Pixie: A Context-Aware Multi-Agent Multi-Modal Large Language Models (LLM) Architecture for Digital Marketing

We present an innovative framework called Pixie, designed to enhance digital marketing operations through a context-aware, multi-agent large language model (LLM) architecture. The framework is versatile, supporting both real- time and offline marketing activities, and leverages multimodal LLM capabilities to increase reliability and performance. Pixie offers a generalized multiagent architecture capable of performing a broad spectrum of digital marketing tasks. This is achieved through cooperation between a primary orchestration agent and task-specific specialized agents. These agents combine different learning techniques, such as zero-shot and few-shot prompt-based learning, with insights derived from marketing data, and proprietary expert marketing strategy guidelines and documentation on our proprietary statistical models. The agents broadly belong to the following categories – Primary Orchestration Agent which coordinates the workflow, understands user queries, and consolidates outputs from all agents; Task-Specific Agents which perform specific tasks such as performance analysis and Reconciliation/Reviewer Agents which evaluate and reconcile outputs from other agents using chain of thought reasoning.

We experimented with different agentic workflows, ranging from single agent to multi-agent systems. The framework was evaluated through a combination of human expert assessments (trained marketers) and measuring LLM output variance. The evaluation criteria focused on accuracy, consistency, and the ability to solve domain-specific challenges. The results demonstrated that Pixie’s multi-agent framework significantly improves decision quality and operational efficiency in digital marketing workflows. We evaluated the efficacy of the Pixie framework on some of the most frequent digital marketing processes and it can potentially reduce the time spent on tasks such as root cause analysis and reporting, campaign research and planning. The framework is flexible, allowing for the integration of additional specialized agents tailored to specific marketing tasks. This expansion will further enhance the system’s capabilities and its ability to adapt to the evolving demands of digital marketing.