Monthly window completed. 18,400 valid reads routed to billing, 420 non-communicating meters with substitute read, 180 classified anomalies.
AGENT · SMART METER READER COORDINATOR
Smart meter reads feed billing without manual intervention.
Smart Meter Reader Coordinator orchestrates the flow of smart meter reads (electricity, gas, water) into the utility's billing system. It identifies missing reads, consumption anomalies, and atypical peaks. It coordinates with the field team for manual interventions when a smart meter is unresponsive. Integration with the smart metering system and the billing system is delivered during the project.
Smart Meter Reader Coordinator at work.
Taking the 180 anomalies. The 40 with a new-business pattern go to the commercial team, the rest to customer service.
Field ticket opened for the 420 non-communicating meters, list ordered by geographic zone. Event traced in audit.
Why it exists.
Utilities with smart meters installed (second-generation electricity meters, electronic gas meters, digital water meters) manage an automatic read flow that feeds the billing system. The operational challenge is not the standard read — it is exception handling: non-communicating meters, reads that deviate from historical consumption, peaks that need investigation. Exceptions handled manually absorb resources and generate billing errors.
How it coordinates the flow.
The agent activates within configured read windows. For each customer: it retrieves the read from the smart metering system, validates against the configured consumption history, activates the billing flow for valid reads. For non-communicating meters: it activates a substitute read based on historical consumption. For consumption anomalies — suspicious peaks, atypical reads — it flags the case to the customer service team with full context.
Handling stays with the operations team.
The agent does not call the customer, does not suspend supply, does not close cases. It sets up the separation between the standard flow and the exceptions flow, classifies exceptions by urgency. Intervention decisions — calling the customer, sending a field technician, opening a faulty meter case — stay with the operations team.
Who it serves and where it applies.
Utility operations team
Recovers the time spent manually handling read exceptions. Exceptions arrive already classified and prioritised; the team focuses on the actions that require human intervention — calling the customer, sending a field technician, opening a faulty meter case — not extraction and classification.
SCADA · live 2 anomalies
Substation SE-104 voltage out of range
Line LV-22 current spike
NIS2 classify material · 24h timer
Alert to CISO · ticket opened
Billing team
Has confidence that reads entering the billing system have been validated against history. Substitute reads are documented and distinguishable from actual reads in customer billing. Billing goes out consistent with historical consumption; disputes go down.
Ongoing outage ETA 35'
Zone Northeast cluster B
Field team 3 crews · in transit
Customer SLA reset · 20'
Regulator report in preparation · 4h
Customer service manager
Receives consumption anomalies relevant to the customer — unexpected spikes, atypical usage — with the context already prepared for handling. The conversation with the customer changes nature: it stops being reactive, it becomes proactive on significant anomalies.
14-day forecast 48 sections
MV-North section high
MV-East section medium
Assets > 25 years 12 replacements
Weekly maintenance plan
A concrete example.
On the first Monday of the month, the agent starts.
For a multi-service utility, the monthly read window opens on the first Monday of the month. The agent activates across the full customer list in the smart metering system. Within a few hours it processes every meter. For 18,400 out of 19,000 meters, reads arrive as expected and route directly to the billing system.
420 non-communicating, 180 anomalies.
For 420 meters, communication failed: the agent activates a substitute read for each (estimate from seasonal consumption history) and creates the ticket for the field team — list ordered by geographic zone. For 180 meters, the read shows an anomaly: consumption more than 40% above the twelve-month average for the same period.
The 180 anomalies come out already classified.
The agent classifies the anomalies: 40 cases with an anomaly consistent with the start of new production activity (flagged to the commercial team as a possible contract update), 140 cases for the customer service team to investigate (proactive contact with the anomaly detail). The read summary — valid, substitute, anomalies — arrives in the operations team's channel by that morning.
Configuration and technical resources.
The rules are declarative. The utility's operations and billing team defines, in a readable format, the read windows, anomaly thresholds by commodity and consumption band, substitute read estimation rules. The rules live in the customer's repository, versioned, validated at agent startup. The estimation rules for non-communicating meters follow the criteria configured by the operations team, consistent with the applicable regulatory guidance per commodity.
- Language
- TypeScript (Node.js)
- LLM model
- customer's choice: Anthropic, OpenAI, Mistral, open source models hosted internally, AWS Bedrock for a private model
- Built-in controls used
- pii-detector, tool-rate-limit
- Native delivery channels
- Slack, Telegram, HTTP OpenAI-compatible
- Smart metering system integration
- dedicated adapter built during delivery
- Billing system integration
- dedicated adapter built during delivery
- Memory
- persistent per instance
- Registry
- append-only, queryable with a standard SQL client
Frequently asked questions about the agent.
The substitute read for non-communicating meters follows the estimation rules configured by the utility's operations team, consistent with the applicable regulatory guidance for each commodity. The estimation rules are declarative and versioned in the repository. The audit registry traces every substitute read generated, with the rationale, available for inspection.
The agent compares the period's consumption against the twelve-month baseline for the same period and against the previous month's consumption. Anomalies classified as potential new-activity have a consistent growth pattern; anomalies to investigate have isolated peaks or discontinuities. The classification rules are declarative and refined by the operations team on real data.
The typical pattern is 12-18 weeks. Discovery 2 weeks, validation rules and threshold configuration with the operations team 3-4 weeks, smart metering system and billing system integration 5-8 weeks, hand-off. The effective duration is defined in discovery on the real case.
From a 30-minute conversation to the squad in production.
A 30-45 minute conversation to understand how Smart Meter Reader Coordinator would configure to the utility's case. Which smart metering system, which billing system, how many commodities are in scope, which anomaly thresholds are already in use.