Oregon Eligibility Partnership’s Eligibot: agency chatbot shows early efficiency gains, internal-only access
Loading...
Summary
Oregon Eligibility Partnership and OIS described Eligibot, an internal large-language chatbot launched 06/03/2025 to support eligibility workers; presenters reported roughly 265,000 staff queries in the first 150 days, estimated savings over $1 million, and ongoing tuning that raised answer accuracy from about 75% to 93.5%.
Jeff Howe, contractor supporting Office of Information Services engineering for ODHS and OHA, and Kelly Wilfong, content manager for the Oregon Eligibility Partnership, briefed the committee on Eligibot, a curated policy chatbot designed to help eligibility staff find vetted internal resources.
Wilfong said Eligibot is a large language model that staff access through an internal web link; it is not connected to eligibility case systems and is limited to internal documents OEP controls. "Eligibot is available 24 hours a day, 7 days a week, supporting workers during the eligibility determination process," she said, noting the tool went live on 06/03/2025.
Howe and Wilfong described the knowledge curation required: Eligibot was populated with roughly 550 unique internal resource documents (including about 330 eligibility guides and a 1,500-word policy manual referred to as 'open'). ODHS uses a formal review process before publishing documents to the bot and maintains a content manager role to update and retire materials so the bot's knowledge base remains accurate.
On performance, the presenters reported that in the first 150 days the bot answered about 265,000 staff questions, estimated time savings of more than 35,000 hours and fiscal savings over $1,000,000, and that the initial computing costs were roughly $8,000 in that period. "LGBOT has answered a little over 265,000 questions with an estimated savings of over 1,000,000 dollars," Howe said. The team said the bot answered about 75% of questions in June and has since improved to about 93.5% after tuning and staff training.
Presenters emphasized the tool is intended to support workers, not to replace staff judgment or determine eligibility. They described a feedback loop — a thumbs-up/thumbs-down option and a feedback form — to collect user reports and a question set for continuous evaluation and drift detection. Committee members welcomed the innovation but asked how the tool preserves discretion for vulnerable clients; presenters reiterated that Eligibot supports staff during the process and that workers retain decision authority.
