Fluency Deploys Agentic AI to Manage $3 Billion in Ad Spend
Digital advertising platform Fluency is integrating large language models with its deterministic automation systems to safely scale agentic ad campaigns without risking client budgets.
Fluency, a digital advertising operating system managing approximately $3 billion in annual ad spend across more than 250,000 monthly campaigns, is integrating probabilistic large language models into its deterministic execution platform. Eric Picard, Fluency's Senior Vice President of Product, explained that while generative models excel at creative reasoning, they are inherently unreliable for deterministic tasks like budget management. To solve this, Fluency separates the creative, probabilistic reasoning of AI from the strict, rule-based execution layer that actually deploys the campaigns.
This dual-layer architecture allows advertisers to establish firm guardrails. Picard, who previously scaled Microsoft's ad-tech division from 200 to 1,500 engineers and grew its revenue from $750 million to over $5 billion, is helping align Fluency with emerging industry standards. The company is focusing on the Agentic Advertising Management Protocols (AdCP) managed by AgenticAdvertising.org, which automates account-level campaign management and creative workflows. Fluency also plans to support the IAB TechLab's Agentic Advertising Management Protocols (AAMP) to proxy AI into real-time bidding infrastructures.
For practitioners, this setup solves the daily operational grind. Picard noted that automating these repetitive tasks eliminates the "death by a thousand cuts" of manual campaign management. For example, a car dealership can link its inventory system to Fluency to automatically launch ads for specific vehicle models and pause them the moment they sell. The platform's SOC2 compliance ensures that every automated action is fully logged and auditable. If an AI agent suggests an action that falls outside of pre-defined risk tolerances, the system automatically routes the decision to a human supervisor.
Picard believes this architecture serves as a blueprint for other highly regulated sectors, such as financial services, healthcare, and enterprise procurement. By keeping the decision-making layer probabilistic and the execution layer strictly deterministic, organizations can leverage generative AI without risking financial or compliance errors.
This is our own summary of reporting by Unite.AI
