How can utilities prove staff are trained to oversee AI agents?

Utilities should build their proof around vulnerable customers, where regulators look first. For every agent-prepared bill correction, payment plan, refund or disconnection step, a utility should be able to show the person who approved it, their current certificate checked at source, and the vulnerability information they had in front of them. The same record then serves auditors, ombudsmen and complaints.

Why utilities need careful evidence

Energy and water companies supply essential services. Billing errors, aggressive debt recovery and wrongful disconnections fall hardest on people least able to absorb them. Regulators and complaint schemes pay close attention to how suppliers treat vulnerable customers.

Agents in utilities typically handle meter reading queries, bill explanations and corrections, payment plans, refunds of credit balances, debt letters and moves in and out of properties. Most of this is routine. The exceptions, such as a back-bill after a meter fault or a customer in arrears who has disclosed a health condition, are where the evidence matters.

Utilities that are essential or important entities under NIS2 in the EU also carry cybersecurity risk management duties that extend to how their systems are operated.

What evidence should a utility keep?

• An action map. Each agent capability, its depth, the utility's own limits, and the role that approves held actions. Debt and disconnection actions should always be held for a person.
• A certificate register for approvers, checked at the issuer.
• Rights linked to certificates, so only certified staff approve back-bills, payment plans for flagged customers and any step towards disconnection.
• Vulnerability flags in the record, showing what the approver saw.
• The filed record of each verdict and approval.
• Briefings when tariffs, debt policy or regulatory rules change.

The AI agent training plan template for utilities sets out training per role.

What should utility staff be trained on?

1. Vulnerability. Recognising signs in a customer's messages that the agent may not have flagged, and knowing what support options exist.
2. Back-billing. When a corrected bill can be issued and how far back, under the utility's rules, and checking the agent applied them.
3. Affordability. An agent's payment plan may clear the debt on schedule and still be unaffordable. Approvers judge affordability, not arithmetic.
4. Meter evidence. A smart meter reading from the utility's own system is sealed. A customer's photo of a meter is declared. An estimated reading is neither. Approvers should know what each can support.

An example from an ombudsman complaint

A customer complains that her energy supplier set up a payment plan she could not afford after a back-bill. The record shows the agent drafted the plan at depth 2, based on the debt and the supplier's standard terms. The approving adviser was certified and current. The customer's earlier message, mentioning she had just lost her job, was in the contact history but was not flagged.

The adviser approved without reading the history. The ombudsman finds against the supplier. The supplier's own review uses the record to show the failure was one of practice, not training, and changes the agent's brief so that any mention of job loss or illness holds a payment plan for review with the message highlighted. It also briefs all advisers on the case.

Which controls make the evidence sound?

• All disconnection steps held for certified staff, without exception.
• Vulnerability information displayed at the point of approval.
• Sampling of approved payment plans for affordability.
• An append-only record that complaints teams and regulators can review.

Under the EU AI Act, deployers of high-risk systems must assign oversight to competent people; see Article 26 deployer duties in energy and utilities and Article 4 AI literacy in energy and utilities. For comparison, see how banks prove staff are trained and how wealth managers do it. The guide to AI literacy and certification covers the overall approach.

Where the Academy fits

Pryme Intelligence Academy's Operator track covers human review, reading the workspace and what to do when something goes wrong, with module tests set in the utility's own industry and role. Certificates are checkable at source, and a business can make approval rights depend on a current certificate, depth by depth.