AI use case

AI agent for utility billing, payments, meter readings and move in or move out

An AI agent for energy and water customers that explains bills and tariffs, takes meter readings, sets up or changes payments, and handles move in and move out (final reads, closing one account and opening the next), across phone, messaging, email and the app, while anyone in payment difficulty, in a vulnerable situation or with a complaint is handed to a person.

By Len Debets · Last verified 26 September 2026 · 5 public deployments

75%
Reported containment rate
Aydem Energy, organization claim.
5.6%
Reported contact deflection
EDF, organization claim.
EUR 400,000 to EUR 4.8 million
Indicative value per year
An energy retailer with 1 million residential accounts. Worked example, see how it is calculated.

What problem does it solve?

Alongside outage calls, utility contact centres handle a steady stream of routine account tasks: why is my bill so high, when will I be charged, here is my meter reading, I am moving house. Each one needs the customer's own account, meter and tariff data, which is why static FAQs and old IVR menus rarely finish the job. Volumes are also spiky. PolyAI reports that PG&E saw daily calls in the tens of thousands during weather emergencies, and Aydem Energy says its call volumes rise sharply in seasonal peaks, faster than it can add staff.

Home moves are among the most involved routine tasks. The supplier needs the right date and final readings, has to close one account and open the next without a gap in supply, and often has to deal with an unknown new occupant. Errors here can turn into estimated bills, back billing and complaints months later. Billing is already the largest complaint category at the UK Energy Ombudsman.

  • The UK Energy Ombudsman reports that billing related disputes remain the most common complaint category, accounting for 58% of the 46,532 cases it accepted in the first half of 2026.Energy Ombudsman H1 Data 2026 (2026)

How does it work?

  1. Identify the customer and the account. The agent verifies the customer and finds the supply points, meters and tariffs involved, matching an inbound phone number to the account where the rules allow.
  2. Explain from the customer's own data. Bill questions are answered from the actual bill lines, readings, tariff and payment history, not from generic content, and the agent shows how the amount was calculated.
  3. Take readings and payments. It validates a meter reading against the expected range, asks for a photo when the value looks wrong, and sets up or changes a payment date or amount within the limits the supplier allows.
  4. Run the move as a workflow. For a move out or move in it collects the date, final or opening readings and the forwarding address, closes or opens the account and confirms each step, with a human check before anything irreversible.
  5. Answer policy questions from approved content. Tariff terms, price cap rules and support schemes come from retrieval over the supplier's approved documents.
  6. Hand over at the right moments. Signs of payment difficulty or vulnerability, disputes, complaints and safety issues (such as a gas smell) go straight to a person or the emergency line, with the conversation attached.
Audience
Customer facing
Autonomy
Supervised agent
Adoption
Early adopters
Channels
Phone and voice, Web chat, Mobile app, WhatsApp, Email, SMS

What is it worth?

Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.

Value benchmarks for AI agent for utility billing, payments, meter readings and move in or move out
KPIMedianReported rangeData pointsClaimed by
Interactions handledNot pooled
1000 to 9.2 million
31 organization, 2 vendor
Containment rateToo few to pool
67% to 75%
21 organization, 1 vendor
Contact deflectionToo few to pool
5.6%
11 organization
Customer satisfactionToo few to pool
76%
11 organization
Hours savedNot pooled
35,000 hours
11 vendor
Satisfaction upliftToo few to pool
22%
11 vendor

Value drivers: Lower cost to serve, Customer experience, Speed and cycle time, Inclusion and access.

Indicative value

An energy retailer with 1 million residential accounts

EUR 400,000 to EUR 4.8 million

Human handled contact cost avoided per year

How this is calculated

Formula: accounts * contactsPerAccount * inScopeShare * containment * costPerContact. The low scenario uses every low input, the high scenario every high input.

InputLowHighBasis
Residential accounts accounts, accounts1,000,0001,000,000The reference retailer.
Assisted contacts per account per year contactsPerAccount, contacts per account per year12Editorial assumption, replace with your own contact volume.
Share of contacts about bills, payments, readings and moves inScopeShare, fraction of contacts0.40.6Editorial assumption; billing is the largest complaint category at the UK Energy Ombudsman, but check your own contact reasons.
Share of in scope contacts the agent resolves containment, fraction of in scope contacts0.250.5Conservative against the evidence on this page (PolyAI reports 67% containment at PG&E and Aydem Energy reports 75% of WhatsApp inquiries resolved), because moves and disputes are harder than FAQs.
Cost of a human handled contact costPerContact, EUR per contact48Editorial assumption for a blended phone, email and chat contact. Replace with your own fully loaded cost.

