Benefit theft is a serious issue nationwide, including California. While California counties, who administer benefits, work as quickly as possible to reimburse theft victims, this has been a persistent issue that impacts critical safety net programs, such as CalFresh and CalWORKs.
California, in partnership with federal and local officials, continues to take steps to mitigate the theft of benefits and investigate those who are engaging in these illicit activities.
As part of these continued efforts to fight benefit theft, the California Department of Social Services (CDSS) recently partnered with the Office of Data and Innovation’s (ODI) data science and data operations teams to find an innovative and efficient solution. Together, they built an AI-powered machine-learning model that is providing a new way to help the state detect and respond to benefits theft.
Specifically, it flags benefit theft more quickly and accurately, with the ultimate goal of better informing theft mitigation efforts and supporting ongoing and future investigations.
A key component of this effort is data-driven research, which can reveal patterns and insights into how these crimes occur.
Overcoming data challenges
CDSS worked with the ODI to create the infrastructure to enable rapid processing of data that can help the state identify suspicious activity and take action to protect client benefits.
One of the biggest hurdles in combating EBT theft has been access to timely and comprehensive information. CDSS researchers needed a way to more quickly identify suspicious transactions and patterns. ODI built our partner department a technical foundation, specifically, a comprehensive data pipeline to detect suspicious transactions and patterns.
—Monica Bobra, Principal Data Scientist, ODI
Beyond just flagging suspicious transactions, the model incorporates explainability features, helping the team understand and explain why the model flagged specific transactions as illegal or legitimate. This transparency is crucial for building trust with both CDSS staff and the communities they serve.
The power of machine learning
ODI is home to subject matter experts in data science and data engineering. Our unique role in the state, as a government-to-government service, not only advocates for upskilling state teams in this effort, but also making more efficient use of data to improve services.
A machine learning model was trained to predict which transactions were likely fraudulent. This model used various features, such as location characteristics, retailer information, and temporal factors, to identify unauthorized transactions with a high degree of accuracy.
ODI developed a model that significantly improves the detection of EBT theft. The initial pilot showed impressive results:
| Speed | Detection time dropped from 2 months to just 72 hours — a 95% improvement that means families get help faster |
| Data collection | Automated (saving 2,160 staff hours annually) |
| Theft identification | 82% accuracy in identifying theft |
| Theft location | Precise geographic data provided |
What’s next
- Looking at if adding the transaction’s time could increase accuracy
- Extending the model to cover CalFresh
- Improving the data pipeline’s geographic data coding
- Monitoring the impact of upgrades to EBT card technology: California is the first state in the nation to roll out Chip and Tap Cards for EBT benefits.
CDSS is in the middle of a once-in-a-generation modernization of our data infrastructure along with a push to hire data scientists and engineers. ODI dramatically accelerated our efforts to use our new infrastructure and people to ingest, analyze, and extract insights from vast quantities of data – insights that will improve services and protect benefits for Californians.
—Joaquin Carbonell, Research Data Supervisor II, CDSS
How did we do it? We invite you to read the technical paper on this partnership.And you are also invited to read the case study.