Automating energy billing: A new beginning

Author: Sasank Chilamkurthy

One of our earliest customers was a large electricity distribution company responsible for millions of consumers. Their billing department was drowning in paperwork, manual meter readings, and disputes. Each month, the process of generating bills required hundreds of staff members working overtime to reconcile readings, validate tariffs, and respond to complaints.

The Challenge

The existing workflow looked something like this: field staff captured meter readings on paper, the data was manually entered into a centralized system, and then a team of accountants verified every entry against the applicable tariff slab. Errors were common, disputes were frequent, and the entire cycle took weeks to complete. Consumers were frustrated, and the utility was bleeding money.

Privacy was another major concern. Billing data contains sensitive consumer information, including addresses, usage patterns, and payment history. Storing this data on a public cloud was not an option because of regulatory requirements and internal security policies.

Our Approach

We deployed a JOHNAIC unit inside the customer's data center. Because JOHNAIC is an on-premise personal AI computer, the data never left the building. This single decision addressed the privacy concern completely.

Next, we built a pipeline that combined computer vision and natural language processing:

1. Meter reading sheets were scanned and digitized using an OCR model fine-tuned on local fonts and handwriting.
2. The digitized readings were validated against historical patterns to catch outliers before billing.
3. A language model generated human-readable explanations for any anomalies, helping the dispute-resolution team respond faster.

The Result

The billing cycle shrank from three weeks to three days. Disputes dropped by 40% because consumers received clearer, more accurate bills. Most importantly, the entire workload ran on hardware owned and controlled by the utility.

This project taught us a valuable lesson: AI does not have to live in the cloud to be powerful. When data sovereignty matters, on-premise AI is not just an alternative, it is the only responsible choice.

Published on 15/09/24