---
title: "An AI Agent in Energy and Utilities: Meter Readings, Outages and Enquiries"
description: "When the bill arrives or supply drops, everyone calls at once. Where an agent absorbs the peak and where its answer would be outright dangerous."
canonical: "https://interfase.sk/en/news/an-ai-agent-in-energy-and-utilities-meter-readings-outages-and-enquiries"
lang: "en"
last_updated: "2026-11-16"
---

# An AI Agent in Energy and Utilities: Meter Readings, Outages and Enquiries

> When the bill arrives or supply drops, everyone calls at once. Where an agent absorbs the peak and where its answer would be outright dangerous.

Published: 2026-11-16

- Load in utilities comes in bursts — billing and outages create a peak a call centre cannot cover.
- Submitting a meter reading and answering billing questions are ideal tasks: repeated, with an unambiguous answer from data.
- A report of a gas smell, damaged cabling or a leak is not an enquiry — the agent escalates it at once and points to the emergency line.
- The agent must not explain why a bill is high without the data to support it; an unsupported guess is taken as the supplier's position.

The bill goes out and the phone line is blocked for three days. Supply drops in one district and three hundred people call within the hour with the same question. Meanwhile nobody with a different problem gets through.

Utilities carry a load that is bursty by nature — exactly the environment where automation earns its keep. It is also an environment where one badly phrased answer can have physical consequences.

## First, what the agent must not do

Start with the line, because here it is sharper than in any other sector.

If an enquiry mentions **a gas smell, sparking, damaged cabling, water escaping under pressure or work on electrical equipment**, it is not a customer enquiry. The agent **stops handling anything else**, points to the fault and emergency numbers, and escalates the report immediately.

It does not ask for details, does not reassure, does not judge severity, and under no circumstances advises what to do with the equipment. This is configured as a hard rule outside the model — not as a sentence in a prompt.

## Where the saving is largest

**Submitting meter readings.** Taking the reading, checking whether it falls in the expected range, confirming. A repeated operation with unambiguous rules. If the value is out of range, the agent asks for verification or a photograph rather than accepting it and causing a wrong bill.

**Billing questions.** Which period the invoice covers, when it is due, how the advance and the balance split, when the next one arrives. The answers are in the system and they are exact.

**Changing the advance payment, address and contact details.** Ordinary administration that today takes up most of the time at the counter.

**Information about planned outages.** When, where, for how long. This can even be got ahead of — a notice sent in advance cuts call volume far more than any answer after the fact.

## "Why is my bill so high"

The most common question after billing, and the trickiest.

The agent may **show the data**: a comparison with the same period last year, the consumption curve, when the advance changed, whether the reading was actual or estimated. That usually satisfies the customer and it is verifiable.

It must not **infer a cause**. "You probably have a new appliance" or "a tap is likely leaking" is a guess the customer takes as the supplier's position. If the data does not answer it, the case goes to a person.

We describe the same principle in [how to prepare company data for AI](/blog/ako-pripravit-firemne-data-pre-ai): what is not in the material does not get asserted.

## A peak can only be handled by preparing for it

During an outage, everything changes within minutes. The agent has to answer from **the live fault state**, not an overnight copy — otherwise it will insist everything is fine while people sit in the dark.

The practical answer is simple: customer reports are collected with an address, the system groups them, and on the next call from the same area the agent says straight away that the fault is known and gives the expected restoration time. That drops call volume fastest.

## The phone has to stay

Some utility customers will never use a chat, and some of them are in a situation with neither electricity nor internet. The agent is there to relieve the line, not replace it — and the route to a person must not be hidden.

## What to measure

- **Share of meter readings submitted without intervention.**
- **Number of missed calls** during the billing period.
- **Share of fault reports correctly escalated** — this is a safety check, not a performance one.
- **Repeat calls about the same fault** after proactive notification was introduced.

## Where to start

With meter readings. The rules are unambiguous, the risk is low and the effect shows in the next billing cycle. Billing questions come after, once it is clear which data the agent may show a customer.

If you are working on this, [get in touch](/kontakt). More on [AI and automation](/riesenia/ai) and examples in our [case studies](/referencie).


---

Web version: https://interfase.sk/en/news/an-ai-agent-in-energy-and-utilities-meter-readings-outages-and-enquiries
llms.txt: https://interfase.sk/llms.txt
