What do energy and utility customers contact you about?
Most messages to an electricity, gas or water supplier concern a bill, a meter or a move. Agent can resolve those once it can read the account and write a meter reading back. Safety, affordability and complaints go to trained people.
| Conversation type | Can Agent resolve it? | Hand off when | Data Agent needs |
|---|---|---|---|
| Submitting a meter reading | Yes: validates it against the meter and posts it | It fails validation twice, or the meter may be faulty | Meter registers, last actual reading |
| "Why is my bill so high?" | Yes, when an estimate, rate change or winter use explains it | The customer still disputes it, or it back-bills over 12 months | Bill lines, reading history, tariff dates |
| Balance, due date or direct debit amount | Yes: explains and recalculates within your rules | The customer says they can't afford the new amount | Balance, payment schedule, usage forecast |
| Moving home | Yes: records dates and readings, closes or opens the account | Debt at the premises, or the account holder has died | Premises, meter points, account status |
| Struggling to pay | Partly: shares support options, books a callback | Always | Balance, arrears, vulnerability flag |
| Smart meter not sending readings | Partly: checks status, books an engineer | A fault code, or no supply | Head-end status, appointment slots |
| Prepayment top-up not on the meter | Partly: confirms the payment and how to apply it | At once if off supply or vulnerable | Top-up transactions, meter credit |
| Power cut or no water | Yes for logged incidents: shares the restoration estimate | No incident logged, or powered medical equipment at home | Incident feed by postcode or ZIP |
| Smell of gas or sparking meter | No: emergency number in the first reply | Always, immediately | Safety script in Knowledge |
| Complaint | Logs and acknowledges it; never closes it | Always | Complaint log, account history |
What does an AI procedure for estimated bills look like?
Bills top the problems Ofgem's complaints guidance tells households to raise with their supplier, and many high-bill messages trace back to estimated readings. Smart meters have not ended that: at the end of June 2026, 72% of meters in Great Britain were smart or advanced, yet about 3.1 million smart meters were running in traditional mode, per government smart meter statistics. A billing team could write the procedure like this.
Example procedure: high bill after estimated readings
- Verify the customer with the account number plus postcode or ZIP code, or a logged-in app session. After two failed attempts, hand off.
- Read the latest bill, the reading history (actual or estimated) and any rate change in the period.
- If the bill used an estimate, ask for today's reading and a photo of the meter display.
- Validate the reading: the serial number must match, it cannot be below the last actual reading, and daily use must sit between half and three times last year's.
- If it passes, submit it, request a re-bill and say when the corrected bill will appear. If it fails twice, open a meter investigation task and hand off with the photo.
- If a rate change or winter use explains the bill, say so from the bill lines. Never forecast bills or recommend a tariff.
- If the bill charges for energy used over 12 months ago, hand off to billing to apply the back-billing rules.
- If the customer can't pay, or mentions illness, medical equipment or a bereavement, stop discussing the bill and book a same-day callback.
- Spend limit: £0. Agent never applies credits or changes payment plans. Words such as complaint or ombudsman log a complaint and hand off.
Example conversation
Customer: My bill says £318. It's normally about £120. What's gone wrong?
Agent: I can look into that. Please confirm your account number and postcode.
Customer: 40021337, LS6 2QT.
Agent: Thanks. Your last two bills used estimates because we haven't had a reading since August. If you send today's reading and a photo of the meter, I can correct it now.
Customer: [photo] 23817.
Agent: That matches your meter and your normal use. I've submitted it, and a corrected bill of £131.40 will replace the £318 one by tomorrow.
Customer: Thanks. It's been hard since my mum died. She used to deal with all this.
Agent: I'm very sorry about your mum. Someone from our support team will call you before 6pm today to talk through extra support.
What guardrails does an energy or utility AI agent need?
Agent puts safety first, never applies collection pressure, never decides affordability or complaints, and always says it is AI.
What Agent must never do
- Troubleshoot a gas smell. The HSE domestic gas FAQ tells anyone who suspects a leak to call 0800 111 999 in Great Britain, open doors and windows, and shut the supply at the meter if they know how. Agent gives your emergency number first and stops.
- Discuss disconnection or prepayment installs. Ofgem's prepayment rules require at least 10 contact attempts and a home visit before an involuntary install, so debt recovery stays with trained staff.
- Give financial, legal or medical advice, including debt options, eligibility for schemes such as LIHEAP, or whether it is safe to be without power with a medical condition.
- Agree a payment plan or decide a complaint.
What triggers a handoff
- Keywords such as gas, smell, sparking, no power, oxygen, can't afford, warrant, complaint, ombudsman or deceased.
- Customer tier: accounts on your Priority Services Register, prepayment and business accounts.
- Two negative messages in a row, or anything Agent cannot verify.
Ofgem's contact guidance suggests prioritizing contacts linked to register accounts or containing words like "emergency", which is what keyword and tier rules do.
Complaints start a clock
A GB energy supplier has 8 weeks to resolve a complaint before the customer can go to the Energy Ombudsman (Ofgem). Water companies in England make fixed payments under the Guaranteed Standards Scheme when a written complaint, an account query or a request to change payment arrangements misses its deadline. In the US, state commissions such as the CPUC take complaints a utility could not resolve. Agent logs the complaint and its date the moment dissatisfaction appears.
