OpenAIPublic SectorCustomer Service AutomationDocument ProcessingAI and Machine LearningContact CenterGenerative AI / LLMCustomer ExperienceProductivity ImprovementRisk ReductionScalability2026OpenAIOA#10a37fPublic SectorPennsylvania, United StatesThe Commonwealth wanted to improve the speed, clarity, accessibility, and accuracy of public services while reducing manual effort for state employees handling drafting, research, hiring support, document review,…175 employees across 14 agenciesApr 15, 20266Customer Service Automation|Document ProcessingBoard of Pardons online tracking system · COMPASS · ChatGPT Enterprise02026-04-156
OpenAI Public SectorPennsylvania, United StatesCompany newsroom
Commonwealth of Pennsylvania
Customer Service Automation · Document Processing
The Commonwealth wanted to improve the speed, clarity, accessibility, and accuracy of public services while reducing manual effort for state employees handling drafting, research, hiring support, document review, benefits interactions, and case processing.
Pennsylvania scaled governed AI use from a pilot to statewide operations, saving employee time, improving document intake quality, increasing automated resolution in unemployment support, and using AI-assisted enhancements to improve customer service, reduce SNAP error rates, and speed application processing.
Customer Experience
Productivity Improvement
Risk Reduction
Scalability
Solution
The Commonwealth of Pennsylvania expanded a governed generative AI program built around ChatGPT Enterprise and agency-specific AI-enabled service workflows. Employees across 35 agencies use ChatGPT Enterprise for drafting communications, summarizing materials, conducting research, brainstorming, and helping HR staff analyze state job requirements against comparable federal positions, with human review kept mandatory. In service delivery, the Department of Human Services uses AI to scan client-uploaded documents for blurriness, image quality, and relevance before caseworkers continue processing, and related AI-assisted enhancements feed into COMPASS to improve customer service, lower SNAP errors, and speed applications. The Board of Pardons uses AI to transfer data from handwritten clemency applications into its online tracking system. Labor & Industry rebuilt unemployment chatbots so AI maps broader user phrasing to curated answers and escalates unsupported questions to live representatives.
Data flow
Commonwealth employees send drafting, research, and analysis prompts through ChatGPT Enterprise and then review outputs before use. Public users upload documents to DHS systems, where AI checks image quality and relevance before caseworkers continue processing and COMPASS workflows consume the results. Handwritten clemency applications are converted into the Board of Pardons online tracking system, while unemployment users query AI-enabled chatbots that either return curated answers or route the interaction to live representatives.
Solution architecture
5 components · 1 layer
Serving
ChatGPT EnterpriseProvides governed generative AI assistance for employee drafting, summarization, research, brainstorming, and HR job-analysis tasks.
COMPASSReceives AI-assisted enhancements used by DHS to improve customer service, reduce SNAP error rates, and process applications faster.
Board of Pardons online tracking systemStores information transferred from handwritten clemency applications with AI assistance.
UC benefits chatbotHandles unemployment benefits questions by matching varied user phrasing to preapproved responses and escalating exceptions to live staff.
UC taxes chatbotHandles employer unemployment tax questions using the same curated-response and escalation pattern.
Architecture clues
Governance is enforced through a Generative AI Governing Board, policies, guidelines, and required training.
Human review remains essential for employee use of generative AI.
The source describes AI embedded into existing Commonwealth systems rather than a newly named standalone data platform.
The unemployment chatbots do not generate novel answers and escalate out-of-scope questions to live representatives.
Evidence from the source
"175 employees across 14 agencies reported an average 95 minutes saved per day"
"DHS and OA incorporated this input into a suite of enhancements to COMPASS to improve customer service, reduce Pennsylvania’s SNAP error rate, and process applications faster."
"Since its launch in October 2025, nearly 400,000 documents have been scanned, with approximately 30 percent flagged for issues that could delay processing."
"Since its relaunch, the number of customers whose interactions were handled entirely by the UC benefits chatbot increased by approximately 65 percent."
"The Board of Pardons is using AI to assist with transferring information from handwritten applications for clemency into the Board’s online tracking system."
"The Commonwealth’s year-long pilot of ChatGPT Enterprise demonstrated clear benefits when AI is deployed with strong governance and an employee-centered approach."
"The Department of Human Services is using AI to scan documents uploaded by clients and inform them of potential issues with blurriness, image quality, and relevance."
"The Department of Labor & Industry rebuilt its chatbots ... with AI that can match a wider range of user phrasing to preapproved, curated responses."
"the Shapiro Administration has expanded the use of generative AI tools to over 3,000 state employees across 35 agencies"
Review how this reference was scored and trace each normalized claim back to its documented source.
01
Scoring methodology
How scores are calculated
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Customer reference value100/100
Problem clarity20/20
Measures whether the customer's operational problem is specific enough to compare.
Implementation depth25/25
Rewards named components, data flow, integrations, and architecture detail.
Outcome specificity20/20
Measures whether the reported benefit is concrete rather than a generic claim.
Quantified impact25/25
2 metrics describe measurable impact.
Deployment scale10/10
Rewards evidence of a live or scaled deployment and meaningful integration scope.
Evidence confidence94%
Source authority25/25
The deployment is described by the adopting organization or a government source.
Claim traceability21/25
10 of 12 key fields link to a named source.
Direct evidence20/20
9 direct source quotes retained.
Date provenance10/10
The publication date comes from the source page or publisher.
With your OK, we use Google Analytics cookies to see which pages and searches are useful. No ads, no profiling, and
nothing is loaded until you accept. You can change this any time under “Cookie settings” in the footer.