Customer Service Automation · Document Processing · IT Operations Automation
Arizona DCS needed to reduce manual casework, document handling, policy navigation, and development overhead so staff could make child-safety decisions faster and spend less time on administrative work.
Reported impact
Arizona DCS improved classification accuracy, sped policy retrieval, accelerated development workflows, and reduced document-processing effort so caseworkers and analysts could focus more on child and family outcomes.
Arizona DCS and IBM Consulting used a phased Microsoft Azure and Azure OpenAI transformation to modernize several operational workflows. In the central-registry workflow, Azure OpenAI classifies and subcategorizes allegation narratives to improve retention-period decisions and reduce manual review. A separate virtual assistant built on Azure generative AI helps caseworkers navigate policy documents, retrieve reliable answers with source links, and understand updates faster. The program also integrated TachyonGPT with Azure DevOps to generate user stories, acceptance criteria, and test cases, while Power Platform AI Builder automates document uploads and form prefilling. IBM Consulting Advantage supported internal workflows and program materials throughout the effort. Together these components created a governed operating model where classification, policy search, delivery work, and document processing move faster without removing human oversight from child-safety decisions.
Case narratives, policy documents, DevOps work items, and uploaded forms flow through Azure-based generative-AI services, where classification models organize allegations, the virtual assistant retrieves policy answers, TachyonGPT prepares delivery artifacts, and AI Builder reads documents and prefills forms before humans review and act.
Data retrieval has become 40% more efficient
The team worked on improving categorization using Microsoft Azure OpenAI tools to generate subcategories for classified allegations
With the power of AI Builder, the team has significantly improved the efficiency of document processing
an AI-powered virtual assistant was created to simplify the process of accessing and understanding policy documents.
the team worked to integrate TachyonGPT ... with an Azure DevOps Board to generate and refine features, user stories, acceptance criteria and test cases.
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
Google Cloud Sample Archived · Feb 11, 2026 Wonderful Wonderful was looking for a technology partner that could combine advanced multilingual models with globally distributed infrastructure, at a cost that scaled with their growth. They also needed strong reliability,… Gemini powers expansion into 30 multilingual markets Cloud Storage · Gemini · Google Kubernetes Engine (GKE) 
Microsoft Sample Dec 3, 2024 SACE Done manually, these processes take about half a day. With Copilot, they take an hour. "It was very traditional," says Gianfranco Chimirri, who joined SACE as Chief People Officer in 2023. "The… Microsoft Azure Microsoft Sample Apr 18, 2024 Petrobras Brazilian oil, natural gas, and energy corporation Petrobras used Azure OpenAI Service to create ChatPetrobras, a text generation tool using natural language processing to streamline workflows, reduce manual tasks, and… In November 2022, Petrobras employees started to use new AI services for office tasks like summarizing and… Azure OpenAI Service 