What it leaves out: Gross avoided contact cost only. It leaves out the cost of the AI and the billing system integration, fewer estimated bills and complaints from better readings and cleaner moves, and the peak capacity the agent adds during outages and price changes.

Who already uses it?

5 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.

EDF

United Kingdom · Energy and utilities · 2026

ProductionGrade B

EDF in the UK launched Bill Explainer, an AI powered tool in the customer account that breaks each bill down step by step with explanations tailored to the individual account. After a pilot made available to more than 58,000 customers, EDF reports that billing related contacts fell by 5.6% among customers who used it, and it announced a rollout to its 3.4 million residential and small business customers.

  • Contact deflection: 5.6%, billing related contacts among pilot customers who used the tool
    "Early results indicate a positive shift in how customers engage with their bills, with billing related contacts decreasing by 5.6% among those who used the tool."
    Claimed by: organization

Octopus Energy

United Kingdom · Energy and utilities · 2026

ProductionGrade B

Octopus Energy and Kraken built Arlo, an AI assistant that answers straightforward customer emails about tariff renewals, payment dates and account details. Every AI written email is labelled, the customer can ask for a human at any time, and vulnerable customers, sensitive cases and complex complaints always go to the human team. In a three month UK trial Arlo handled around 8,000 emails a week (4% of customer emails) and scored 76% customer satisfaction against 72% for comparable human replies; Octopus then started a wider rollout while human experts keep reviewing its messages.

  • Customer satisfaction: 76%, three month trial, versus 72% for comparable human replies
    "Arlo achieved a 76% customer satisfaction score, beating comparable responses from human advisors, which scored 72%."
    Claimed by: organization
  • Interactions handled: about 8000, per week during the trial
    "During the three-month trial, Arlo handled around 8,000 emails a week"
    Claimed by: organization

Aydem Energy

Türkiye · Energy and utilities · 2024

ProductionGrade C

Aydem Energy, which supplies electricity to around 6 million customers in Türkiye, built a WhatsApp assistant on Azure OpenAI with Softtech. It opens with data protection consent and an AI disclosure, matches the caller's phone number to CRM data, gives location specific outage updates, guides meter reading submission, checks outstanding bills and processes compensation claims, and escalates urgent or angry customers to a human. Aydem reports about 1,000 inquiries a day, 75% of them resolved without an agent, and it restricts the assistant to the knowledge relevant to each scenario to control cost.

  • Containment rate: 75%, WhatsApp inquiries
    "Of the inquiries that come through WhatsApp, 75% are fully resolved by the digital assistant without the need for live agent support"
    Claimed by: organization
  • Interactions handled: about 1000, per day
    "Today, the digital assistant manages about 1,000 customer inquiries daily through WhatsApp, which represents 10% of total customers calling customer service."
    Claimed by: vendor

Pacific Gas and Electric Company

United States · Energy and utilities · 2024

ScaledGrade C

PG&E, which receives about 16 million calls a year with sharp peaks during storms and outages, deployed a PolyAI voice agent named Peggy. It authenticates customers, gives location based outage updates, answers billing questions and FAQs in English and Spanish and texts links for self service, with integrations into Oracle, Cisco and in house systems. PolyAI reports 67% overall containment, 35,000 labour hours saved and a 22% increase in CSAT on outage calls. Start and stop service, appointment setting and expanded billing were named as the next use cases, not yet reported as live.

