It is worth spending twenty minutes understanding why the demand charge exists, for a practical reason rather than an intellectual one. The rationale predicts the charge's behavior. Once you know what it is trying to measure, its odder features — the interval, the ratchet, the coincident variants — stop looking arbitrary and start looking like design decisions with trade-offs you can reason about.
The problem it solves
A utility's costs do not all behave the same way.
Some scale with how much electricity is delivered. Fuel burned, power purchased on the wholesale market, energy lost as heat in the wires: all of these move with volume, and all of them are naturally recovered through a rate per kilowatt-hour.
The rest do not scale with volume at all. The generating capacity that must be available at the worst moment of the year, the transmission lines that must carry the peak flow, the substation transformer sized for the maximum load on its feeder, the service conductors to your building sized for your own maximum — these are sized for the peak and cost the same whether that peak lasts a season or a quarter of an hour. They are, from the utility's point of view, largely fixed once built.
So a utility recovering all its costs through a volumetric rate would be recovering a capacity cost through a mechanism that has no relationship to what drives capacity cost. And when customers have very different load shapes, that mismatch is not academic.
The subsidy that would otherwise exist
Consider two customers, each consuming the same energy in a month.
The first runs steadily, twenty-four hours a day, at a constant load. The second runs two intense production shifts a week and is otherwise idle, reaching a peak several times higher for those hours.
The system must be built to serve the second customer's peak. That capacity — generation, transmission, the local transformer — exists because of them, and sits underused the rest of the time. The first customer imposed a much smaller capacity requirement for the same energy.
Under a pure volumetric rate, both pay the same. The steady customer therefore pays part of the cost of the capacity the spiky customer required, and the spiky customer receives a discount on the most expensive thing they caused. That is a cross-subsidy, and eliminating cross-subsidies between customers is one of the things a regulator is explicitly there to do.
The demand charge is the mechanism. Volume is billed volumetrically; capacity is billed against the peak that made it necessary. Whatever else is true of it, it is an attempt to charge people for what they cause.
Where the argument is strongest
The rationale is at its most convincing at the distribution level, and specifically for the equipment dedicated to a single customer.
The transformer and conductors serving your site are sized for your maximum, full stop. No averaging with anyone else applies, because nobody else uses them. A non-coincident demand charge — one that bills your own highest interval whenever it occurs — maps almost exactly onto that cost. If your peak is at three in the morning, the transformer still had to be built for it.
Where it is contested
Higher up the system, the argument gets weaker, and this is where rate cases actually get argued.
Transmission and generation are built for the system peak, not yours. If your maximum occurs at four in the morning when the regional system is at its trough, you contributed nothing to the peak that sized the transmission network. A non-coincident charge nevertheless bills you as though you did. This is the central objection, and the response to it is the coincident peak charge — allocating capacity cost by what you drew when the system peaked. See coincident and non-coincident demand.
The ratchet is a rough instrument. It is defended on the basis that capacity dedicated to serving your peak remains dedicated whether or not the peak recurs. That has real force for equipment that cannot be redeployed. Whether a specific percentage over a specific look-back window is a good proxy for the actual cost of standing ready is exactly the sort of question a rate case exists to argue about, and the answers differ between jurisdictions. See ratchet clauses.
A single interval is a noisy signal. Billing an entire month against the worst fifteen minutes gives enormous weight to one measurement. A commissioning test, an unusual production run or a control fault produces a charge unrelated to the site's normal contribution to system cost.
It can work against other policy goals. A demand charge levied on a facility that has done something genuinely useful — installed storage, shifted load off the system peak, added on-site generation — can obscure the value of that action if the charge is measured non-coincidentally. Standby charges for sites with their own generation raise the same tension in sharper form: standby and supplemental charges.
None of these is a knock-down argument, and none of them means the charge is illegitimate. They mean it is an approximation, chosen because it is administratively workable, with known weaknesses that rate design continues to grapple with.
What follows for you
Three things, all practical.
The charge is a price signal and it can be answered. It exists to make the cost of capacity visible to the customer who causes it. A site that flattens its load is genuinely reducing the system cost it imposes, and the saving is not a loophole. That is the mechanism working.
Understanding the rationale tells you which measures will work. The charge prices the height of your load curve, not its area. Anything that reduces height reduces the charge; anything that reduces area but not height does not. That is why six efficiency measures that do not cut your demand charge is a real list rather than a contrarian one.
The tariff is a regulated document, not an offer. You cannot negotiate the rate. You can change which schedule you are on, and you can change the determinants. That is the whole of the strategy set out in how to reduce peak demand charges, and it starts from the definitions in demand charges explained.
