Eric Rosenfeld on AI Cost Control, Token Use, and Measuring Real Business Value

In this conversation, Eric Rosenfeld, Chief Information Officer at ServRx, discusses how his team uses AI to improve pharmacy billing, claims management, and business process workflows in the workers’ compensation and personal injury space. He explains how ServRx moved from early document extraction challenges to more reliable AI-assisted systems that combine LLMs, Textract, orchestration layers, database checks, and human review. The discussion covers the limits of generative AI, the need for governance, the importance of keeping humans in the loop, how to measure AI value, and why companies should focus on practical use cases rather than replacing people. Eric also shares his view on the future of AI, including model selection, cost control, workload routing, and the need to compare outputs across multiple models. Takeaways ServRx helps pharmacies get reimbursed faster and more fairly in workers’ compensation and personal injury cases. The company’s first AI use cases focused on interpreting and extracting data from complex explanation of benefits documents. Early LLM-only approaches were not accurate enough for financial workflows, which led the team to combine document extraction tools with LLM interpretation. Human review remains critical when AI handles claims, payments, legal documents, or other high-risk workflows. AI should support people by reducing manual work and speeding up decisions, not replace human judgment. Orchestration layers help control AI systems by adding checks, database validation, and workflow routing. Poor inputs can limit what AI can do, especially when documents are blurry, incomplete, sideways, or missing key information. AI cost control matters because token use can grow fast in complex or agentic workflows. ServRx measures AI value through new revenue, faster decisions, improved payer follow-up, and work that could not be done before. The future of enterprise AI may involve sending workloads to multiple models, comparing results, and choosing the right model for each task. Today's guest, Eric Rosenfeld, can be found at: Linkedin:   / ericrosenfeld   Your host, Hunter Jensen, can be found online at: Website: https://www.barefootsolutions.com/ LinkedIn:  / hunterjensen   Keywords AI, generative AI, LLMs, pharmacy billing, workers compensation, personal injury, claims management, explanation of benefits, EOB, document extraction, Textract, orchestration, human in the loop, prompt engineering, hallucinations, governance, database verification, business process outsourcing, legal workflows, payer communication, revenue recovery, token costs, model selection, workload routing, cost control, automation, healthcare technology, claims database, decision making, AI strategy