Leading enterprises aren’t scrapping their proven systems for AI — they’re creating powerful hybrids that outperform either approach alone
Leading enterprises aren’t scrapping their proven systems for AI—they’re creating powerful hybrids that outperform either approach alone.
United Health’s recent AI update reveals an intelligent approach to integrating LLMs into their claims processing. While they now have 1,000 AI applications in production, what’s most interesting is how they’ve carefully positioned AI within their claims adjudication process.
The company explicitly does not use LLMs to evaluate claims. Instead, they rely on complex rules-based software for auto-adjudication of 90% of claims. They leverage AI specifically for the remaining 10% that can’t be auto-adjudicated, where AI assists human adjusters by searching across systems to track down missing information and documentation.
This combination leverages the strengths of both approaches. Explicit rules ensure consistent and transparent outcomes — a claim either meets specific, defined criteria or it doesn’t. The LLMs excel at natural language processing and information retrieval across disparate sources.
Aitomatic, featured in a recent Meta case study, demonstrates similar principles in their work with an IC design company. They developed a Domain-Expert Agent (DXA) that combines fine-tuned LLMs with codified expert processes and methods. The DXA captures decades of specialized field engineering knowledge while maintaining rigorous decision-making frameworks.
Aitomatic IC Results
The results are compelling - their IC design implementation achieved 75% first-try success rates (compared to 15-20% from generic AI tools) and 3x faster issue resolution. This success stems from carefully integrating AI capabilities with existing expert knowledge and processes rather than attempting to replace them entirely.
These examples highlight how companies are realizing that while LLMs can be tremendously helpful, they aren’t ideal for every task, especially deterministic processes. The future lies in preserving what works well and thoughtfully augmenting it with AI capabilities. As United Health’s chief digital officer notes, the goal is to be “super pragmatic and super responsible” in applying AI to enhance rather than replace proven systems.