A factory owner told me his solar array had cut consumption by 28% and his bill by 11%. He assumed something was wrong with the array. Nothing was wrong with the array. He was paying demand charges, and solar had not touched them.
Commercial and industrial tariffs generally bill two things: energy consumed (kWh) and peak demand (kW). The demand component charges you for the highest rate of consumption recorded in the billing period, usually measured over 15 or 30-minute intervals.
The logic is that the network has to be built to serve your maximum draw, whether you use it for one interval or continuously. Depending on the tariff and region, demand charges commonly represent 25-45% of a commercial electricity bill.
Solar reduces demand only if it is generating at the moment your peak occurs. Three common reasons it is not:
The peak happens outside daylight. Winter morning startup at 06:00, or an evening shift. Solar output is zero, so the peak is unaffected.
The peak happens on a cloudy day. Your annual peak is a single interval. It only takes one heavily overcast day with full production for the peak to be set with no solar contribution.
The peak is sharp and short. Starting a large motor or a furnace produces a brief, enormous draw. Even at full sun, solar is a fraction of that spike.
Some tariffs also apply a ratchet – your demand charge for the year is set by the highest peak in any month, sometimes carried forward for eleven months. Under a ratchet, one bad interval in February costs you all year.
A battery discharges on demand, regardless of sun. That is the entire point. When the site draws above a set threshold, the battery supplies the difference, and the meter never sees the peak.
For this to work, three things have to be right:
This is where systems succeed or fail, and it gets far less attention than the cells.
A naive controller watches demand and discharges when it crosses a threshold. That works for gradual ramps and fails for sudden steps, because by the time it responds the interval average is already elevated.
Better controllers forecast within the interval – tracking consumption rate against elapsed time and projecting where the interval will land, then discharging early enough to hold it under target. Better still, they learn site patterns and pre-position for known events like shift changes.
When evaluating proposals, ask specifically how the controller decides when to discharge, and what happens on an unforecast step change. Vague answers here predict disappointing results.
Pull twelve months of interval data. For each billing period, find the peak. Then simulate: with a battery of X kW and Y kWh, applying a threshold, what would the new peak have been?
Multiply the reduction by your demand rate. That is the annual saving, before any arbitrage or backup value.
Do this for several battery sizes. The curve typically flattens – the first 200 kW of shaving captures most of the value, and each additional increment captures less. The knee of that curve is your economic size, and it is frequently smaller than what gets proposed.
Before buying storage, look at whether the peaks can be avoided operationally. Staggering equipment starts, sequencing compressors, shifting a batch process by thirty minutes – these cost nothing and sometimes remove a large fraction of the peak.
It is not glamorous and no one sells it, but I have seen sites cut recorded demand by 15% through scheduling changes alone. Do that first, then size storage against what remains. Otherwise you are buying a battery to solve a problem a spreadsheet could have solved.