Customer data platform maker GrowthLoop Inc. today introduced a composable artificial intelligence analytics platform designed to help better understand the causal drivers of customer behavior.

The New York-based company said its Composable AI Decisioning platform is intended to operate directly on enterprise data clouds, enabling marketers to analyze and act on customer data without moving it into separate systems. These real-time capabilities enable businesses to direct messages to customers in near real time while they are browsing a website or on the phone.

“Composable means it cooperates really well with data clouds, data lakes, data warehouses, the likes of Google, BigQuery, Databricks, Snowflake and many others,” said former Chief Executive Chris O’Neill, who recently stepped aside while continuing on the company’s board. “As opposed to taking data and having to hydrate it and move it into a tool, this sits on top.”

The launch reflects a broader shift in marketing technology away from correlation-based analytics toward systems that attempt to establish causality, a capability made more practical by artificial intelligence. O’Neill argued that most competing approaches fall short of the goal of understanding what drives customer behavior.

GrowthLoop’s platform applies a combination of AI techniques, including reinforcement learning, to address that gap. O’Neill said its approach is “multi-armed bandits that essentially create different treatments at the individual level, expose consumers to different stimuli, and then test in a controlled manner.”

He said GrowthLoop maintains statistical rigor using “proven statistical significance applications from the machine learning realm and mixing them with the specific artificial intelligence to show what variables would be most conducive to a causal relationship.”

A key differentiator, he said, is the platform’s ability to operate in near real time without relying on a separate customer data cache, which is commonly used by CDPs. “We use technology like [publish and subscribe] or Kafka queues to poll the website itself and then trigger something in real time,” he said.

That approach enables what the company describes as “same-session personalization,” allowing marketers to adjust offers and messaging while a customer is actively engaged.

The platform also introduces continuous measurement capabilities intended to overcome the limitations of traditional experimentation models that rely on static control groups. “We’re snapshotting actual interactions constantly and the buckles right back into the database,” O’Neill said. This “always-on” measurement approach allows marketers to track outcomes and refine strategies over time.

Another component of the system is what GrowthLoop calls an “agentic context graph,” which integrates multiple data sources to inform decision-making.

Operating directly on cloud data platforms introduces potential latency tradeoffs, but O’Neill said those have largely been mitigated. “We’ve eliminated that tradeoff,” he said. “We have the ability to use the data clouds without the round-tripping latency being a problem.”

Audit controls address concerns about AI model behavior as automated decisioning becomes more common. “It literally is a snapshot with IDs,” O’Neill said. “Every interaction gets snapshotted back into the data cloud.”

More broadly, O’Neill positioned the platform as part of a shift in how marketing organizations operate, moving from linear campaign models to iterative learning systems. “We talk about loops, not funnels,” he said. “You go through a loop faster and then lather, rinse and repeat so you can continue to compound and get steadily better. It’s a partnership between agents and humans to measure causally and overcome the bad habits of the last several decades of correlative marketing.”

Photo: Unsplash