What makes Hawaii’s success story remarkable is not only the technical innovation itself, but also the fact that these advancements were largely led by internal staff who were initially unfamiliar with AWS. Through iterative learning, partnership with AWS, and a clear mission to improve compliance and efficiency, the department has implemented a series of impactful projects that improve operational performance while reducing administrative burden.
Project Overview
The Hawaii Department of Taxation focused on practical, high-value opportunities to introduce AI into its operations. Four standout projects demonstrate how a smart, incremental approach can deliver outsized impact:
1. AI-Enhanced Compliance Reviews with Entity Resolution
To identify unreported income, the department compares publicly available datasets, such as federal contractor records from USASpending.gov and Airbnb listings, with internal tax return data. One challenge: inconsistency in how names and addresses are formatted across sources.
To resolve this, the department developed a fuzzy matching engine using AWS AI tools that assigns confidence scores to potential matches. This allows auditors to quickly zero in on high-probability leads, reducing the review pool to only the most promising cases, allowing the agency to streamline workloads and increase the likelihood of revenue recovery.
2. Intelligent Document Processing (IDP) for Imaged Tax Documents
Each tax season generates a flood of supporting documents that often go underutilized due to the manual effort required to process them. To change this, the department is piloting Intelligent Document Processing (IDP) using Amazon Bedrock, enabling automated extraction of relevant data from previously scanned forms.
This allows the department to enhance its compliance reviews and research efforts without slowing down the return processing timeline, unlocking valuable insights from documents that were once siloed in image formats.
3. Automated Software Testing with Bedrock-Generated Test Scenarios
Testing tax software each year is a critical but time-consuming task. Business analysts must ensure that every field on each return accepts only valid values, which traditionally requires crafting hundreds of test scenarios and inputs manually.
Now the department can generate these test cases automatically based on business rules and field definitions. This automation significantly reduces preparation time, improves accuracy, and enables the team to focus on refining functionality and user experience.
4. Internal AI Chatbot for Staff Support
As a forward-looking initiative, the department is piloting an internal chatbot to help employees answer tax-related questions. This tool is expected to reduce internal support time, improve knowledge access, and increase staff productivity, especially during peak filing seasons.
Challenges and Solutions: Learning by Doing
The Hawaii Department of Taxation team faced steep learning curves, limited budget, and technical barriers within its legacy systems. However, a few key strategies helped them succeed:
- Experimentation as a learning model: Staff were encouraged to test multiple AWS services using their own datasets, helping build hands-on experience and identify the most suitable tools for each use case.
- Strategic support from AWS: Weekly and biweekly touchpoints with AWS experts provided critical guidance and troubleshooting, accelerating progress and knowledge transfer.
- Right-sizing AI projects: Instead of jumping into complex AI use cases, the team began with small, well-scoped datasets, refining models and prompts to ensure manageable resource consumption and meaningful results.
Outcomes and Benefits
- Audit efficiency: The AI-powered entity resolution solution reduced the manual workload by confirming that only 30% of reviewed contractor data required further scrutiny, which allows auditors to focus on high-value cases.
- Data accessibility: The IDP pilot promises to unlock structured data from large volumes of scanned documents, without disrupting core operations.
- Testing productivity: Automating test case creation accelerates software validation, enhances accuracy, and shortens the testing lifecycle, ensuring greater reliability and readiness ahead of tax season.
- Scalability: The entity matching and document processing models are designed to scale — applying to new datasets or use cases without major redevelopment.
- Empowered staff: Internal employees, once new to AWS, now lead AI development initiatives, demonstrating how skill-building and ownership can flourish even in resource-constrained environments.