Transparency and data
Ofgem's ethical AI guidance asks energy companies to make people aware when they are interacting with AI such as a chatbot, to let them challenge its output and to keep access to redress. Agent introduces itself as AI, verifies identity before showing account data and acts only through revocable connector permissions. Give it the vulnerability flag, not health details. Aftersales offers a GDPR data processing agreement and hosts all data in the US. It does not make you compliant; it gives your compliance team procedures, escalation rules and Monitors results to review.
Which channels matter most for utility customer service?
Phone remains essential for emergencies and payment difficulty, while app and web chat take most readings, smart meter questions and moving dates.
- Phone. Ofgem expects customers who are struggling to pay to reach their supplier without extra cost, with a freephone number as one way to do it, and a 24-hour service for anyone off supply because of a meter fault, per its contact guidance. Growth and Scale add warm and cold transfers, recording consent and per-brand numbers.
- App and web chat. Customers already send readings and meter photos here, so this is where Agent resolves the most.
- Email. Long billing disputes, landlord and letting agent requests, and executors settling a deceased customer's account.
- SMS. Reading reminders and outage updates. In the US, FCC rules require consent before autodialed texts to a mobile number, with an exception for emergencies involving danger to life or safety, so keep safety alerts and routine reminders apart.
- WhatsApp and social DMs. Storm nights and price changes bring public questions. Agent replies in 31 languages from the same Knowledge your team uses.
Ofgem's opening-hours analysis left out channels with no human representatives behind them, such as a fully virtual assistant, so count Agent's overnight cover as extra help, not staffed hours.
Which utility systems should the AI agent connect to?
Connect billing first, then metering and outages. Read access to billing plus write access for meter readings covers most of the conversation table above. Permissions are set per connector, so the billing connection can submit readings without being able to change a payment plan.
| System | What Agent reads or does | How it connects |
|---|---|---|
| Billing and customer information system | Account, balance, bills, tariff, payment schedule; submits readings and requests re-bills | Aftersales API or webhooks |
| Meter data and smart meter head-end | Last communicated reading, communication status, prepayment credit and top-ups | Aftersales API or webhooks |
| Outage management system | Open incidents by postcode or ZIP code, restoration estimates | Webhooks that push incidents as they open and close |
| Field service and work orders | Appointment slots for installs, meter exchanges and leak visits | Aftersales API or webhooks |
| Vulnerability records | The flag and needs codes that drive escalation rules | Aftersales API or webhooks |
| Salesforce, Zendesk or Freshdesk | Existing cases; Agent can also run inside them | Named integrations |
| Slack and Jira | Side conversations with metering or billing operations; a Jira issue when one billing fault hits many accounts | Named integrations |
Billing platforms differ widely between suppliers, so confirm specific connectors on a call. Put tariff explanations, price change letters, outage FAQs and the emergency script in Knowledge so Agent, your team and the help center give the same answer.
What does AI customer service cost for an energy or utility team?
A utility service team of 15 handling 12,000 conversations a month would pay about $6,915 a month at list prices, roughly $0.58 per conversation, on these assumptions:
- 15 service reps on Growth, billed annually.
- Agent resolves 45%, well below the 71% median Aftersales customers reach after 90 days, because payment support, complaints and safety contacts always go to people.
- Copilot for every rep, Insights Pro for a custom complaint-handling scorecard, and 8 billing, metering and complaints specialists as free Lite seats who comment and approve.
| Line item | Calculation | Monthly cost |
|---|---|---|
| 15 service seats on Growth | 15 × $69 | $1,035 |
| Agent resolutions | 5,400 × $0.99 | $5,346 |
| Copilot for 15 reps | 15 × $29 | $435 |
| Insights Pro | Flat add-on | $99 |
| 8 specialists as Lite seats | Included on Growth | $0 |
| Total per month | $6,915 |
Resolution costs follow winter and price-change peaks while seats stay flat; conversations Agent cannot help with are free. Utilities needing SSO or SCIM would choose Scale at $109 per seat. Teams staying in Salesforce Service Cloud, Zendesk or Freshdesk can run Agent on its own at $0.99 per resolution, with a minimum monthly commitment and no seats. See pricing.
How do you roll out an AI agent at a utility in 30 days?
Start with readings and moves, prove the safety paths, then add billing.
- Week 1. Tag three months of conversations by intent and choose three types, usually meter readings, moving home and balance questions. Connect read access to billing, and agree the handoff list with your safety, vulnerability and complaints leads.
- Week 2. Write procedures or import macros. Replay past conversations, including every gas, off-supply and bereavement case you can find, and confirm each one hands off on the first message.
- Week 3. Go live on app and web chat for readings and moves, evenings and weekends first. Turn on write access for readings and run Monitors on every conversation; Helio Energy saw quality scores climb 14 points once Monitors replaced sampling.
- Week 4. Add estimated-bill explanations with re-bills, smart meter status and outage updates. Report resolution rate, handoff reasons and complaint logging to compliance, using the definitions in our resolution rate guide, and plan for the next price change or winter peak.