Building Enterprise AI Agents: Beyond the Proof of Concept

Many companies are experimenting with AI agents, but few have tackled the challenge of deploying them reliably at scale. TransUnion, a global credit reporting company, has taken a unique approach that could serve as a model for others in regulated industries like finance and healthcare.

Having built its own agentic AI platform called OneTru, TransUnion demonstrates that while creating simple agents is relatively easy—taking just days with current tools—establishing the necessary governance, security, and scalability requires significant investment and architectural rethinking. The company allocated $145 million to develop OneTru, which combines deterministic “expert system” approaches with generative AI capabilities in a layered architecture.

Layered Approach for Reliability

TransUnion’s design separates core decision-making—which relies on updated expert systems operating under well-defined rules—from tasks where generative AI can add value. For example, when an agent encounters a new situation, an LLM analyzes the problem and suggests a rule update that human experts review before implementation.

This approach creates built-in checks and balances while limiting any single component’s potential impact. As TransUnion’s Chief Technology Officer Venkat Achanta explains, “With the neural reasoning layer—the LLM—we put humans in the loop. When it’s a symbolic reasoning layer, which is logic and machine-learning-driven, we let it be automated.”

The result is a system that combines the flexibility of generative AI with the predictability and auditability required for enterprise applications. TransUnion has already seen $200 million in cost savings from OneTru and is using it to build customer-facing solutions like its AI Analytics Orchestrator Agent powered by Google’s Gemini models.

What lessons can other organizations draw from TransUnion’s experience? Building a security foundation around agentic AI requires focusing on the “seams” where different components connect, rather than treating the entire system as a black box. By designing for reliability from the outset, companies can unlock the transformative potential of AI agents while mitigating risks.

Tags: ai-governance, enterprise-ai, transunion, united-states, infrastructure