Clear the billing queues. Keep people on the decisions.
Energy suppliers, municipal utilities and network operators run large back offices: meter-read and billing exceptions, moves and switches, connection requests and field work orders. We look for the workflows where agents can check and prepare that work against the meter data, the tariff and your rules, with controls that stop a wrong bill before it goes out, and people keep every decision about a customer's supply and every operational or safety call.
Six back-office workflows with high volume and strict rules.
These are the workflows we would examine first in an energy supplier, a municipal utility, a water company or a distribution network operator's customer and connections teams. They run through the billing system, meter data management, the CRM, market message systems, field service tools and shared inboxes. We work on the back office, not on grid or plant control.
| Workflow | Manual today | AI-native version | Stays human | Verification check |
|---|---|---|---|---|
| Meter-read and billing exceptions | Billing staff work a queue of implausible reads, missing reads and bills held by validation rules, one account at a time. | An agent checks each exception against read history, the meter record, the tariff and recent contacts, classifies the cause and proposes the fix. | Estimated bills above a threshold, back-billing, disputed reads and any account flagged as vulnerable. | Proposed consumption must sit inside a plausible range for that meter. Corrected bills are re-run through the billing system's own validation. |
| Moves, switches and market messages | Staff fix rejected switches and move-ins, chase missing meter readings and correct mismatched market messages by hand. | An agent matches each rejected message to the account, identifies the reason and prepares the correction. | Disputed supply start dates and cases needing contact with another supplier or operator. | Corrections must follow the message format and process rules. Supply dates must match the evidence on file. |
| Payment difficulty and collections | Collections staff send reminders, agree instalment plans and review accounts before any escalation. | An agent applies payments, prepares reminders by your rules and summarizes each account's history for the case handler. | Payment plans, hardship cases, vulnerable customers and every decision to restrict or disconnect supply. | Accounts with a vulnerability flag, a plan or an open complaint are blocked from automatic escalation. No agent can start a disconnection. |
| Grid connection requests | Connection teams check applications for solar, heat pumps, storage and charging points for missing documents and route them to engineers. | An agent checks completeness, extracts the technical data into a structured record and routes the application. | Technical assessment, capacity decisions and connection approval. | Required documents and data fields must be present before engineering review. Applicants are told exactly what is missing. |
| Field work orders and job reports | Field technicians send notes, photos and readings. The office retypes them into the work management and billing systems. | An agent turns job notes and photos into a structured completion record and flags follow-up work. | Safety findings, defects and every sign-off of work on live equipment. | Meter numbers and readings must match the asset record. Safety keywords always route to a supervisor. |
| Customer complaints | Complaints arrive across channels and are logged, classified and tracked against response deadlines by hand. | An agent logs and classifies each complaint, gathers the account history and drafts a response for a case handler. | The resolution, any compensation and replies to vulnerable customers. | Every complaint has an owner and a deadline timer. Replies make no commitments outside fixed templates. |
Six numbers that show whether it worked.
The Audit measures these before anything is built. The Pilot is measured against the same numbers.
| Metric | Definition |
|---|---|
| Minutes per billing exception | People time spent resolving a held or implausible bill, from queue entry to bill released. |
| Exception backlog age | Number of open billing and switching exceptions, and how many days the oldest have waited. |
| Bill correction rate | Bills reissued because they were wrong, per 1,000 bills sent. |
| Switch and move rejection rate | Market messages rejected or corrected, per 1,000 processes started. |
| Connection application cycle time | Days from application received to complete file ready for engineering review. |
| Complaints past deadline | Complaints not answered within your internal or required response time. |
What can go wrong in a utility back office, and the control for each.
Energy and water are essential services. Decisions that affect a customer's supply, and any operational or safety decision on the network, stay with people. The EU AI Act treats AI used as a safety component in critical infrastructure as high-risk, and we do not build agents for that role. These are the failures we design against first.
| Failure mode | Control | How we'd detect it |
|---|---|---|
| Wrong bill sent | Proposed corrections pass plausibility limits and the billing system's own validation. Large changes and back-billing need a person. | Bill correction and complaint rates tracked every week. |
| Vulnerable customer escalated automatically | A vulnerability flag blocks every automatic escalation step. Those accounts always go to a trained case handler. | Weekly check of escalation steps against the vulnerability register. |
| Supply restricted or disconnected in error | Agents cannot start a restriction or disconnection. Every such step needs a named person and a recorded reason. | Audit trail of every restriction request and who approved it. |
| Switch processed with the wrong date or meter | Supply dates and meter numbers are checked against the evidence and the asset record before a message is sent. | Rejection and correction rates per message type. |
| Safety finding lost in a job report | Safety keywords and defect categories always go to a supervisor. Agents never sign off work on equipment. | Sampled review of job reports against the defect log. |
| Customer or consumption data exposed | Agents see only the fields a task needs. Self-hosted, EU-hosted deployment is possible. | Access logs per agent and per task. |
Billing exceptions, from diagnostic to pilot.
