Energy volatility belongs in the operating model, not a footnote

Electricity costs are not one number. Consumption is charged per unit, but a significant share of the bill often sits in network charges, capacity or demand charges, and taxes and levies. These behave differently from the energy component.

Demand charges are typically set by the peak power drawn in a period, not by total consumption. A site can reduce total consumption and still see costs rise if its peak demand is unchanged or higher. Conversely, shifting a load off a short peak can reduce cost without reducing consumption at all.

Load shape matters for a second reason: it determines exposure to time-of-use pricing. A process that can be shifted, batched or run outside peak periods has an option that a continuous process does not. Whether that option is worth exercising depends on the process — and for some, particularly those with high thermal mass or long start-up sequences, it is not.

The tax and levy component varies by jurisdiction and by the characteristics of the consumer, and it changes. Any model should carry it as an explicit line rather than folding it into an average unit rate, because it responds to different drivers than the energy component does.

The useful output of an energy sensitivity is not a single number but a set of thresholds. At what price does a particular process become uneconomic to run? At what point does it become worth invoking an alternative arrangement? What is the maximum duration of an interruption the site can absorb?

Thresholds agreed in advance are operationally useful because they remove the need to make decisions under pressure. A site that has decided in March that it will switch to an alternative operating pattern above a stated price can act immediately. A site that has not will convene a meeting, and act later and less well.

For many processes, the cost of an unplanned stop exceeds the energy cost that triggered it. Restarting a furnace, re-establishing a thermal equilibrium, clearing a line or disposing of work in progress all carry cost, and some carry scrap.

These costs should be modelled explicitly rather than treated as an operational nuisance. Once quantified, they frequently change the answer: a site that appears to have a strong incentive to curtail consumption during a price spike may find that the restart cost exceeds the saving.

The same applies to start-up energy itself. Some processes draw heavily during start-up and settle at a much lower steady-state consumption. For those, increasing the number of starts to chase lower off-peak prices can increase total consumption even while the average unit price paid falls — a result that surprises sites which model price but not start-up behaviour.

It is also worth distinguishing between a planned and an unplanned stop. A planned shutdown can be scheduled into a low-price window and prepared for. An unplanned one occurs at whatever moment the supply fails, and the recovery cost is usually higher because the process was not in a state prepared for interruption. If interruption is a realistic scenario, the question is whether the site has a defined, rehearsed response and what that response costs to hold ready.