Before and After: Collecting and Analyzing Time Expenditure Data To Assess Impact of Human-In-The-Loop Generative Ai On Routine User Support For Economic Data Services

Supported by the Federal Reserve Bank of St. Louis, FRED (https://fred.stlouisfed.org) is an economic data aggregation service that presently compiles more than 800,000 data series from more than 100 public and private sector sources. Users can access data via web browser, download, Excel add-in, or API.

Serving more than 18 million users annually, from around the world, FRED receives thousands of user-support requests per year. These questions come from users representing many different occupations, educational backgrounds, and technological skill levels. The problem: questions come in via email, and are often duplicative in nature. Responding takes time, and the task is not easily automated; the team must remain close to the product and to the users experiencing it.

This presentation will review basic data collection and analysis efforts conducted in the summer of 2024 and summer of 2025 to determine potential time savings offered by an internal-only generative AI solution. A human in the loop ingests well-suited user-support questions in bulk, edits automated drafts, provides response-quality feedback to AI developers, and sends appropriate responses to users at the level of detail to which they are already accustomed.

The analysis demonstrates 44 percent reduction in time spend, with the potential for much more in the future. Personnel capacity previously reserved for user support is now more available for higher-level technical work in service to FRED operations.