PrymeIntelligence

Training teams for AI agents·Playbook

How to run an AI agent readiness assessment in a utility

Run an AI agent readiness assessment in a utility: test billing and metering teams on meter reads, vulnerable customers and depth, then pick the agent.

Assess the billing and metering team before an agent touches customer accounts. The test, set in energy and water, covers meter reads of different provenance, vulnerable customers, held actions that are hard to undo, and the record. The report shows the gaps by concept and decides whether the first agent explains bills, validates reads or drafts payment plans.

What is the assessment for in a utility?

Utilities bill millions of households and businesses, mostly from meter readings. Agents can validate readings, explain why a bill is estimated or higher than usual, set up payment plans within policy, and draft letters about arrears. A few actions in this business are hard to undo and can cause real harm, such as switching a meter to prepayment or starting a disconnection process. Those belong with people, under strict rules.

The assessment checks whether the billing and metering staff who will review an agent's held actions understand what is in front of them. The head of billing operations usually owns it, with the lead for customers in vulnerable circumstances. For the concept, see what a readiness assessment is.

Who should take it?

One team of up to 25:

  • Billing advisers who answer bill queries and approve adjustments.
  • Metering analysts who validate and correct readings.
  • Collections and payment plan staff, if the first agent touches arrears.
  • The vulnerable customers lead or a specialist from that team.

Field and network operations teams are separate; their agents, if any, schedule work and read sensor data. Advice on order is in who in a utility needs AI agent training first.

The steps

  1. Person (head of billing operations): chooses the team and lists actions the agent will never take alone: prepayment switches, disconnection steps, write-offs above a set amount, and changes to a priority services record.
  2. Person (head of billing operations): asks for the test set in utilities.
  3. Person (each member): sits the test alone.
  4. Person (head of billing operations and vulnerable customers lead): read the report by concept.
  5. Person (head of billing operations): chooses the first agent and its depth. A common start is explaining estimated and high bills at depth 3 from approved answers, with adjustments drafted at depth 2.
  6. Person (team leaders): train the weakest concept.
  7. Agent, depth 1: reads a month of past bill queries in shadow, proposing explanations and adjustments for comparison with what advisers did.
  8. Person (cohort): reassesses before customer-facing replies.
Step The agent does Depth Stays with a person
Choose team Nothing yet None Never-do list
Sit the test Nothing yet None Individual answers
Read the report Nothing yet None Gaps, with vulnerability lead
First agent Named as candidate None Agent and depth
Train the gap Runs in practice labs Practice Learning
Shadow Proposes explanations on past queries 1 Comparing with real replies
Reassess Nothing new 1 then 2 or 3 Go-live

What does the test usually find in utilities?

Meter reads of different provenance. A reading sent automatically by a smart meter is close to sealed. A reading taken by a meter reader is signed by a named party. A reading the customer typed into an app is declared. A photo of the meter sent by email is external. Metering analysts know the difference instinctively; billing advisers often do not. The test shows whether staff understand why the agent weights them differently and holds a bill built on a declared read that jumps sharply.

Vulnerability signals. A customer disputing a high bill mentions they rely on powered medical equipment. Staff know this matters when told. In the test, many answer as though the bill dispute is the whole case. The agent's bounds should hold such contacts for a person, and the reviewer must then involve the vulnerable customers team.

Hard-to-undo actions. Staff sometimes expect that a trusted agent could start a prepayment switch. The test makes clear this stays with a person and why.

What does the report change?

It picks the first agent. A team weak on vulnerability signals should start with bill explanations, not payment plans or arrears letters.

It sets bounds. The never-do list and the vulnerability triggers become part of the agent's bounds, so those contacts are held whatever the topic.

It identifies training for billing advisers on provenance before any agent validates reads.

What evidence should the utility keep?

The never-do list.

The report by concept.

The first-agent decision and its depth.

The reassessment before live replies.

What should the head of billing operations measure?

  • Scores by concept, comparing billing and metering staff.
  • During shadow, bills the agent would have held on a declared read that advisers let through.
  • After go-live, time from a held vulnerability contact to the specialist team picking it up.

Where does it go wrong in utilities?

Starting with arrears. Arrears letters carry the most risk for vulnerable customers.

Leaving metering out. Metering analysts understand provenance best and are the people who can teach it to billing.

Assessing just before winter. Volumes and vulnerability peak in cold months. Assess and train in the quieter season.

Utilities in the EU may also fall under NIS2 as essential entities, which affects cybersecurity risk management and incident reporting for the systems agents connect to. This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel. Banks and fintechs face similar questions about money and evidence; compare the bank version and the fintech version. How the scoring works is in what a readiness assessment for AI agents measures, and the topic guide is training teams for AI agents.

Questions people ask

Should field staff take the assessment?

Not at first. If agents start scheduling field work or sending instructions to engineers, assess the people who approve those instructions.

Can the agent set up a payment plan?

Within the utility's own limits and policy, at depth 3, once it is certified to that depth. Plans for customers flagged as vulnerable should be held for a person.

How often should a utility reassess?

Before each new capability, before any rise in depth, and ahead of each winter.

Train the people who run your agents.

Pryme Intelligence Academy: free courses on the real product, audio lessons, labs in a free workspace, and certificates anyone can check on VerifyDoc.ai. Teams get seats, an admin view and a free readiness assessment.