The Reality of AI Investment: Lessons from Tatari's Approach

Few will disagree that "AI" has become the buzzword du jour. Across Alphabet, Microsoft, and Meta, AI was mentioned no less than 200 times in the most recent earnings reports. AI is a term that has caught the imagination of industries far and wide, symbolizing innovation and futuristic vision. However, as the dust settles, it becomes clear that not all endeavors into AI are created equal. Sadly, for many companies, AI is used to sugarcoat questionable business results or earnings reports.

The advertising technology sector, in particular, has been quick to proclaim its AI capabilities. Publicis announced it has spent $11 billion on AI. That’s about the same amount of money that OpenAI has raised to date, yet Publicis has little to nothing to show for it. WPP's Mark Read, in turn, highlighted a $318 million AI investment (annually!) and a strategic partnership to procure NVIDIA chips. Forgive us for stating the obvious, but what will a traditional agency lacking tech chops possibly do with “AI chips” (and how will it drive value for WPP)?

True AI integration into the business fabric demands a dual foundation: a wealth of structured data and a cost-efficient technological backbone. Absent these two pillars, AI's presence is relegated to fleeting mentions in corporate earnings, devoid of substantive business impact. Let us delve into these two prerequisites.

At the heart of efficacious AI deployment lies the rich bedrock of structured data. For Tatari, this means leveraging over seven years of data encompassing millions of media transactions and billions in TV  dollars (and the outcomes related to it for hundreds of brands). This data doesn’t live in ChatGPT, the AI tool so often touted by agencies. This proprietary data fuels our media tools (amongst other parts of our business), enabling precision, efficiency, and speed that dramatically outpaces traditional human-led media buying processes.

Equally critical is the technological scaffolding that underpins AI functionality; Tatari’s foundation is the Databricks Data Intelligence Platform. It is a platform that captures the company’s structured, semi-structured, and unstructured data to power (in a cost-efficient manner) ETL, BI, and ML/Gen AI use cases. This technical backbone enables us to train LLM models on proprietary (unstructured) data, thus returning more tailored and relevant outputs. We furthermore take advantage of many other end-to-end Gen AI Ops capabilities such as native vector databases, AI governance/lineage with Unity Catalog, and model monitoring. In the future, we plan on leveraging MLFlow LLM Evaluate and native LLM/GPU model serving as well. The entirety of Tatari’s data is integrated into a Graph Neural Network that utilizes LLM models for node feature creation – this in itself will power further GenAI applications. At the periphery of it, we have built an MLOps infrastructure that fully integrates with the Data Intelligence Platform, allowing for rapid model iteration and improvement without sacrificing site and platform reliability.

AI at Tatari is more than just two letters. It’s even more than a technology. We see it as a business model. Leveraging our unique data assets and robust infrastructure, we generate (and execute) complex (and above all, successful!) media plans in seconds. This allows us to not only scale the existing business but also enter new markets. For example, Tatari’s agency business has grown from nothing to 10% of revenues this year, and Growth clients (i.e. smaller advertisers new to TV) sign up on Tatari’s platform on a daily (!) basis. Without the leverage of machine learning, managing such a deluge of new business would have been unfeasible. Meanwhile, our average revenue per employee was up 35% in 2023 (and in case you wonder: no layoffs).

Our belief is bold but founded: AI gives Tatari the potential to expand the total US TV advertising market beyond its current $90 billion annual. It is a vision not rooted in the superficial allure of AI as a buzzword; it is rooted in data, technology, and expertise – with the use of AI as part of another important acronym: our company DNA.