How to prepare your energy data for AI before the tool is approved

There’s one in every organisation. Forty tabs open, a spreadsheet of their own that nobody else quite understands, and a habit of testing new tools on last month’s figures before anyone has asked for it. On a random Tuesday, you’ll find them standing by someone’s desk with a laptop in their arms: “look at what I’ve built in Copilot”.

…but then comes the question of where the data ends up if it goes into an AI tool. And after that, they either get a straight no, or a list of questions nobody has time to answer.

“I often meet property owners who really want to. They see the value straight away, and they have usually already tried something small on their own. Then the internal directive arrives that no information at all may be shared with the AI tool, because nobody knows what might be sensitive. So the tool sits there, but it’s not much use. Nobody wants to go against internal directives, so you end up with your hands tied instead”, says Joel Torkelsson, CCO and co-founder of Metry.

When there is little to go on, many choose the safe option

The AI Act became applicable in a broader sense on 2 August 2026. The heaviest requirements on high-risk systems were at the same time pushed back to 2 December 2027 by the amending regulation known as the Digital Omnibus, (EU) 2026/1744, which entered into force on 27 July 2026. The transparency requirements, by contrast, have applied since 2 August 2026 as planned, and the next change comes as soon as 2 December this year, when two new prohibitions take effect.

So the rules have moved several times in a short period, but none of those moves has fully settled, and many companies still lack clarity. The national picture adds to that. Member states were required to name their AI Act supervisory authorities by August 2025, and more than a year past that deadline several still have not.

When IT has to assess the AI tool (who runs it, what the agreement says about storage and model training, who can access the conversations afterwards) there is still no established list of questions to work through, so two companies with the same conditions can arrive at different answers. And when there is little to go on, most people choose the safe option.

One thing is often overlooked here. Most of your data is probably already sitting with a cloud provider. Where the AI tool is part of the same suite, email, documents and spreadsheets are often in the same cloud that the AI tool runs in, and for the business versions broadly the same agreement and the same boundaries apply to both.

The question is therefore rarely whether the data leaves the building, but under which agreement it is handled and who can reach it.

“People are landing in different places on this right now, and there is no established best practice to lean on yet. I recognise it from the discussions about cloud services ten years ago”, says Joel Torkelsson.

“Back then a lot of people wanted everything on their own server down in the basement, and it took time before customers accepted storing certain data in the cloud.”

But that doesn’t mean the work has to stand still.

Get ready for AI while you wait for an answer

The effort you need to put into the data layer looks the same whichever tool is approved in the end. It also has value if the answer turns out to be that no tool is approved at all, because the same structure is used for energy management, tenant billing and sustainability reporting.

To be able to work with your energy data in an AI tool, here is what to look at:

1. Take stock of your meters. Which meters exist across the portfolio, which ones deliver data today, and which ones do you only have an invoice for? Make sure the list is complete, so that you have coverage for what you want to be able to analyse.
2. Decide on the structure while you still remember why it looks the way it does. Which meter belongs to which building, what is common-area electricity and what is the tenant’s, who is responsible for what. That knowledge often sits with a handful of people and needs to be documented before an AI can use it.
3. Find the history. If the grid company has kept five years of consumption, then five years of history can be retrieved (with Metry, at least!). By collecting as much historical data as possible you can make better comparisons backwards and better forecasts forwards.
4. Make sure you own your own data. Export in an open format and access through a documented API are what make it possible for the data to reach whichever environment ends up being approved.

Metry gives you control over your energy data collection

In Metry you structure your meters in a tree by location, property and meter, and add tags, areas of responsibility and metadata. Every value is logged with its source and timestamp, so you can show where it came from and when it was retrieved.

That structure and quality assurance looks the same whether the data goes on to the energy management system, to the sustainability report or to an AI tool that gets approved six months from now.

Are you waiting for a decision on an AI tool? Get in touch and we will go through what your collection can do today and what needs to be in place by the time the decision arrives.

…but if your company has already approved an AI-tool internally, there’s now an easier way to access your energy data in it – Metry MCP server! Click to learn more.