ICF

Federal agencies have largely moved beyond asking whether AI works. The challenge now is determining how to deploy the right AI capabilities in the right places to accelerate mission outcomes while navigating the realities of procurement, governance, workforce readiness, and operational complexity.

New research reveals that while agencies are confident in AI’s potential, many still face significant delays moving from pilot to production and struggle to measure success beyond technical implementation. The primary bottleneck is rarely the AI model itself; rather, it is fragmented data infrastructures, misaligned organizational workflows, and a lack of rigorous, enterprise-grade governance. For instance, pilots often use small, immaculately cleaned, and curated datasets. In contrast, live production environments contain messy, fragmented, and siloed data across legacy systems. When models hit this noisy real-world data, accuracy plummets.

Learning Objectives:

• Identify the differences between a pilot program environment and the real-world environment the pilot program is supposed to scale to
• Review the workflow used by the pilot and compare to the workflow of users not participating in the pilot to find where there are snags and obstacles
• Evaluate the requirements for investing in machine learning operations, security, compliance testing, and scaling compute costs – necessary for budgeting, planning, and timeline management