A commercial account’s bill is a stack of separate charges that happen to arrive on one page. They behave differently, they respond to different things, and a battery touches them unequally. Reading the bill as a single number is the most common mistake in behind-the-meter storage analysis, and it is the one that survives longest, because a wrong total still looks like a total.
What a commercial bill is made of
Energy charges bill the quantity of electricity consumed across the billing period, priced per unit of energy. On a time-of-use tariff that quantity is split across defined windows: on-peak, off-peak, and often a shoulder or mid-peak period, priced differently in each. The windows are defined by the tariff, they usually change by season, and they are frequently redrawn when the tariff is revised.
Demand charges bill capacity rather than energy. They are priced per unit of power and applied to the highest demand the meter recorded during the period, averaged over the interval the tariff specifies. A single coincidence of equipment can set the demand charge for an entire month. Many tariffs bill more than one demand quantity: a facility or maximum demand covering the whole period, plus an on-peak demand that looks only inside the peak window.
Fixed charges are the customer charge, service charges, and metering charges that do not move with consumption at all.
Riders and adjustments are the layer that surprises people. Fuel and purchased-power adjustments, capacity and transmission cost recovery, efficiency-program surcharges, franchise fees, and any number of jurisdiction-specific items. Some ride on energy, some on demand, some are flat. Individually small, collectively material, and revised on their own schedules.
Taxes apply on top, sometimes to a subset of the lines above rather than all of them.
The structural fact underneath all of it: a battery does not act on “the bill.” It acts on specific billing determinants, and those determinants flow into the lines above through the tariff’s own arithmetic. Until you know which determinants a site is billed on, you do not know what a battery is worth there.
Ratchets, and why last summer can bill all winter
A demand ratchet sets a floor under billed demand using a peak from an earlier period. A tariff might bill the greater of this month’s measured demand and some percentage of the highest demand recorded over a look-back window, often weighted toward the summer season.
Consider Site A, on a summer-peaking tariff with a ratchet. One August afternoon sets the highest demand of the year. For months afterward the account is billed against that floor even though measured demand never approaches it. The value of shaving that one peak is therefore not one month of savings: it is the difference between two annual billing paths.
Ratchets vary in shape. Percentage, look-back length, whether they apply to facility demand or on-peak demand, whether season matters, whether they reset. They are ordinary tariff text, and they are an ordinary omission in analysis. Missing a ratchet makes every subsequent month wrong in the same direction, quietly, which is a worse failure than being wrong loudly.
Which lines a battery can move, and which it cannot
Directly movable. On-peak energy charges, by shifting consumption out of expensive windows into cheaper ones and paying round-trip losses for the privilege. Demand charges, by discharging into the site’s peaks so the meter never records them. And anything that rides proportionally on those determinants.
Not movable. Fixed customer and service charges. Flat monthly riders, which do not care that a battery exists. Taxes move only insofar as the taxed base moves.
Conditional. Export compensation and standby or supplemental service provisions can appear the moment a battery starts exporting or the site takes service under a different provision. A tariff can also reclassify an account entirely: crossing a demand threshold, adding export capability, or changing the service configuration may move the site onto a different rate schedule with a different structure. That reclassification can dominate everything the dispatch does, in either direction.
The battery’s own load. Round-trip losses and standby draw appear on the meter as consumption. Charging at the wrong time can set a new demand peak. This is not hypothetical: it is the most common way a poorly controlled installation gives back part of what it saved.
The bill and the interval data have to be read together
Neither one is sufficient, and the reason is worth stating plainly.
The bill is ground truth for structure. It tells you which schedule the account is on, which determinants were billed, which riders applied, and what was actually charged. What it cannot tell you is shape. A demand charge is one number. It does not say when the peak occurred, how long it lasted, how often the site came close, or whether the month contained one such event or thirty.
The interval data is ground truth for shape and nothing else. It does not know which tariff the account is on, that a ratchet applied, or that a rider changed in April.
Storage value lives in the interaction between them. How much a battery can shave depends on the duration and frequency of peaks, which only the interval data reveals, priced by determinants only the bill defines. Analysis built on one without the other is describing half a site.
The practical test is reconstruction: take the interval data, apply the tariff as written, and check whether the calculated bill matches the delivered bill line by line. If it reconciles, the model is anchored to something real. If it does not, a problem has been found before it reached a credit memo rather than after.
When the bill and the model disagree
Assume the model is wrong first. Reconciliation failures cluster in a few places.
Tariff version. Tariffs are amended, sometimes several times a year, and the amendment that matters is often a rider rather than the headline price. Confirm that the version applied covers the service period printed on the bill, not the version in effect today.
Determinant definition. Billed demand is frequently not measured demand. Ratchets, contract demand minimums, power factor adjustments, and demand measured only inside a window all change the quantity that gets priced. When a demand line is off by a clean factor, this is usually why.
Window boundaries. Season change dates, holiday treatment, and whether the on-peak window follows local clock time through a daylight saving transition. Small definitional errors here move on-peak energy in a way that reads like a data problem.
Meter coverage. Gaps, duplicated periods, a billing period that does not align with the interval file’s boundaries, an account served by more than one meter, or a meter exchanged mid-period. A partial month reconciles badly, and totals alone rarely reveal why.
Sign and channel conventions. Whether export is negative consumption or a separate channel, whether reactive quantities have been mixed in, whether the file records energy per interval or average power. These produce wrong answers that look entirely plausible.
Estimated bills. Some bills are estimates, trued up later. A model reconciling against an estimate is reconciling against a guess.
The discipline is sequential: fix the model until it reproduces the historical bill, and only then run a scenario forward. A model that cannot reproduce the past has no standing to describe the future.
What decision-grade means for bill analysis
Decision-grade is a short list, and none of it is clever.
The tariff version applied is stated, and it matches the service period. Every billed determinant is reconstructed from the interval data rather than assumed. Ratchets, minimums, and reclassification thresholds are modeled explicitly, including the ones that do not bind today but would bind under the battery’s operation. The calculated bill reconciles to the delivered bill line by line, and where it does not, the difference is explained rather than tolerated. Every number traces back to a source document and a version of the method that produced it. And the whole analysis can be run again next quarter, against the same inputs, and return the same answer.
That is unglamorous work, and it is where storage cases are won or lost. Utilyzed builds the bill layer to that standard, and the methodology behind it is readable without a login.