Why AI matters at this scale

The North Carolina Department of Revenue (NCDOR) is a state government agency responsible for administering tax laws, collecting state taxes, and ensuring compliance. With a workforce of 1,001–5,000 employees, it processes millions of tax returns and handles vast amounts of financial data annually. At this scale, manual processes become costly and error-prone. AI presents a transformative opportunity to automate routine tasks, enhance analytical capabilities, and improve service delivery, allowing the agency to do more with existing resources while meeting rising public expectations for digital government services.

Concrete AI opportunities with ROI framing

1. Automated Tax Return Screening: Implementing machine learning models to pre-screen returns can drastically reduce the manual review burden. By flagging anomalies and high-risk returns for auditors, the agency can increase audit yield and recover more revenue from non-compliance. The ROI comes from higher collections with the same audit staff, potentially generating millions in additional revenue annually.

2. Intelligent Fraud Detection: AI systems can analyze patterns across returns, payments, and third-party data to identify sophisticated fraud schemes like identity theft or organized refund fraud. Early detection prevents revenue loss and protects taxpayers. The ROI is direct: every fraudulent refund blocked is money saved, and the system improves over time, reducing future losses.

3. NLP for Taxpayer Correspondence: Natural language processing can categorize, route, and even draft responses to taxpayer letters and emails. This cuts response times from weeks to days and frees staff for complex cases. The ROI includes reduced overtime costs, higher taxpayer satisfaction scores, and improved staff morale by eliminating repetitive work.

Deployment risks specific to this size band

As a large public sector organization, NCDOR faces unique AI deployment challenges. Legacy mainframe systems may lack modern APIs, requiring costly middleware or phased integration. Data silos across different tax types (e.g., income, sales, corporate) must be unified for AI models to be effective. Public procurement rules can slow vendor selection and pilot projects. There is also heightened sensitivity around algorithmic fairness and bias in tax enforcement; models must be transparent and regularly audited to maintain public trust. Change management is critical—training thousands of employees on new AI-assisted workflows requires significant investment in communication and support. Finally, cybersecurity and data privacy are paramount when handling sensitive citizen financial information, necessitating robust governance frameworks around any AI deployment.