  • Containment rate: 67%
    "35,000 labor hours have been saved by the PolyAI agent, increasing CSAT by 22% and achieving 67% containment, 6% higher than the legacy IVR"
    Claimed by: vendor
  • Hours saved: 35,000 hours
    "35,000 labor hours have been saved by the PolyAI agent, increasing CSAT by 22% and achieving 67% containment, 6% higher than the legacy IVR"
    Claimed by: vendor
  • Satisfaction uplift: 22%, outage calls
    "35,000 labor hours have been saved by the PolyAI agent, increasing CSAT by 22% and achieving 67% containment, 6% higher than the legacy IVR"
    Claimed by: vendor

Dubai Electricity and Water Authority

United Arab Emirates · Energy and utilities · 2017

ScaledGrade C

DEWA introduced its Rammas chatbot in 2017, which Microsoft describes as the first chatbot based on Microsoft AI at a government utility, and later integrated Azure OpenAI Service to make its answers more accurate and more natural. Microsoft reports that Rammas has responded to more than 9.2 million customer inquiries autonomously since launch, for a utility with more than 1.2 million customers. DEWA also runs AI based high water usage alerts, and is exploring voice conversations for customers.

  • Interactions handled: at least 9.2 million, cumulative since 2017
    "Since then, Rammas has responded to over 9.2 million customer inquiries autonomously."
    Claimed by: vendor

How do you implement it?

A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.

Data you need

  • Account, bill line, tariff and payment data reachable through APIs
  • Meter data with expected reading ranges per meter
  • Approved tariff terms, regulatory rules and support scheme content
  • Vulnerability and priority services flags with the rules for using them
  • Contact reason data to choose the first intents

Systems to integrate

  • Billing and customer information system
  • Meter data management
  • Payments and direct debit platform
  • Industry registration or switching processes for moves
  • Contact centre platform for handover and callbacks

Complexity: Medium

Answering is simple; acting on the account is where the work lies. Moves touch the billing system, meter data, industry registration processes and sometimes a credit check, and every write needs clear limits and a way back.

  1. 1

    Start with readings, bill explanations and payment dates

    These are high volume, low risk and easy to check. Octopus Energy started its email assistant on tariff renewals, payment dates and account details and kept complex complaints, sensitive cases and vulnerable customers with people.

  2. 2

    Ground every bill answer in the account

    Give the agent tools that return the bill lines, readings and tariff for this customer, and make it show the calculation. Generic answers about "how bills work" do not reduce repeat contact.

  3. 3

    Treat moves as a workflow with checkpoints

    Model move in and move out as steps with validation (date, readings, address) and a confirmation at each step, and keep a human approval for account closure until error rates are known.

  4. 4

    Build vulnerability detection into the flow

    Define the phrases and signals (missed payments, health conditions, distress) that route to a specialist, and check the Priority Services Register or its local equivalent before any change.

  5. 5

    Plan for the peak

    Outages and price changes bring the surges. Make sure the agent can carry outage updates at volume, as Aydem Energy and PG&E do, so billing questions still get through.

Guardrails

  • No disconnection, debt or payment plan decisions by the agent; those go to trained staff
  • Payment changes only within limits the supplier sets, with confirmation to the customer
  • Readings outside the expected range rejected or sent for review, never billed as given
  • Safety issues such as a gas smell routed to the emergency line immediately
  • AI disclosure on every channel and an easy route to a person

KPIs to instrument

  • Containment per intent, counting repeat contact within seven days as not contained
  • Estimated bills and back billing cases after agent handled moves
  • Customer satisfaction for AI handled versus human handled contacts
  • Handover rate and reasons, including vulnerability referrals
  • Complaints that mention the assistant

Human in the loop

Specialists handle payment difficulty, vulnerable customers, disputes, complaints and failed moves. Operations leads approve each new intent and action, and a sample of contained conversations and AI written emails is reviewed every week. Octopus Energy says its human experts keep checking the messages its assistant sends.

Common failure modes

Wrong reading, wrong bill
A mistyped or misread meter value flows into billing. Validate against expected ranges and ask for a photo when in doubt.
Broken moves
The old account is closed before the new one is confirmed, or readings are missing. Use a workflow with checkpoints and human approval for closures.
Missing vulnerability
A customer in difficulty is handled like any other and put on a payment plan they cannot keep. Detect signals early and hand over.
Generic bill explanations
The agent explains how bills work in general instead of this bill, and the customer calls anyway. Give it the bill data and measure repeat contact.

What are the risks and rules?

