Making end-to-end advertising conversational with an agentic AI copilot powered by Claude

Metadata had built a powerful B2B advertising platform capable of managing the complete campaign lifecycle across multiple major advertising channels. One of its key differentiators was the experimentation model. The platform combines different audiences, channels, offers, creative assets, copy variations, and ad types into dozens—or even hundreds—of individual campaign experiments, compares their performance, and redistributes budgets toward the strongest results.

This breadth created a substantial learning curve. Customers needed to understand not only digital advertising strategy but also the extensive platform workflows, terminology, navigation, configuration options, and experimentation model. Many customers therefore relied on Metadata’s internal marketing strategists to operate the platform on their behalf. This limited self-service adoption and made the complexity of the product a growth constraint.

Metadata needed a way to preserve the full power of the platform while radically simplifying how customers interacted with it.

To solve this, Proxet and Metadata created an AI copilot directly within its platform, introducing a conversational layer through which users could explore data, make decisions, and operate sophisticated advertising workflows using natural language.

Initially the copilot functioned as an analytical assistant. Proxet and Metadata engineers built and iterated on the copilot’s underlying agents using Claude Code, which accelerated development of the individual platform action agents. Existing agents and platform actions did not operate together as one reliable process. With Claude, the copilot became an orchestrator for these specialized agents. Proxet helped analyze the gaps between these components, test their interactions, resolve missing steps, and connect separate actions into cohesive workflows. 

Depending on the objective, it could now call the appropriate audience, creative, campaign, analytics, or optimization capabilities and connect them into an end-to-end workflow. Sample use cases include: conversational campaign creation, sales acceleration, experimentation with creatives at scale. Human in the loop was built in, so the customer was always in control–combining the speed of agentic automation with human oversight of brand, targeting, campaign structure, and budget decisions.

• 6 hours recovered per Metadata marketing strategist per day.
• Enabled more customers to use Metadata in a self-service or hybrid model without losing access to its advanced capabilities.
• Metadata reports at least 400+ hours of work saved per client, now automated.
• Gave customers control over large campaign portfolios while preserving human review before launch and other consequential actions.
• Repositioned the user experience around marketing objectives rather than software navigation: users could describe what they wanted to accomplish, and the copilot translated that intent into execution.
• Strengthened Metadata’s differentiation as a platform for creating, launching, analyzing, and optimizing advertising across multiple channels through a single agentic interface.