CMS built a waitlist for its AI chatbot — and that drove momentum — official says

The Centers for Medicare and Medicaid Services had employees join a waitlist to test out its generative artificial intelligence chatbot when it first rolled out the tool — an approach that helped build momentum for the product’s eventual agencywide adoption, according to a CMS official. 

Andres Colón Pérez, the chief technology architect at CMS, said the agency’s embrace of its internal chatbot evolved out of the promises of generative AI tools, but with a recognition of their potential drawbacks. Pérez made the remarks during an ACT-IAC event on Wednesday. 

“And our approach — when we decided, ‘okay, we're actually going to launch this for the organization’ — was that we’re all very excited, but we were going to be very honest about its limitations, right? Primarily because many of our users had never touched a system like ChatGPT,” Pérez said.

Given concerns about employees’ comfort with using an AI-powered chatbot, as well as their potential lack of familiarity “with some of the drawbacks, some of the gaps,” Pérez said CMS initially moved to prioritize voluntary participation in testing the tool. 

“Instead of releasing it to everybody, we made it a ‘join waitlist,’" he said. “And so that essentially allowed those that were very excited to try — like they were already using it in their private lives and we didn't know about it — they were reaching out and saying, you know, ‘I want to join the waitlist.’”

This approach, Pérez said, “started building momentum within the organization” for adoption of the tool, because those employees “would do things faster [and] they would share with their colleagues.”

Pérez said there were still some lapses in understanding the chatbot during this pilot period, as well as some issues with the tool itself that needed to be worked out. But he said the waitlist-based testing process “set up the right foundations for proper training and broader rollout.” 

The agency launched CMS Chat in December 2024. In its 2025 AI use case inventory, the Department of Health and Human Services said the chatbot “generates text-based responses including drafted content for emails and reports, document summaries and analysis, synthesized findings, brainstormed ideas, and answers to queries - all delivered through a conversational interface.”

In version 4 of its AI playbook, the agency said “100% of CMS employees have been granted access to use CMS Chat, with participation from all CMS components and the Office of the Administrator.” CMS also noted that it launched a “micro-training program,” known as AI Ignite, that has helped train more than 4,700 CMS employees in using CMS Chat, as well as using other generative AI tools. 

Pérez said that 80% of users have given the tool high marks for helping them save time.

Many federal agencies have also launched their own internal chatbots to assist their workforces with basic tasks. The State Department rolled out its generative StateChat tool in 2024, although it took some time for the service to be adopted by its employees. 

During an event last September, State’s Chief Information Officer Kelly Fletcher said StateChat was initially used by 3,000 beta testers and that — as of that time — it was being used by between 45,000 to 50,000 employees out of the department’s roughly 80,000 workers. 

“Something I wildly underestimated with AI is the amount of training and education and conversation required to get folks who would benefit greatly from it to use it,” Fletcher said.