The Data Exam Behind Every Megawatt

When I talk with people from the energy sector about data and AI, the conversation usually starts at the glamorous end: algorithms that forecast price spreads on the spot market, trading engines that decide in real time where a megawatt earns the most, machine learning on top of everything. I understand the attraction, because that is where the money is visibly made. But my experience from years in this industry tells me that the fate of such projects is decided much earlier and much lower in the stack, at a layer nobody puts on a conference slide, and I want to show this on a case that I find more instructive than any general argument about digital transformation.

One virtual power plant, three grid operators

Consider an operator of a virtual power plant in Central and Eastern Europe, aggregating distributed renewables, battery storage and flexible industrial consumers across Poland, Czechia and Hungary. The business idea is sound: pool the flexibility of many small assets and sell it to the markets as if it were one power plant, for example as frequency regulation services for the Polish grid operator PSE, which has joined the European PICASSO platform for balancing energy. On paper, this is a trading and aggregation business. In practice, before this company sells a single megawatt of balancing service, it has to pass certification with the grid operator, and it is worth reading carefully what that certification actually tests.

It does not test the elegance of the trading strategy. It tests whether the operator can receive telemetry from every battery and every solar farm with a sampling rate of 1 to 4 seconds, whether the data pipelines that carry measurements and control signals reach 99.99 percent availability, because from the grid operator’s point of view this is critical national infrastructure, and whether the company can speak the specific protocols and settlement file formats the operator requires. And since the assets sit in three countries, the same company must map the data standards of three different grid operators, PSE, ČEPS and MAVIR, each with its own identification codes, its own settlement periods and its own reporting structures, onto one coherent internal model. In my opinion, the honest name for this is not a trading project at all. It is a data engineering exam, administered by a regulator, with financial penalties for failing.

Where the exam is really passed or failed

Three requirements from this case have stayed with me, because each sounds technical and each is, in fact, about money.

The first is that when the operator sends its price bids and activation volumes to the grid operator, the pipeline can lose or duplicate nothing, because a lost packet can mean a contracted megawatt that was never delivered, and that means penalties. The second is what happens when data goes bad in the middle of the night: if a meter on a Czech solar farm starts reporting negative or empty values, the system has to catch it, quarantine it, and substitute a defensible estimate before a corrupted profile reaches the grid operator, without waking anyone up. The third is the audit: at any moment, the operator must be able to prove, second by second, which specific asset the energy used to balance the public grid came from. I have watched companies in logistics treat data lineage as a nice-to-have for years. In energy balancing it is a condition of staying in the market.

Only when this unglamorous layer holds does the attractive part begin to pay: the forecasting engines that predict spreads 24 and 48 hours ahead, and the decision, taken in a fraction of a second, whether it is more profitable to sell energy from storage on the Hungarian exchange or to activate reserve for PSE in Poland. There is also a quieter benefit that I think deserves more attention than it gets: one central data team running pipelines for three countries, instead of three separate national IT systems doing the same work three times.

Why I believe energy is ahead of logistics here

In my last article I wrote that logistics has a homework problem: the AI is capable, but the data underneath is fragmented and nobody is forced to fix it. What strikes me about the energy sector is that the forcing mechanism already exists. A freight forwarder with messy data loses margin quietly, over years, and can keep telling itself a comfortable story. A flexibility operator with messy data does not get certified, or gets fined, or loses the audit, and the regulator does not accept comfortable stories. This is why, in my opinion, energy companies are learning faster than most industries that the difference between an AI ambition and an AI business is a boring, reliable, provable layer of data plumbing.

If your company sits anywhere in this chain — generation, storage, aggregation, or energy-intensive industry — my advice is the same one I gave the forwarders, only with more urgency: before you budget for the intelligent layer, find out whether your data would survive an exam in which every second is recorded, and every gap costs money. That is a question with a number as an answer, and it is exactly the number we calculate at Cadvel.

Grzegorz Hinc spent his career in freight forwarding and logistics and in the green energy sector, and today he runs Cadvel, a company that designs, builds, and maintains data platforms for logistics and energy. If this sounds familiar, we will calculate what this problem costs your company.