Plan dynamically, save energy
A bakery, a data center, a CNC manufacturing plant and a sports park have little in common at first glance. Look again, and they share one thing: all four consume electricity in patterns that are hard to predict. This is exactly where Detecon’s challenge at Energy Data Hackdays came in.
The forecast looks back, the business looks ahead
We see it every day in our consulting practice: energy-intensive companies are caught between volatile markets and consumption planning that keeps getting more complex. Prices fluctuate. New assets come online. Production schedules change. And the forecast meant to absorb all of this is based on a historical baseline.
That cannot work. A baseline knows nothing about next week’s change in the weather. It doesn’t know about the new PV system on the roof or the charging stations in the parking lot. It doesn’t notice that the second shift will be dropped from October. It is planning for a company that no longer exists in that form.
The consequences hit the bottom line directly. Too much or too little electricity purchased. Budgets that are already off by the second quarter. ESG reports based on estimates rather than measured values. This gap runs through entire industries. We set out to close it.
One hackathon, one startup, 36 hours
Every year, the Energy Data Hackdays in Brugg-Windisch bring data scientists, energy experts and companies to the same table. Detecon took part as a challenge owner. At our side: TEELS, an energy efficiency startup with a special backstory. It was born at this very hackathon the year before.
We deliberately set an ambitious task: Build a universal electricity consumption forecasting engine. Not for one customer, not for one site. For all of them.
The raw material came from the TEELS platform: 15-minute load profile data from real-world operations. Data centers with their constant background hum. CNC manufacturers with sharp load peaks. Bakeries that work through the night. Sports parks that live by the weather and the weekend. It hardly gets more heterogeneous than that. And that was exactly the point.
What really makes forecasts better
Then 36 intense hours began. The team dove into the data. And into what makes the difference in data science: feature engineering. Which factors actually improve the forecast? Weather data, calendar events, public holidays, operational specifics.
The real hurdle, however, was generalizability. A model that works perfectly for the bakery but fails in the data center is a one-off, not a tool. The approaches that emerged showed a different way. First, companies are clustered by their consumption structure. Then a dedicated, optimized forecasting model is built for each cluster. Stage one selects the model, stage two runs the numbers. Both stages can be optimized with AI and machine learning. But without this two-stage design, there is no generalizable tool.
From forecast to decision
An accurate forecast is not an end in itself. It only creates value when someone decides differently because of it. Building exactly this bridge from data science to day-to-day business was the real goal of the challenge. And it pays off in several places at once.
Procurement comes first. Companies that forecast their future demand reliably reduce volume risks and gain planning certainty on electricity costs. In finance, driver-based forecasts replace the static baseline. The budget reflects what is really happening in the company. Not the situation from three years ago.
In sustainability, high-quality planning data becomes the foundation for auditable ESG reports. Emissions can be measured instead of estimated. In energy efficiency, reliable, automated planning shows what individual measures contribute to future electricity consumption. Variance analyses reveal whether they are working.
What remains after 36 hours
Intelligent load analytics and forecasting are becoming standard tools of modern, data-driven energy management. The Energy Data Hackdays showed how much potential lies at the intersection of data science, domain expertise and business analytics. And how quickly an idea becomes a working approach when the right people work together.
“We were born at the Energy Data Hackdays ourselves a year ago. Now our tool and our data are coming back and becoming the foundation for the next idea. For our customers, one thing matters in the end: load profile data has to translate into decisions and actions. The challenge proved that this is possible,” says Roman Marty, CEO of TEELS AG.
Detecon expects demand for forecasting capabilities like these to rise sharply. A universal electricity consumption forecasting engine enables energy-intensive companies to anticipate their energy consumption while realizing both economic and sustainability benefits. The bakery, the data center, the manufacturing plant and the sports park show that it can be done. Static baselines are a thing of the past. The future belongs to dynamic forecasting.
Detecon's forecasting challenge.
Cut costs with dynamic energy forecasting
Those who can reliably forecast future electricity demand can procure power more strategically. Volume risks decrease because companies procure neither too much nor too little electricity. This is especially important in volatile markets. New facilities, changes to shift schedules, or sudden weather changes are directly factored into the forecast. This provides planning certainty regarding electricity costs, rather than making decisions based on an outdated baseline.
Realistic budgets
Static baselines are based on a company that no longer exists in that form. As a result, budgets are already off by the second quarter. Driver-based forecasts provide a solution. They link energy consumption to the factors that actually determine it, such as production, operating hours, or weather. The budget thus reflects what is actually happening in the company, rather than the situation from three years ago.
Auditable ESG reports
Many ESG reports today are still based on estimates. This is increasingly insufficient for auditable sustainability reports. High-quality planning data based on 15-minute load profile data provides a robust foundation. Emissions can be measured rather than estimated. This gives companies greater transparency with auditors, investors, and customers and enables them to provide verifiable evidence of their sustainability goals.
Making measures quantifiable
Whether it’s a new solar power system, more efficient machinery, or adjusted operating hours: every efficiency measure should pay off. Reliable, automated planning shows how individual measures will impact future electricity consumption. Variance analyses reveal whether they are actually effective. This transforms energy efficiency from an assumption into a measurable metric, allowing for targeted management of investments.
