Utility Bill Validations

Overview

This document outlines the comprehensive validation strategy employed for processing utility bills. Our approach integrates deterministic pre-processing checks, manual reviews by the data team, and adaptive, antagonistic AI evaluations to ensure data accuracy, integrity, and reliability. This multi-layered framework minimizes errors, flags potential anomalies for human oversight, and continuously improves through learning mechanisms, thereby supporting robust decision-making for utility management.

Pre-Processing Deterministic Validations

These validations consist of rule-based, deterministic checks applied prior to automated bill processing. They serve as safeguards to identify inconsistencies or irregularities that may indicate data entry errors, incomplete information, or deviations from expected patterns.

If any validation fails, the bill is automatically routed to a queue for human review and manual processing. Note that failures do not necessarily imply invalidity; they function as proactive warnings requiring confirmation to maintain data quality.

Validation Criteria

• Charge Summation: Verifies whether the individual charges aggregate to the stated billed total, ensuring arithmetic consistency.
• Non-Zero Charges: Confirms the presence of at least one non-zero charge, preventing processing of potentially blank or erroneous bills.
• Usage Application: Checks for the application of a usage value, ensuring core consumption data is present.
• Demand Application (applicable to relevant accounts only): Validates the inclusion of a demand amount where expected.
• Usage Threshold: Assesses if the current usage is within a factor of 3 compared to the previous month’s or year’s value, detecting unusual spikes or drops.
• Demand Threshold: Evaluates if the demand amount is within a factor of 2 relative to the prior month’s or year’s value, flagging potential anomalies in peak usage.
• Duplicated Usages: Scans for any repeated usage entries that could indicate redundancy.
• Subset Summation for Duplicates: Determines if any subset of usage amounts sums to another usage value, providing an additional layer of duplication detection.
• Service Period Cadence Match: Ensures the bill’s service period aligns with the expected billing cadence (e.g., monthly or quarterly).
• Invoice Date Validity: Confirms the invoice date is in the past, avoiding future-dated anomalies.
• Invoice Date Sequencing: Verifies the invoice date follows the service period end date chronologically.
• Service Period Adjacency: Checks if the service period start date directly abuts existing validated utility data (adjacent to the day), maintaining continuity in historical records.
• Active Account Status: Ensures the bill pertains to an actively tracked account, preventing processing for inactive or irrelevant entities.

Post-Processing Reviews

Following initial processing and commitment of data, post-processing reviews provide an additional audit layer. This combines human expertise with AI-driven analysis to validate structured data against raw inputs, ensuring end-to-end accuracy.

Manual Review by Data Team

Data review is a major daily task for our team.

• Categorized History Graphs: On a daily basis, the data team examines categorized historical graphs for all accounts with newly processed data. This visual inspection easily identifies outliers, trend disruptions, or categorization inconsistencies.
• Ad Hoc Review: In addition to the systematic validations described here, invoices are regularly audited manually by our data team.

AI Evaluation (Olive)

Our AI system, Olive, performs non-deterministic, flexible validations on all committed structured bill data by cross-referencing it against the original raw file contents. This adaptive process focuses on key attributes and evolves over time:

• Account Verification: Confirms the bill corresponds to the correct account.
• Usage Accuracy: Verifies that usage amounts are fully and precisely captured.
• Service Period Validation: Ensures accuracy and completeness of service periods and invoice date.
• Charge Accuracy: Checks for complete and accurate representation of all charges.

Olive assigns a numerical confidence value, along with instructions of how to validate any doubts it encountered. This generates a “short list” for focused human review.

With time, Olive becomes more familiar with accounts. She learns from flagged exceptions, whether confirmed as errors or validated as acceptable, enhancing precision and actionability of future evaluations.