EU AI Act

Depends on design

A customer service agent for bills, readings and moves falls under the transparency duty for systems that interact with people (Article 50(1)): customers must be told they are talking to AI. If the agent assesses creditworthiness, for example to set a deposit when a new customer moves in, that part falls under Annex III point 5(b) and is high risk; keep credit decisions in separately governed systems. The agent is not a safety component in the operation of the gas, water or electricity supply (Annex III point 2), so safety reports such as a gas smell go straight to the emergency line rather than being handled by the agent.

Rules that apply

Guidance

  • Ethical AI use in the energy sector (Ofgem, Europe). Good practice guidance for energy companies, updated in May 2026, including a section on AI in consumer interactions with transparency and proportionate explainability.

Controls to put in place

  • AI disclosure and a documented route to a human on every channel
  • Action allow list with limits for payments, readings and account changes
  • Audit trail of every account change the agent made, with the verification used
  • Vulnerability and priority services checks before any change
  • Change control and regression tests for each new intent

Frequently asked questions

Do customers accept AI answers from their energy supplier?
Early evidence suggests so, when the AI is transparent and a human is one reply away. Octopus Energy reports that its email assistant Arlo scored 76% customer satisfaction in a trial, against 72% for comparable human replies, with every AI written email labelled.
What share of utility calls can an AI agent resolve?
PolyAI reports 67% containment for PG&E's voice agent, which handles outage and billing calls, and Aydem Energy reports that 75% of the inquiries reaching its WhatsApp assistant are resolved without an agent. Moves and disputes are likely to resolve less often than outage updates and FAQs.
Should the agent handle move in and move out end to end?
It can collect everything and run the steps, but keep a human check before closing an account until you know the error rate. Public outcome data for AI handled moves is still scarce; PolyAI lists start and stop service at PG&E as a use case being built, not as a result.
Can AI help with bill confusion without a chatbot?
Yes. EDF's AI Bill Explainer breaks each bill down step by step in the customer account, and EDF reports 5.6% fewer billing related contacts among customers who used it in the pilot.

How to cite this page

Blits.ai AI Use Case Library, "AI agent for utility billing, payments, meter readings and move in or move out", last verified 26 September 2026, https://www.blits.ai/ai-use-cases/utility-billing-and-move-agent. Licensed under CC BY 4.0. Method: how we verify use cases.

Changelog
  • 27 September 2026: First published

Related use cases

Telecommunications

AI agent for telecom bill explanation and billing disputes

An AI agent that explains a customer's telecom bill line by line, in plain language and on any channel, answers why a charge changed or appeared, corrects clear errors within set limits and opens a billing dispute with the evidence attached when a human has to decide.

Deployments
4 public, best grade B
Reported first contact resolution
60%
Vodafone, organization claim
Cross industryBanking

AI agent for early collections and hardship support

A voice and messaging agent that contacts customers in early arrears and answers their inbound calls, takes payments and sets up payment arrangements within preapproved rules, and recognises signs of hardship or vulnerability so those customers go straight to a trained person.

Deployments
2 public, best grade C
Reported cost reduction
75%
SameDay Auto Finance, vendor claim
Cross industryHealthcare

AI agent for outbound reminders and confirmations by voice and messaging

An AI agent that contacts customers about something they already booked or ordered (an appointment, a delivery, a reservation or a service visit) to remind them, confirm attendance and let them cancel or move it in the same conversation, by phone, SMS, WhatsApp or email. It is operational service outreach, not marketing: nothing is sold, and success is measured in kept appointments and reused slots, not in conversion.

Deployments
4 public, best grade B
Autonomy
Supervised agent
Cross industryBanking

AI agent for first line contact centre service

An AI agent that answers the first line of inbound customer contact on phone, chat and messaging, resolves general and routine questions end to end in the customer's own language, and routes everything complex, sensitive or regulated to the right human team with the context attached.

Deployments
18 public, best grade B
Median containment rate
47%
7 deployments
Cross industryBanking

AI agent for complaints recognition, investigation and response

An AI agent that recognizes when a customer interaction is a complaint, logs it against the regulatory definition, classifies its root cause and severity, gathers the evidence, drafts the acknowledgement and the response for a human handler to approve, and tracks every statutory deadline until the case is closed.

Deployments
2 public, best grade B
Reported time saved per task
about 5 minutes
Lloyds Banking Group, organization claim