Agentic Advertising Explained - Fluency on AI Agents MCP AdCP and Governance at Scale

Introduction to Agentic Advertising and Fluency (0:05)

This session introduces Eric Picard, Senior Vice President of Product at Fluency, discussing agentic advertising. The goal is to understand the good, bad, and ugly aspects of AI agents and how to deploy them safely in the advertising environment. The current landscape is compared to the early days of internet advertising, marked by rapid startup growth and intense board-level interest in AI strategies.

The Dangers of Unconstrained LLMs (3:16)

Giving Large Language Models (LLMs) agency—the ability to act on your behalf—is a powerful but potentially dangerous concept, especially in critical systems like advertising where money is spent. Just as one wouldn't let an LLM write a blog post without editing or manage personal finances without oversight, it's risky to fully unleash an LLM on important ad systems. While early experimentation is crucial, larger organizations require enterprise-class capabilities, auditability, and the ability to roll back changes, which goes beyond what an individual experimenting with an LLM can provide.

The Rise of Agentic Advertising Protocols (6:06)

There's significant momentum and investment in using AI agents for advertising. Two key protocols driving this are MCP (Messaging and Content Protocol), started by Anthropic, and Agent-to-Agent Protocol, started by Google, which are becoming standard for inter-agent communication. Within advertising, two major initiatives built on these protocols are ADCP (AgenticAdvertising.org), an open-source project focused on accounts, campaigns, and organizational workflows, and AMP (IAB Tech Lab), which primarily targets programmatic advertising, exchanges, and transactional levels. These initiatives are seen as collaborative, solving slightly different problems, and are actively being adopted by large and small companies, with campaigns already being transacted through them.

Probabilistic vs. Deterministic Systems in Advertising (12:02)

LLMs are inherently probabilistic; asking the same question five times yields five different answers. This feature is valuable for creation, problem-solving, and intelligent conversation. However, for critical operations within an ad platform, such as spending money or instantiating campaigns, deterministic business rules are essential. You need a system that consistently makes the same decision given the same inputs. While LLMs offer creative power, a deterministic layer is required to ensure predictable, controlled, and scalable ad campaign management, particularly for large budgets and frequent, small changes, where LLM costs and unpredictability become prohibitive.

Fluency's Deterministic Agentic Platform (14:36)

Fluency operates as a deterministic agentic platform and operating system across the ad ecosystem, managing hundreds of thousands of campaigns and approximately $3 billion in spend monthly for eight years. It's designed for large brands, agencies, and multi-local businesses (e.g., car dealerships, real estate, franchises). Fluency connects to various ad platforms, providing automated campaign management, multi-channel reporting, and insights. The platform hosts data for LLM decisioning and plans to become an agent gateway, allowing third-party agents to plug in and use Fluency as their conduit to various ad platforms, operating at scale with built-in governance.

Blueprints: The Guardrails for AI Agents (17:49)

Fluency's core capability is "blueprints," which serve as the deterministic "DNA" or "tracks" for AI agents. While probabilistic AI acts as the powerful engine, blueprints provide the governance and guardrails needed for predictable, reliable, and safe outcomes at scale. Blueprints predefine how campaigns are built, dictating stages, tools, data usage, and conditional logic. For example, a blueprint can instantly launch localized ad campaigns for power generators if a hurricane is approaching and stores have inventory, or pause irrelevant ads. This ensures business logic is followed, reducing unpredictable results and enabling auditability through detailed logs. Blueprints can scale across various businesses within a vertical while allowing for unique targeting and naming conventions.

Customization and AI-Driven Blueprint Creation (26:41)

Blueprints are highly customizable and differ for every business, agency, or holding company, aligning with their unique strategies and best practices (e.g., B2C, B2B, channel-specific, vertical-specific). Fluency is also developing new capabilities where LLMs can automate the creation of these blueprints by ingesting an agency's existing playbooks and naming conventions, significantly streamlining the setup process that was previously hand-coded.

A Three-Step Process for Agentic Implementation (29:03)

Successfully implementing agentic advertising involves a three-step process:

1. Invest in the Right Infrastructure: Utilize open-source frameworks like AgenticAdvertising.org (which offers certification via Atti, the AAO agent) or partner with platforms designed for scale.
2. Run Deterministic Automation: Implement pre-established business rules and automation to handle routine, error-prone workflows. This can range from human oversight in low-volume scenarios to sophisticated platform capabilities like Fluency's blueprints and budget management engines for high-volume operations.
3. Enable Safe Agent-to-Agent Interactions with Governance: Establish a robust system of record that defines rules, safely interacts, and manages campaigns. Ensure LLMs analyze properly processed data rather than attempting to process raw data themselves.

Addressing Client Concerns and Scaling Decisions (33:22)

Addressing potential client damage is managed through platforms built with automatic rollbacks and searchable audit logs for every change. For small teams or direct-to-consumer brands, building a custom system with an LLM for a single platform might be feasible. However, for larger organizations requiring scale, compliance (e.g., SOC 2), and enterprise-class capabilities across multiple platforms, partnering with a platform like Fluency becomes essential. This avoids the need to build an entire infrastructure from scratch for each platform.

Deterministic Orchestration and Human Involvement (36:16)

The "connective deterministic tissue" can be built using various approaches, from orchestration layers like N8N to custom-written software engines like Fluency's. The goal is to codify human wisdom and expert decisions into scripted rules. When experts analyze reports and make decisions, these insights can be translated into automated, deterministic workflows. The concept of "human at the helm" or "human on the loop" emphasizes that AI should augment human capabilities, allowing experts to focus on strategic thinking rather than routine tasks.

Team Roles, Legal Guidance, and Compliance (41:23)

Successful adoption of agentic platforms often requires adjusting workflows and even staffing within an organization. Companies that embrace these tools can significantly increase efficiency and campaign performance by removing distractions and hair-on-fire emergencies through automation and guardrails, allowing teams to focus on strategic improvements. When dealing with legal and compliance, it's crucial to educate them and view them as consultative partners. Legal guidance provides recommendations, but the ultimate business decision takes that advice into account.

Reconciliation, Billing, and Fluency's Future Vision (45:05)

While Fluency doesn't have direct integrations with systems like Media Ocean for reconciliation and billing, it facilitates easy import and export of data in compatible formats for large agency clients. Looking ahead, Fluency's corporate vision is to become an "agent gateway" for scaling LLMs across the entire advertising ecosystem. Fluency focuses on account-level integrations, configuring campaigns, pulling data, and managing operations in a non-real-time capacity, supporting the broader adoption of deterministic agentic advertising.