CASE STUDY:
Code For America Transforms Lives:
How AI-Powered Entity Resolution Cleared 350,000 Criminal Records in Utah
When Utah passed its Clean Slate law, the goal was clear: help people with past criminal records move forward with their lives by automatically identifying and clearing eligible records, without any action needed on behalf of the person with the record. The challenge? Implementing this groundbreaking legislation would require examining records at a scale that included data from nearly 150 different courts and 3 million records โ a task that seemed insurmountable through traditional, manual means.
Enter Code for America, a civic tech nonprofit with over 15 years of experience partnering with governments at the state and local levels. The organization works โto improve benefits delivery through human-centered technology,โ explains Jenn Thom, Senior Director of Data Science.
A Mountain of Data, A Sea of Variations
The scale of implementing the Clean Slate law was staggering. “Utah’s courts were directed to examine millions of case records spanning multiple judicial districts to determine whether each individual record qualified for expungement… they needed to construct a complete history for every defendant in their system,” explains Chris Correa, Director of Solutions Engineering for Criminal Justice at Code for America.
Making matters more complex, these records lacked unique identifiers, such as a Social Security number or driverโs license number. The same person might appear multiple times across different courts with varying spellings, aliases, addresses, or dates of birth. As Correa notes, “Every record was associated with a name. Most had a date of birth and address… We needed to handle common data entry errors and not just errors, but natural variations that occur.”
Code for America Delivered
99.9% Accuracy
To tackle this challenge, Code for America partnered with Senzing for entity resolution technology to accurately match, deduplicate and resolve records. Privacy and data security were paramount considerations throughout the implementation, with robust protocols ensuring sensitive information remained protected while still enabling efficient processing.
During validation testing, Senzing Insideโข entity resolution achieved 99.9% accuracy in identifying matching records, outperforming even expert human reviewers. “We found it’s much more likely for a person โ even a person with domain knowledge and skills โ to make an occasional mistake… than it is for the state-of-the-art entity resolution solution,” Correa explains.
The impact of this technology-driven approach was immediate and substantial. What would have taken court staff decades to review manually was accomplished in days. The system processed over 3 million records, consolidating them into approximately 1.2 million unique individuals โ effectively identifying and clearing eligible records at an unprecedented scale.
“We found it’s much more likely for a person โ even a person with domain knowledge and skills โ to make an occasional mistake… than it is for the state-of-the-art entity resolution solution.”
โ Chris Correa
Director of Solutions Engineering
Code For America
“The Clean Slate law will affect nearly 350,000 people in Utah,” says Ron Gordon, Utah State Court Administrator. “Code for America played a critical role in the implementation process, developing the technology and processes that will identify the cases and people who qualify for automatic expungement. This allows for convictions to be automatically cleared at scale. We’re grateful for their expertise and the technology tools that will now help to give second chances to thousands of Utahns.”
Engineering Excellence for Social Change
The technical implementation leveraged sophisticated tools while maintaining strict data governance standards. Senzingยฎ entity resolution provided multi-cultural name matching trained on more than 800 million names. This allowed the solution to handle complex name variations, including different cultural naming conventions and common misspellings, assuring eligible individuals wouldn’t miss out on relief due to simple clerical errors or data inconsistencies.
Unlike many AI and ML techniques that must initially be trained using extremely large data sets, Senzing entity resolutionโs purpose-built AI is pre-trained, pre-tuned and highly accurate from day one.
Senzing entity resolution AI does not use LLMs and is not any other form of GenAI. It is a smart principle-based system and produces consistent and explainable results with accuracy that outperforms humans. “We see AI as a tool in our overall toolbox,” notes Jenn Thom, “and we’re really cautiously optimistic about the application of AI to help improve benefits delivery.” This measured approach to technology adoption, combined with a strong focus on privacy and ethical considerations, helped ensure the project’s success.
A Clean Slate Model for the Future
Utah’s success story offers a blueprint for how other states can implement similar automatic record clearance programs. Code for America continues to expand this work, providing support to other states as they implement their own Clean Slate initiatives.
“One of the things that we find to be helpful when we first work with states and localities on different solutions is to try to get in early and prototype and try to get a sense of the data we’re working with,” shares Michael Cowden, Senior Director of Solutions Engineering at Code for America.
The project demonstrates how leveraging advanced entity resolution technology can transform government services, making them more efficient, more accurate and more equitable. In Utah, 350,000 residents now have a clearer path to housing, employment and other opportunities that might otherwise have remained out of reach.
“They delivered meaningful relief for thousands of residents who no longer had to navigate a costly and burdensome petition-based process to seal their records,” Correa shares. “And perhaps most impressively, Code for America quickly and easily leveraged Senzing entity resolution for more accuracy than a traditional manual review.”
Perhaps most impressively, Code for America quickly and easily leveraged Senzing entity resolution for more accuracy than a traditional manual review.
โ Chris Correa, Director of Solutions Engineering
This article is based on Better Government for All: A Practical Introduction to Entity Resolution with AI, an InnovateUS workshop led by Jeff Jonas (CEO, Senzing) and Michael Cowden, Jenn Thom and Chris Correa from Code for America.
