Peak Load Management

Peak load management has an unusual property among cost-reduction projects: the theory is simple, the savings are large and recurring, the capital requirement is often zero — and almost nobody does it well. The reason is not technical. It is organizational. Managing peaks requires somebody to make a disruptive operational decision on a few hours' notice, several times a year, on days that cannot be scheduled in advance. In most companies that authority does not exist, so the intervals pass unmanaged and the bill arrives.

This is a guide to building the program rather than to the billing mechanics. If you need the underlying concepts first, start with understanding demand charges and understanding capacity charges — this piece assumes them.

You Are Billed on Three Different Peaks, and One Action Hits All Three

The single most useful thing to understand is that "your peak" is not one number. Depending on your market and rate schedule, up to three separate peaks are being measured against you, on different clocks:

• Your facility peak. The highest demand interval in the billing month, which sets the demand charge on that month's bill. Measured against you every month, and often subject to a ratchet that carries a bad month forward for a year.
• The system capacity peak. Your demand during the hours the regional grid hits its annual maximum sets your capacity obligation for the following delivery year — the PJM peak load contribution tag, the ISO-NE capacity tag, the NYISO installed capacity requirement. A handful of hours, once a year, priced for twelve months.
• The transmission peak. In some markets transmission cost is allocated the same way on its own schedule, most famously the ERCOT Four Coincident Peak method, which uses one fifteen-minute interval in each of June, July, August and September.

These overlap heavily. Hot summer weekday afternoons tend to be when all three occur. That is the leverage: a curtailment executed on the right August afternoon can simultaneously reduce that month's demand charge, next year's capacity tag and next year's transmission allocation. One decision, three separate line items, and two of the three pay out for a full year afterward.

Why Programs Fail

Before the method, the failure modes, because they are consistent:

• No owner. The alert goes to a distribution list. Everyone assumes someone else will decide. Nothing happens.
• No pre-agreed authority. The facility manager is not sure they can stop production without a plant manager's approval, and the plant manager is in a meeting.
• No playbook. Even with authority, "reduce load" is not an instruction. Which equipment, in what order, for how long, and who confirms it came back?
• No verification. The curtailment happened, nobody measured what it saved, and the following year finance cannot tell whether the disruption was worth it. Unverified programs get canceled.
• Attacking the wrong peak. A site that curtails diligently every month to shave demand charges, but does nothing on the four days that set the capacity tag, is doing the harder work for the smaller prize.

Step 1: Get Interval Visibility

You cannot manage what you see only in monthly aggregate. The starting requirement is interval data — fifteen-minute or hourly consumption for at least twelve months, which your utility holds and will release to you or to an advisor with a Letter of Authorization.

What you are looking for in that data is not the average. It is the shape of the top 1% of intervals. Specifically: how far above your typical demand your peaks sit, how many intervals account for the peak, whether the peaks cluster by time of day and day of week, and whether they coincide with the grid's peaks or are driven by something internal and idiosyncratic like a morning startup surge. A site whose peaks are internal and predictable is a much easier program than one whose peaks track ambient temperature.

Step 2: Rank Your Load by How Movable It Is

Build an inventory of significant loads and sort them into four honest categories:

• Shiftable. Can run at a different time with no loss of output — battery charging, water pumping into storage, some batch processes, non-urgent compressed air demand.
• Reducible. Can run at lower intensity temporarily — HVAC setpoint drift of two or three degrees, dimmable lighting, reduced ventilation rates outside occupied hours.
• Interruptible with cost. Can stop, but you lose something — a production line, a curing oven, a shift's output. These have a real price per hour and should be quantified rather than assumed unavailable.
• Firm. Cannot move at any price — life safety, critical process, refrigeration below a hard temperature threshold, data center IT load. Take these off the table and stop discussing them.

Attach an estimated kW and an estimated cost-per-event to each item. The output is a ranked curtailment ladder: what you shed first, second, third, and where you stop. Most facilities discover more shiftable load than they expected and less interruptible load than they feared.

Step 3: Define the Trigger

A program needs an unambiguous rule for when to act, decided in advance and not by committee on the day. Practical triggers:

• Grid operator day-ahead load forecast above a defined threshold.
• Forecast high temperature above a defined threshold, in the relevant months.
• A real-time alert from a curtailment service, an energy management platform, or an advisor watching the market on your behalf.
• For monthly demand charges specifically: an internal alarm when running demand approaches the month's current maximum, which is a different and more frequent trigger.

Set the threshold to over-call. If the trigger fires on twelve days and four of them turn out to be the ones that mattered, the program worked. If it fires on four days and misses one that mattered, you lost a quarter of a year's capacity savings to protect eight afternoons of full production.

Step 4: Write the Playbook, Then Rehearse It

One page. Who receives the alert, who makes the call, what the call authorizes without further approval, the ordered list of loads to shed with target kW for each, who executes, who confirms execution, when load is restored, and who logs it. Then run it once on a non-peak day, in advance, to find out which of those steps does not survive contact with reality — because at least one will not.

Step 5: Verify and Attribute

After each event, pull the interval data and record what actually happened: baseline demand, achieved demand, kW avoided, duration, and any operational cost incurred. At the end of the season, reconcile against the bill — the demand charge lines, the following year's capacity tag, the transmission allocation.

This step is what keeps the program alive. Peak management asks operations to accept real disruption for a benefit that shows up months later on a document they never see. Unless somebody closes that loop with a number, the cooperation erodes within two seasons. Verification is not administrative overhead; it is the thing that buys you next year's participation.

Where Automation Fits

Manual execution is fine for a handful of large loads and a facility with staffed operations. Automation earns its cost in three situations: when the curtailment must happen faster than a human can act, when it must happen across many sites at once, and when the loads are numerous and small enough that manual sequencing is impractical. Building automation systems, dedicated demand controllers and storage dispatch all belong to this category.

The order matters. Buy controls after you know what you are controlling and what it is worth. Sites that buy the platform first frequently end up with excellent visibility into a peak nobody is authorized to reduce. For the broader question of what software solves versus what advisory work solves, see energy management systems versus energy management services.

What It Is Worth

The return varies with rate structure and load, but the shape is consistent: near-zero capital, a real but bounded operational cost on a handful of days, and savings that recur annually without further investment. On a demand-metered industrial or large commercial account, this is routinely the highest return-per-hour-invested project available, and it is the one most likely to still be undone five years from now. If the same curtailment capability is enrolled in a paid program, it becomes revenue as well — see demand response in PJM and NYISO.