3-minute read
A large California natural gas utility set out to improve eligibility review for the California Alternate Rates for Energy (CARE) program, which helps income-qualified households manage their energy costs. After several years of using a predictive model to support automatic renewal and identify customers for additional eligibility verification, the utility sought to validate the underlying data and identify opportunities to improve prediction performance.
Logic20/20 assessed the available customer data, refined the model, and automated the pipeline for retraining it. Model evaluation showed approximately 77 percent fewer eligibility ranking errors than the previous model, giving the utility a stronger basis for selecting customers for additional verification or automatic renewal.
We brought our expertise in
- Utility customer program analytics
- Data assessment & validation
- Predictive modeling & feature selection
- Model evaluation & deployment
- Automated retraining & performance monitoring
Delivering energy to millions of Californians
More than 20 million consumers depend on this investor-owned utility for energy delivery across hundreds of California communities. Its service territory spans approximately 24,000 square miles of diverse terrain, supporting households and businesses across major metropolitan areas as well as smaller communities.
Keeping eligibility review aligned with customer needs
Administering the CARE program requires balancing continued access for eligible customers with responsible management of program resources and costs borne by ratepayers. Regulatory requirements govern enrollment and post-enrollment verification. Automatic renewal supports continuity of assistance, while carefully targeted verification helps confirm continued eligibility and limits unnecessary documentation requests to customers.
An established predictive model supported customer selection for automatic renewal or additional eligibility verification. After the model had been in use for several years, the utility sought a fresh assessment of the model’s underlying data and prediction performance to identify opportunities for improvement.
Relevant information resided across three databases, with different stakeholders contributing knowledge of the sources and their history. Evaluating the model required:
- Consolidating institutional knowledge
- Validating the available data
- Determining which inputs best supported eligibility predictions
Align data investments with operational priorities
Evidence-based utility decisions: Where data platform value begins explores how utilities can establish dependable data foundations, integrate analytics into daily operations, and measure the business contribution of their data investments.
We will never sell your data. View our privacy policy here.
Strengthening the foundation for CARE eligibility review
Logic20/20 combined data assessment and model refinement with deployment support and automated retraining to strengthen eligibility predictions and ongoing model maintenance.
Assessing the data supporting eligibility predictions
Working with utility stakeholders, Logic20/20 identified relevant data sources, assessed the available information, and addressed gaps in knowledge of the data and its history. The assessment covered customer information such as reported income, household size, enrollment method, and program history. Third-party demographic data provided an additional reference for evaluating the utility’s internal information.
Refining the model's predictive inputs
Logic20/20 used quantitative feature selection to identify the inputs most useful for predicting eligibility and examined correlations among inputs to reduce redundancy. The data science team also evaluated modeling approaches suited to eligibility prediction and validated the updated model’s performance.
Operationalizing model improvements
Logic20/20 supported deployment of the updated model into the utility’s existing production environment and automated the retraining pipeline. Data engineers can incorporate updated data and retrain the model through a standardized process, reducing the effort required for future updates. Performance monitoring helps the utility detect declines in prediction quality and identify the need for model updates.
Improving prediction quality and sustaining model reliability
Logic20/20’s work strengthened the utility’s CARE eligibility review process through:
- Approximately 77 percent fewer eligibility ranking errors compared with the previous model, based on model evaluation results
- Reduced effort for future model updates through a standardized, automated retraining pipeline. At each retraining, feature selection incorporates findings from data discovery across CARE program systems, helping keep the model current as customer data changes.
- Ongoing visibility into prediction quality to inform model maintenance
More reliable predictions and ongoing model oversight strengthen the utility’s ability to target verification requests and support continuity of assistance through automatic renewal for eligible CARE customers.
Ready to sharpen your utility assistance decisions?
Work with our experts to strengthen eligibility review, streamline model maintenance, and support continued access for eligible customers through predictive modeling.
Rodolfo Fiorini is a Senior Data Scientist in Logic20/20’s AI & Analytics practice, specializing in predictive modeling, data engineering, and workflow automation. He develops and deploys analytical solutions that support operational decisions, with a focus on automated pipelines and MLOps practices that sustain model reliability over time.