-, second- and third-party data. Aggregated media campaign performance, customer records, sales data, cookieless graphs have been ingested, stitched together and visualized through end-to-end ELT and BI pipelines for identity resolution, programmatic activation and marketing mix modeling.
Conversely, very little has been done to understand what drives performance from a creative perspective. As a result, creative assets have largely remained stored in standalone DAMs or dumped in data lakes with no real thought given to how to derive value from them.
Now, multimodal AI models can extract insights from creative files. Frameworks such as Google’s ABCD (Attention, Branding, Connection, Direction) help make sense of that multimedia creative data.
Marketers can connect the content inside their Digital Asset Management (DAM) system with the performance data they already pull from media platforms. This enables them to more effectively measure how creative decisions impact their bottom line.
Brands can also get a more complete understanding of the creative lifecycle, which can guide how much media investment should go toward each set of masters and variations. Being able to ensure that local teams are adapting preexisting assets rather than having their agency generate new sets of creatives is paramount, particularly in a world where each marketing dollar counts.
2. Contextualize large-language models with proprietary data
Large-language models have been trained on trillions of words, but they know nothing about what makes your brand unique. They cannot make up your strategic marketing guidelines.
That’s why AI marketing applications have started using pipelines called Retrieval-Augmented Generation (RAG) to pull specific brand information and feed it into generative AI models to avoid hallucinations and keep the content they create always on-brand and accurate.
Marketing technology and IT organizations must now evolve their data governance and data sharing policies to include data sets that largely have fallen outside of their scope.
Marketing teams can use new solutions such as Google Agentspace, built on top of NotebookLM, to sift through massive amounts of wordy documents, recordings and videos and extract the specific information that should be shared with a third-party AI model for a particular use case. Agentspace then creates an agent to automate the data collaboration process without moving the confidential information from the company’s cloud environment.
3. Evolve data products into agents
The increasing demand for ready-to-use data sets has led marketing technology professionals to create frameworks such as data meshes, where nontechnical users can consume data products, governed by data contracts and described by data models.
These frameworks have successfully underpinned the democratization of marketing analytics across the enterprise. But the underlying plumbing of data products is still based on predefined SQL queries, rule-based automation workflows and dashboards with multiple drop-down menus.
With the rise of reasoning models and AI agents, it is now possible to create agentic applications that will autonomously make sense of a data set’s semantic layer (a business description of the data object) or a data object graph (a combined representation of the data and the processes it is linked to).
These agents can create new data products and proactively alert marketing practitioners. For example, agents could be trained to identify early signs of creative fatigue and make strategic adjustments to media budget allocation. They could also suggest high-performing DMAs or high-inventory SKUs on retail media.
While these evolutions are poised to hit every marketing organization as AI agents become more ubiquitous, data leaders will need to rethink how they structure, document and connect both structured and unstructured data. Hyperscalers and enterprise data solutions such as Snowflake and Databricks are already scrambling to provide solutions to ease this process.
New standards for data communication across agents are also emerging with the likes of Model Context Protocol (MCP) that are now supported by both Anthropic and Open AI.
Gen AI has revolutionized how marketing organizations create content at scale. But remember the old adage: garbage in, garbage out. Only those who rethink their data strategy and data collaboration frameworks will take full advantage of the revolution underway.
“Data-Driven Thinking” is written by members of the media community and contains fresh ideas on the digital revolution in media.
Follow The Brandtech Group and AdExchanger on LinkedIn.
For more articles featuring Hugo Loriot, click here.
