ML models to detect inaccurate or overestimated energy bills

ML models to detect inaccurate or overestimated energy bills


A nationwide energy provider.


As a leading gas provider, our client sought to increase trust by improving billing transparency. However, relying on manual reviews made it difficult to accurately identify inaccurate or overestimated bills at scale.

To take a more data-driven approach, the client wanted to implement a machine learning solution capable of detecting inaccurate billing patterns across thousands of customers. By accurately flagging potentially unfair estimates, the organisation could correct errors prior to sending bills and provide greater visibility into the billing process.

The ideal solution required a model that could analyse large volumes of customer attributes, usage metrics, rates, and other variables to uncover complex relationships indicative of billing issues. Our client selected Audacia based on our track record delivering custom machine learning solutions.


Audacia worked closely with the client team to develop a machine learning application for identifying inaccurate billing estimates.

We began by selecting Scikit-Learn as the core framework for its accessibility, flexibility, and comprehensive tooling for machine learning. Specifically, we implemented a linear Support Vector Machine (SVM) for its proven capabilities in classification use cases.

Key steps included:

  • Importing and cleansing historical customer and billing data
  • Engineering features from raw data for model input
  • Training the SVM model using labelled data indicating inaccurate bills
  • Optimising model hyperparameters for maximum accuracy
  • Building a web application for analysis and results visibility
  • Creating monitoring dashboards to track model performance
  • Retraining the model periodically as new data emerges

The resulting solution analyses customer attributes and usage data, accurately flagging potentially inaccurate or overinflated bills for further review. This provides greater visibility into the billing process and enables proactive corrections to ensure fairness.


The machine learning solution delivered:

  • Improved accuracy in predicting inaccurate estimates
  • An increase in billing transparency for customers
  • Engineering and manual review time savings
  • Enhanced usage analytics and customer insights
  • Successful pilot of Scikit-Learn for future ML projects

The objective of the project was to leverage predictive capabilities to ultimately strengthen trust and engagement through fair, accurate billing practices. The project also highlighted the advantages of Scikit-Learn for rapidly developing and deploying high-value machine learning applications.

AI ML Case Study Energy Model Diagram