Utility Digitalisation with AI Agents

Utility digitalisation, in operational terms, means the recurring cases between customer contact and the billing system run without manual handover steps, while the billing system stays the system of record. We build AI agents and automations that take in meter readings, move-in and move-out notices, feed-in applications and service requests, check them, match them to the right contract account and hand your team a case that is ready to decide. Data is written back only where your systems allow it.

Where utilities actually lose time

Not in the customer portal. Most utilities run one by now. The load sits next to it: meter readings arriving as a phone photo of the reading card, cancellations written as free text, move notices without a metering point reference, feed-in applications with half the plant documentation attached. Every one of those cases starts with the same search. Who is this, which metering point, which contract, which stage of the supplier switch.

And the load arrives in waves. The annual billing run predictably triggers questions about instalments and back payments. A price adjustment letter does the same, compressed into a few days. The smart meter rollout adds appointment questions and opt-out declarations. The pattern is known, yet you cannot double the service desk for a single week. That gap is what we automate:

• Intake: emails, forms and PDF attachments get classified, matched to metering point and contract account, duplicates flagged.
• Completeness: if a feed-in application is missing the plant rating or a certificate, the request for the missing document goes out and is logged.
• Answers: status, instalment amount, appointment. Read access to approved sources only, handover to an agent as soon as a change is involved.
• Escalation: outage reports and hardship cases go straight to the people responsible for them.

The assistant answers, it does not decide

An AI assistant in utility customer service is not a sales channel. It takes in the request, matches it, answers standard questions from approved material and hands over. Tariff commitments, payment plans, disconnections and anything with legal consequence stay with your team, because a wrong statement about supply obligations costs more than any minute saved.

From 2 August 2026, Article 50 of the EU AI Act requires that users can tell they are talking to an AI. We build that in as the default, not as a later checkbox: disclosure in the first reply, handover to a human at any point, a full transcript of the conversation.

Our own infrastructure, because yours is critical infrastructure

A company operating grids or supplying households is measured against different standards than an online shop. Our automations therefore run on self-hosted n8n with PostgreSQL and NocoDB, in Europe, on infrastructure your organisation or we control, not in a US cloud. Where a language model is involved, we tell you which model does the work, where it runs and what data leaves the system at all. For many cases the case text is enough, without customer master data.

About your billing system we ask one question: which interface does it expose. SAP IS-U, Schleupen, kVASy, an in-house build, plus market communication on top. If there is no write interface, the automation writes nothing and instead presents the checked case for approval. That sounds less like full automation and lasts longer in production. The vendor-neutral part, agent architecture, model selection and operations, is documented on AI agent integration. This page covers the utility case.

Getting started

One case type, one wave, measured against handling time in the service centre. Usually that is the inbound channel before the annual billing run, or feed-in customer onboarding. The first call is free and ends with a straight answer on whether the first process is worth doing at all: contact.