The Causal Context Graph (CCG): A memory architecture for marketing AI

Modern marketing organizations have become proficient at collecting data. Customer data platforms retain profiles, campaign management systems preserve execution history, warehouses accumulate events, experimentation platforms record test results, and dashboards summarize business performance. 

Yet despite this abundance of information, organizations repeatedly lose something far more valuable than the data itself: knowledge generated by marketing interventions. In other words, the results of which marketing action (message content, type, etc.) actually changed the customer’s behavior. And despite the abundance of AI solutions available today, this data gap cannot be overcome by out-of-the-box agentic AI. (If you want to dig in more, read the great Yann LeCun for why next-token prediction alone is insufficient to learn an accurate causal model of the world.) 

The solution to closing this data gap is something we call the Causal Context Graph (CCG).

What is a Causal Context Graph?

A Causal Context Graph stores and links business context, treatments, experiments, and their causal outcomes from marketing actions. And because the CCG lives in an organization’s enterprise data cloud, both marketers and AI can access this graph to use as context for campaign optimization.

Importantly, the CCG is not a customer graph. It does not model relationships among individuals, nor is it intended to replace customer data platforms or identity graphs. Instead, it models relationships among marketing concepts: organizations, audiences, journeys, campaigns, decision points, treatments, metrics, experiments, and aggregate outcomes. The graph captures how these concepts relate to one another through measured evidence rather than customer identity.

Causal Context Graph vs. agentic context graph

An agentic context graph functions the same as a CCG. It stores causal data and gives an ever-evolving roadmap for AI to track customer journeys and interventions.

We see agentic context graphs as the broader category of context graphs, an asset that provides context to agentic AI systems. The Causal Context Graph is the proprietary graph that’s unique to GrowthLoop. Every journey or audience executed in GrowthLoop grows the graph, and every experiment or decision run on GrowthLoop gives it its causal “brain.” 

The value of a Causal Context Graph

What does the CCG bring to the table? Consider this scenario: 

A retailer tests two email treatments on a population of customers and observes that one treatment increases the 30-day purchase rate by 0.8%. The result is discussed, documented, and perhaps implemented. 

Six months later, however, the organization has largely forgotten why that result mattered. Which customers were eligible? What content was in the treatment? What was the control condition? Eventually, the experiment survives only as institutional folklore: “that campaign worked.”

From the perspective of GrowthLoop’s CCG, the central purpose of experimentation is not merely to scale or not scale the current campaign at hand. It is to create reusable evidence on how marketing interventions influence customer behavior, and what makes certain individuals respond to specific treatments. 

The Causal Context Graph addresses this problem. Rather than treating experiments as isolated reports or dashboard snapshots, it represents them as connected pieces of organizational knowledge.  

Going back to our initial scenario of the retailer’s email treatment testing, a CCG would not only record which treatment increased the 30-day purchase rate, it would also capture the business context, the experiment design, and audience information, the content of the treatment all contributing to that experiment. Ultimately, this means the retailer has long-term knowledge of why the experiment worked, what it meant for the business, and how to replicate similar results.

Why the CCG is critical for AI decisioning and experimentation

Organizational knowledge without this causal “brain” means marketing teams only have descriptive or correlational knowledge. And the difference between correlational and causal knowledge becomes particularly important when organizations begin deploying AI.

LLM retrieval alone does not distinguish between correlation and causation. If an AI assistant searches historical campaigns and finds that customers receiving a particular promotion generated high revenue, it has learned a correlation. Without information about randomized allocation, control conditions, or incremental lift, the model cannot determine whether the promotion actually caused the observed outcome. 

The CCG addresses this limitation by making causal evidence retrievable by the AI, along with its experimental context. 

How the Causal Context Graph works in GrowthLoop

The CCG has powerful applications throughout the entire GrowthLoop product:

Faster time-to-value in Composable AI Decisioning 

In AI decisioning, organizations face what is commonly known as the “cold-start problem.” Decisioning systems must initially select among possible marketing actions before accumulating sufficient data to determine which interventions perform best. With the cold start problem, that selection is random in the beginning, based on marketers’ intuitions, and driven by past correlational data (or status quo data). That’s why “slow time-to-value” is a common shortcoming of traditional AI decisioning systems. 

But the CCG’s repository of prior causal evidence offers a fundamentally different initialization strategy for AI decisioning. Rather than beginning with arbitrary assumptions, GrowthLoop’s Composable AI Decisioning can retrieve relevant experimental evidence from comparable audiences, journeys, and business objectives within the same organization. This knowledge accelerates learning for the AI decisioning, and will continually update as new evidence accumulates with campaigns. 

Creating content variants from causal knowledge 

Generative AI has transformed how some teams produce marketing content, but it has not solved a more fundamental problem: which content should be generated in the first place? 

Left on its own, an LLM produces variants by extrapolating from its pre-trained knowledge or from uploaded brand guidelines. While these variants may be fluent and on-brand, they are not grounded in evidence about what has historically caused business outcomes. Content generation remains susceptible to the same correlational biases that affect other AI systems — it optimizes for plausibility rather than causal effectiveness. 

The Causal Context Graph provides a different starting point. Rather than generating creative variants from arbitrary prompts or generic examples, GrowthLoop’s Content Studio retrieves prior causal evidence about comparable campaigns, audiences, messaging strategies, and measured treatment effects within the organization. GrowthLoop then lets users create variants of the provided seed content using this organizational causal knowledge.

Outcomes-based audience building 

Traditional audience building is fundamentally descriptive. Marketers specify rules such as “customers from California” or “customers who purchased within the last 30 days,” relying on deterministic customer characteristics to define who should be targeted. 

The underlying assumption is that customers who resemble historically valuable customers are the best candidates for intervention. However,  that similarity doesn’t imply that a marketing intervention will generate the same kind of engagement, revenue, or incremental business value. 

GrowthLoop’s AI Studio, informed by the Causal Context Graph, reframes audience creation around outcomes rather than attributes. Instead of asking, “Build an audience of customers from California” marketers ask, “Build an audience with the greatest potential of reducing churn AI agents then retrieve evidence from prior experiments involving similar audiences, journeys, metrics, and treatments, and recommend outcome-oriented audiences or actions. These recommendations are grounded in measured incremental effects rather than historical correlations alone. 

With the CCG and AI Studio, audience building shifts from describing customers’ current state to identifying customer populations where marketing interventions are expected to produce the greatest causal impact.

The Causal Context Graph: A foundation for marketing knowledge

The Causal Context Graph is not simply another graph technology. It is a proposal for how marketing organizations should preserve marketing knowledge.  Rather than asking AI to infer causal relationships from historical artifacts, the CCG explicitly records those relationships as reusable organizational memory. The result is not merely better retrieval, but a foundation upon which intelligent and effective marketing systems can be built.