Illustrative example — not a client result. The numbers below are assumptions chosen to show the method and the arithmetic. Your Audit replaces them with measured values.
| 01 | Diagnostic | Suppose a municipal utility picks "billing exceptions" on the Other / General Business track. The Snapshot might show a High leverage signal with Medium confidence, friction at the Processing and Decision stages, and Level 2 (workflow augmentation) as the indicated AI depth, because the billing system stays in charge of the calculation. |
| 02 | Audit | Suppose the Audit measures 8,000 billing exceptions a month at 4 minutes each. That is 8,000 × 4 = 32,000 minutes, or about 533.3 hours a month, about 3.3 FTE at 160 hours. It also finds that about 30% of exceptions involve disputed reads, back-billing, large estimates or vulnerable customers, and need a person. |
| 03 | Pilot | Suppose the 70% routine exceptions are resolved by the agent's proposed fix, with a person checking a 5% sample for 3 minutes each. The 30% exceptions that stay with people take 6 minutes each because they are the harder cases. The Pilot measures whether that holds. |
| Line | Assumption | Hours per month |
|---|---|---|
| Baseline | 8,000 exceptions × 4 min | 533.3 |
| Routine, after | 5,600 exceptions, 280 sampled × 3 min | 14.0 |
| Stay with people, after | 2,400 exceptions × 6 min | 240.0 |
| After, total | 14.0 + 240.0 | 254.0 |
| Difference | 533.3 − 254.0 (about 1.75 FTE) | 279.3 |
At an assumed loaded cost of €45 an hour, 279.3 hours is about €12,570 a month. The Audit would also measure the exception backlog and bill corrections, because a held bill delays cash and a wrong one creates a complaint. It would also count the new costs: compute, billing and meter data integration and exception review. If the numbers don't hold, the recommendation is "don't proceed".
Where our proof comes from.
We'll be plain about this. Our production proof is in AdTech, not energy or utilities. We have not yet run these utility workflows for a client.
The architecture is the same one that runs in AdTech production: typed intake, deterministic checks before any action touches money or customers, bounded retries, then a person. Scoring high-volume event data for anomalies in real time, as invalid-traffic detection does, is close to spotting implausible meter reads. The case studies show that work, and five AI plugins built on this architecture are live at reviveadservermod.com.
Our engineering team and delivery network have built related systems: predictive fault detection that pairs anomaly models with a drafted root-cause summary for engineers, and computer-vision anomaly detection for operational inspection. See Selected builds.
That is why utility engagements start with the free diagnostic and then the Audit. The Audit measures your own baseline and records what must never be automated before anything is designed, so any decision rests on your numbers, not on ours.
Questions utility teams ask first.
Will an agent ever decide to restrict or disconnect a customer's supply?
No. Agents cannot start a restriction or disconnection. They can prepare the account history for a case handler, and accounts flagged as vulnerable are always handled by a trained person.
Do you work on grid or plant operations?
No. We work on back-office and customer workflows. Operational control of networks and plants, and every safety sign-off, stay with your engineers and control systems.
Do we need to change our billing system?
No. We work with the systems you have, through their APIs or interfaces. The billing system keeps doing the calculation, and the Audit records where data is re-keyed today.
Find out which back-office workflow is worth redesigning.
Start with the free diagnostic on the Other / General Business track. Move to the Audit only if the Snapshot points to something worth measuring.
| 01 | Free AI Operational Diagnostic | 10 steps, 8–12 minutes, Other / General Business track. Snapshot on screen and as a PDF. |
| 02 | Operational Leverage Audit, from $1,500 | Fixed fee, one workflow, 10–15 working days after data access. Measured baseline, verification architecture, benefit model with stated assumptions, go / no-go. |
| 03 | Pilot implementation, from $6,000 | Fixed scope. The workflow built and run in your environment, with an audit trail, measured against the Audit baseline. Typically 4–8 weeks. |