How Premier Energy cut its forecasting deviation by a third with an explainable model
Addvery turned Premier Energy's power forecasting into an explainable model, validated it against real hourly market data, and automated the daily workflow. Forecasting deviation fell by about a third.
Premier Energy Group is a vertically integrated energy company with operations in Romania, Moldova and Hungary, serving over 2.2 million customers across Romania and Moldova. In Romania, Premier Energy's electricity supply business is the country's fifth-largest; in Moldova, the Group is the leading energy supplier. Across both markets, Premier Energy combines large-scale supply operations with an established in-house forecasting and analytics function.
Addvery turned Premier Energy's power forecasting into an explainable model, validated it against real hourly market data, and automated the daily workflow. Forecasting deviation fell by about a third.
The hidden cost of a forecast you can't question
Every company that commits to the market before it delivers takes on a risk: the gap between what it promised and what actually happened. In electricity, that gap is settled in cash, every hour of every day. Romanian suppliers have to tell the market operator a day in advance how much power will move through their meters, and when the actual volume differs from that forecast, the imbalance is charged back to them, sometimes at prices that turn a small forecasting miss into a serious loss.
Premier Energy was already running its own forecasting models, built and maintained by a strong in-house quantitative team. This wasn't a company starting from zero. What the existing approach couldn't easily do was explain itself: when a forecast missed, the model could produce a number but not a reason, and that made the gap hard to close.
- The daily forecast leaned on manual work: several spreadsheets loaded by hand every morning.
- When a forecast missed, the team had no automated way to trace what drove the deviation.
- Weather-driven renewable output and Romania's mountainous interior made a share of the deviation genuinely hard to predict.
A forecast built to answer questions
Grounding the forecast in the weather
Addvery developed the forecast into a model that brought physics, statistics and meteorology together, rather than a single black-box engine. Romania's geography made this the hard part: mountains running through the middle of the country distort weather prediction. On top of that sat the messier reality of the grid, with seasonal losses, unmetered draw, and consumption that can shift by tens of gigawatts when industrial customers change how they operate. Reading those behavioural patterns mattered as much as the weather.
Testing every assumption behind the model
Rather than relying on a single off-the-shelf network, Addvery's team and Premier Energy's own analysts sat with the data together and refined the model through repeated iteration, testing which inputs mattered and how far back each model should learn from. Along the way they found what tends to cause deviation in the first place: gaps in process, weak spots in data quality, and sometimes the design of the model itself. Progress was measured against reality: the forecast was replayed against actual hourly settlement prices, so every improvement had to show up in the numbers.
From five spreadsheets to one automated routine
The old routine depended on five spreadsheets loaded by hand each morning. Addvery replaced that with a pipeline that runs on its own, removing the risk that comes with any manual process handled under daily time pressure.
A model the team could keep tuning
Because the model can be explained, Premier Energy's own analysts can audit it, see why it produced a given number, and recalibrate as the weather, the data and the market move.
A third less deviation, and a model that keeps improving
Key benefits
- Forecasting deviation down by roughly a third.
- A payback period of roughly two months on the project investment.
- An explainable model the in-house team can audit and keep tuning.
- With the model explainable and the pipeline automated, Premier Energy's own team now owns the improvements going forward.
Facing a similar problem?
Tell us where your numbers stop being trustworthy. We'll tell you whether it's worth a project.
.png){kind=logo}
