The bill arrives at the start of the month, and the figure that stings is rarely the kilowatt-hours you burned. It is the capacity charge. For one fifteen-minute window in the past four weeks, your site pulled harder than at any other moment, and your tariff prices the whole period off that single peak. The compressors, the furnace, a shift change, and the EV chargers all leaned in at once. It may never line up that way again, yet you pay for it as if it does, every month.
The cost hides in one interval
Across Europe, industrial and commercial sites buy electricity in two parts: the energy they consume, and a capacity or demand charge set by their highest interval of power draw. That peak charge is frequently the larger half of the bill, on many sites between thirty and seventy percent of it. Markets name it differently, contracted power, capacity charge, peak demand, but the mechanic is identical, and most European network tariffs add a second incentive on top: keep your peak out of the system's high-load periods and you pay less to use the grid as well. The peak you avoid pays you twice. Knowing that was never the problem. Acting on it in the fifteen minutes that decide it, every time, and being able to prove it afterward, is.
Forecast the peak, shave it
Today the job falls to a person and a screen. An operator watches the load climb and trips something when it nears the line, which means the decision comes late, under pressure, after the curve has already bent, and the costly interval has often passed before anyone reacts. The data that would let you see it coming, your own meter, SCADA, and process telemetry, is the data you least want to ship to a cloud you do not control. So the largest controllable line on the bill gets run by reflex.
Helios moves that decision earlier. It learns your load and your prices from your own data, on your own infrastructure, and forecasts where the expensive peak will land hours before it forms. The prediction comes from a forecasting model in the lineage of our LST-E model work, and a swarm of agents on Lascaris turns it into a concrete plan: move a batch process twenty minutes, discharge the battery into the spike, bring a generator or CHP unit up for the window, throttle a deferrable analytics job, and leave the loads that cannot move alone. The plan runs, and the peak that would have set your bill never forms.
How Helios EMS works
Helios connects to your smart meter management, SCADA, and process systems where they already run, so nothing migrates. It forecasts load and price by interval inside your own boundary, and only the decision, never the raw data, leaves it. The agents on Lascaris work out the cheapest combination of load shifts, storage, and on-site generation to hold the peak under target, execute it, and write every move to an audit trail. You start on a single site and extend across the estate without re-architecting anything.
Why this matters
The capacity charge is the single biggest lever on industrial energy cost, and shaving the peak pulls it directly. Lower network charges follow, because the same restraint that caps your demand charge keeps you out of the grid's costly windows. Compute stops being the problem and becomes part of the answer: a training run is one of the most movable loads you own, so the rack that used to spike your bill becomes one of the things holding it flat. And because flexibility programs, capacity-charge disputes, and reporting under NIS2 and CER all expect evidence, the audit trail Helios keeps is the proof, not an afterthought.
Why Scalytics
We did not learn this from a slide deck. Our founding team led digital transformation at E.ON, building IoT and grid systems across real European energy assets, and co-created Apache Wayang, the open federated engine the platform runs on. The forecasting, the federated design, and the discipline of keeping the evidence all come from operating real energy systems.
Everything runs on a core that is open and Apache 2.0, on-premise or air-gapped, with your energy data never leaving the site. For grid and OT security under NIS2 and CER, see Federated Cybersecurity Analytics.
Sources
- Demand and capacity charges as a share of industrial and commercial electricity bills (NREL)
- Peak shaving and demand-side flexibility for commercial and industrial sites LST-E, federated forecasting model: https://github.com/scalytics/LSTM-NNW/tree/main/LST-E
