Agentic System for Automated Meter Reading Verification and Bill Adjustments

Instead of a customer service agent manually logging into the MDMS, downloading interval data, and comparing it to historical usage—a 15-20 minute task prone to error—an AI agent executes this in seconds. The workflow retrieves the last 30-90 days of 15-minute interval data, normalizes it for weather, and flags anomalies exceeding configurable thresholds (e.g., 50% spike). This direct integration eliminates the investigative bottleneck, freeing agents for complex cases.

The traditional path requires a ticket to the billing department, which manually reviews the case in a days-long queue before issuing a credit. This workflow embeds pre-approved business rules: if the agent confirms a meter fault or unexplained usage spike, it can autonomously generate a prorated adjustment, route it for a one-click supervisor approval if required, and push the credit directly to the CIS and billing engine. Resolution shifts from a multi-day back-office process to a near-real-time customer interaction.

A significant portion of high-bill complaints historically led to a field technician dispatch for a meter test—a costly 'truck roll' often yielding no fault found. The agentic system first performs a remote diagnostic check, querying the MDMS for meter health flags (last gasp, communication failures, reverse rotation). If remote diagnostics confirm a healthy meter and the usage pattern is consistent (just high), the agent can explain this to the customer with data, avoiding a $150-$300 field visit and reallocating crew capacity to real faults.

Customer distrust often stems from opaque processes. This workflow generates a plain-English, data-backed summary for the customer: 'We reviewed your hourly usage from July 1-31. Your consumption peaked between 2-5 PM daily, coinciding with your pool pump schedule. The meter communicated reliably. Here’s a comparison to your June usage.' This transparent evidence, delivered via SMS or portal, defuses frustration. Every agent decision, data point queried, and rule applied is logged to an immutable audit trail, providing defensibility for regulators and reducing complaints to the public utility commission.

The system shifts from reactive complaints to proactive service. By running the verification logic on a scheduled basis (e.g., nightly) for all accounts with bill spikes exceeding a threshold, the utility can identify and notify customers of potential meter issues or unusual usage before they call. This transforms a cost center (the complaints department) into a trust-building, proactive service operation, reducing inbound volume and catching meter faults earlier.

Implementing this as a custom workflow, not a one-off script, means building in the necessary controls for a regulated environment. This includes configurable approval gates (e.g., adjustments over $500 require supervisor review), integration with formal case management systems for tracking, and seamless hooks into the utility's existing governance and reporting frameworks. The architecture—using LangGraph for orchestration, with agents for MDMS querying, analytics, and CIS updates—becomes a reusable platform for other revenue-protection workflows.

The workflow initiates when a high-bill complaint is logged via IVR, web portal, or CRM. An ingestion agent immediately pulls the customer's last 12-24 months of smart meter interval data from the MDMS (Meter Data Management System) and retrieves the disputed bill from the CIS (Customer Information System). This layer normalizes time-series usage data and flags any recent meter health alerts (e.g., communication failures, reverse flow) for downstream analysis.

A dedicated analytics agent performs the core verification logic. It compares current usage against historical profiles, adjusting for weather (via integrated NOAA data) and seasonality. It runs statistical tests for spikes and identifies patterns indicative of meter malfunction, in-home appliance faults, or unauthorized usage. The agent outputs a confidence score and a structured finding (e.g., 'Confirmed Meter Fault', 'Usage Spike Aligns with Weather', 'No Anomaly Detected').

Based on the anomaly finding and pre-configured business rules, an orchestration agent routes the case. For high-confidence meter faults, it can autonomously generate a bill adjustment within policy limits and push it to the billing engine. Cases requiring human judgment (e.g., high-value adjustments, ambiguous patterns) are queued in a supervisor dashboard with the agent's full analysis. This agent also logs all actions for the audit trail.

A communication agent manages all customer-facing interactions. It drafts a plain-language explanation of the findings, detailing usage comparisons and the adjustment applied (if any). It delivers this via the customer's preferred channel (SMS, email, portal) and can handle follow-up questions in a chat interface. Finally, it updates the CRM case and triggers any required field work orders (e.g., for meter replacement).

The entire workflow is instrumented for real-time observability. A monitoring dashboard tracks key metrics: volume processed, auto-adjustment rates, customer satisfaction (post-resolution surveys), and adjustment value. Every agent decision is logged with a complete audit trail, including the data used and the logic applied. Regular drift detection runs on the analytics model to ensure accuracy over time, and all autonomous financial actions require periodic policy re-certification by finance and regulatory teams.

Built on an orchestration framework like LangGraph or Temporal, the workflow integrates via APIs with the MDMS (e.g., Itron, Landis+Gyr), CIS (e.g., SAP CC&S, Oracle), and billing system. For legacy systems without APIs, browser-automation agents can simulate user logins to retrieve data. Deployment is typically phased, starting with read-only verification to build trust, followed by supervised auto-adjustments, and finally full autonomy for clear-cut cases. The architecture is containerized for scalability during complaint surges.