Forecast curves for load and feed-in in a distribution grid, symbol for AI in grid operations
SERVICE · PROTOTYPE AND AI

AI for utilities: from pilot to regular operation, with prototype and governance

Working code in days, AI in grid operations and customer service with a governance that stands up to the regulator.

79 percent of German energy utilities have introduced AI, 61 percent are stuck in the pilot stage, none calls itself advanced. The bottleneck is not the model but data, process and governance. innobu builds the prototype that proves it holds, and the framework that lets it into regular operation.

Summary

AI for utilities at innobu means: a working prototype in days instead of a feasibility study in months, built on your data and in your process, plus the governance that allows the path into regular operation. Typical cases are load forecasting and grid state, clarification automation in market communication, customer service and document processes. The EU AI Act classifies AI in grid operations as high-risk from December 2027, so governance is part of it from the start.

Who this service is for

Head of grid operations and planning

You want to improve grid state, load forecasting or congestion analysis with AI and need proof before you set up a programme.

Head of market communication and billing

Clarification cases, master data errors and documents eat capacity, and a pilot with a vendor changed nothing.

Head of IT and digitalisation

Business units experiment with AI tools, and nobody has the inventory, classification and rules the EU AI Act requires from 2026.

International AI vendors

You want to sell into German utilities and need to know what the EU AI Act, KRITIS and the market roles demand of your product.

What you get

1. Prototype in days

A working prototype on your data and in your process, built with AI-assisted development, so you see whether the idea holds before budgets flow.

2. Use cases with a business case

Load forecasting, grid state, clarification automation, customer service, document processes: assessed by benefit, data situation, risk and effort.

3. Path into regular operation

Data connection, operating model, responsibilities, monitoring and handover to IT or vendors, so the prototype does not stay a pilot.

4. AI governance under the EU AI Act

Inventory, classification, risk class, documentation and approval paths that stand up to the Federal Network Agency as market surveillance authority. The framework is published on innobu.com.

5. Sovereign tool choice

Cloud or own infrastructure, European or American models: assessed by data protection, KRITIS requirements and cost, without vendor lock-in.

How it works

Four stages, a tangible result after each.

  1. Select cases

    We review your candidates, assess benefit, data situation and risk and pick one to three cases with the biggest lever.

  2. Build the prototype

    In a few days a working prototype on real data emerges that business and IT check together.

  3. Set up governance

    In parallel we set up inventory, classification and approval paths so the case passes the EU AI Act.

  4. Transfer into operation

    We plan data connection, operating model and monitoring and hand over to your IT or your vendor.

Backed by

Reference: municipal utility in northern Germany. Since 2025 innobu has led the grid division of a municipal utility in northern Germany as external portfolio and programme lead: grid portfolio 2025 to 2030, the §14a control programme with five sub-projects, the smart meter rollout across electricity, district heating and water, the customer and grid connection portal up to fibre, plus programme office, roadmap and decision paths. Mandate details on request, anonymised for confidentiality.

Related topic fields and tools

Next step

One AI case in mind, or twenty on a slide? In a first conversation we sort by benefit and data situation and say which case is feasible as a prototype in days.

Request a first conversation

Frequently asked questions

Why do AI projects at utilities stay in the pilot stage? +

The PwC study 2026 shows: 79 percent have introduced AI, 61 percent are stuck in the pilot. The reasons are rarely the model but missing data connection, no operating model, no responsibilities and no governance. A prototype on real data with a path into operation addresses exactly these points.

What is a prototype in days? +

A working piece of software on your data and in your process, built with AI-assisted development, that proves or disproves an idea before a programme is set up. Not a click dummy and not a feasibility study.

Which AI cases pay off at utilities? +

Most often: load and feed-in forecasting in the distribution grid, grid state estimation, clarification automation in market communication, customer service with a knowledge base and document processes in grid connection and billing. Which case comes first depends on your data.

What does the EU AI Act require from grid operators? +

Since August 2026 classification and transparency obligations apply, from December 2027 the high-risk requirements under Annex III, which cover AI in critical infrastructure and grid operations. The Federal Network Agency becomes market surveillance authority. Inventory, classification, documentation and approval paths must be in place by then.

Cloud or own infrastructure? +

It depends on data, KRITIS requirements and cost. innobu assesses both paths and European as well as American models without vendor lock-in. For many cases in grid operations a sovereign variant is the better choice.

How does innobu itself work with AI? +

With role-based AI orchestration in development; the human remains the decision maker. The method is described as Agentic Harness Engineering on innobu.com, with the reliability questions that also apply to your cases.

Further reading