The Challenge
Scaling AI across a software platform that serves hundreds of local governments only works if the solutions fit into the hands of the people already doing the work—editors, legal reviewers, material specialists, and project managers—without disrupting what they know. A leading GovTech company specializing in software solutions for local governments was facing a period of significant operational transformation. Their workflows across two major business lines—legal codification and publishing and digital experience for municipalities—were heavily manual, specialist-dependent, and difficult to scale.
The problems were spread across the organization: editors manually cleaning HTML during large-scale website migrations, legal reviewers spending hours cross-referencing constitutional standards, material specialists relying on expensive external OCR vendors to process legislative documents, and project managers pulling Salesforce data by hand to write client status reports. Each workflow was a bottleneck on its own. Together, they represented a structural inefficiency across the entire platform.
The Approach
We started with a structured discovery across both business lines, running parallel workstreams to map workflows, identify high-impact opportunities, and validate the technical feasibility of AI intervention before committing to a full roadmap.
From that discovery, we identified 8 initiatives spanning content, legal review, document processing, indexing, and operational reporting. Rather than treating these as isolated projects, we approached them as a connected transformation—sharing infrastructure, patterns, and learnings across workstreams to compound impact over time.
Two principles guided every initiative: keep humans in control of final decisions, and build for editors and specialists, not engineers. Every tool we built was designed to fit into existing workflows rather than replace them.
Two POCs validated the approach early. The Supplement Editor Assistant showed that an AI agent could ingest new legislation, compare it against a city's existing code, and suggest precise changes — dramatically reducing manual editorial work. The Content Migration tool demonstrated that AI could automatically clean and standardize HTML content from legacy government websites, flagging accessibility issues and surfacing improvement suggestions at scale.
With both POCs validated, we moved into full delivery.
The Solution
The Solution
We designed and delivered 8 AI-powered initiatives across the company's core operations:
Content & Websites Built an AI-driven content optimization workflow that automatically cleaned and standardized migrated website content, suggested improvements, and flagged items needing human review—achieving 93% main content extraction accuracy and 90% HTML cleanup accuracy.
Built an AI-driven site exploration tool that crawled entire websites, identified content types across all pages, categorized them, and surfaced accessibility enhancement suggestions for human review.
Legal & Codification (Code & Supps) Built an AI-powered legal review assistant that surfaces contextual legal suggestions, flags obsolete references, and retrieves relevant precedents—with approximately 80% accuracy on suggestions and 40% of suggestions grounded in retrievable legal references.
Built a supplement assistant that automatically analyzes new legislation, identifies impacted ordinance sections, and suggests updates while keeping editors in full control of final decisions.
Created an automatic indexing solution that analyzes thousands of code sections, identifies the most relevant terms, groups related concepts, and generates consistent cross-references—matching or surpassing human-built indexes at the default configuration threshold.
Built a knowledge-based DOCX-to-XML conversion system that automatically identifies Word styles and transforms them into schema-compliant XML—reducing manual conversion time by 80–90% with 98.6% overall accuracy.
Document Processing Built a high-accuracy in-house OCR and document processing pipeline that replaced costly external vendors—processing 4,692 documents with a 99% success rate and achieving 89% automated classification accuracy across 3,560+ legislative documents.
Operations Built an AI-powered SOW review system that analyzes statements of work across multiple teams simultaneously, highlighting sections needing attention and reducing manual review time by 80–90%.
Built an automated status report generation system that drafts customer-facing project updates from Salesforce data on a configurable schedule, keeping project managers in full control of final edits and delivery.
The Outcomes
What started as a discovery engagement in March 2011 grew into one of Arionkoder's longest and most expansive partnerships. At peak activity, more than 10 projects were running simultaneously across the Websites and Code & Supps workstreams. Over time, the collaboration expanded beyond its initial scope to support additional business units, including establishing a dedicated AI execution team to help the organization deliver on a growing backlog of AI initiatives across multiple lines of business.
Every initiative was built with the same philosophy: AI handles the repetitive, rule-heavy, high-volume work — and humans stay in control of the decisions that matter.
