A battery has two ratings and they do different jobs. Power, in kW, is how fast it can move energy. Energy, in kWh, is how much it can hold. Quoting a system as "a 500 kW battery" tells you half the specification, and it is usually the less important half.
Demand charge reduction is a power problem. You need enough kW to shave the peak, and only enough kWh to sustain that discharge for as long as the peak lasts. If your peaks are sharp 20-minute events, a 500 kW / 250 kWh system might be right – high power, modest energy. Buying 2,000 kWh for that job wastes most of the cost.
Time-of-use arbitrage is an energy problem. You need enough kWh to carry consumption through the expensive window. The power rating just needs to cover your load during that window. A 250 kW / 1,000 kWh system suits a four-hour evening peak.
Backup is usually an energy problem with a power floor. You need enough kW to carry the critical load, and enough kWh to last until the outage ends or a generator starts.
The ratio of energy to power is called duration. A 500 kW / 1,000 kWh system is a two-hour battery. Most commercial applications land between one and four hours, and knowing which you need is the first real decision.
You cannot size storage from a monthly bill. You need interval data, and you need to look at three specific things:
Demand charges are typically set by the single highest interval in a billing period. That means your battery has to catch every peak, not most of them. Missing one event in a month means the charge is set anyway and the battery earned nothing that month.
This has two consequences people underestimate. First, you need margin – sizing exactly to the historical peak leaves nothing for a slightly worse day. Second, the control system matters as much as the hardware. A battery that reacts a minute late has already lost.
I have seen a well-sized battery underperform badly because its controller used a 15-minute rolling average that lagged the site's actual demand. The hardware was fine. The logic was wrong.
Nameplate energy is not usable energy. Manufacturers specify a depth of discharge – commonly 80-90% for LFP systems – and cycling deeper accelerates degradation or voids warranty terms.
A 1,000 kWh nameplate system at 90% DoD gives you 900 kWh usable at beginning of life. After ten years of degradation you might have 700-750 kWh usable. If your application needs 800 kWh in year ten, you have to size for that, not for day one.
Ask for the end-of-warranty usable energy figure, not the nameplate. Good suppliers state it. It is the number your application actually has to live with.
Battery warranties are usually expressed as a combination of years, total energy throughput, and a retained-capacity guarantee. Something like: ten years or 3,500 full cycles, whichever comes first, with at least 70% capacity retained.
That throughput limit constrains your operating strategy. A system cycled once daily uses about 365 cycles a year, so 3,500 cycles is roughly ten years – matched. A system cycled twice daily for aggressive arbitrage burns the warranty in five years.
Which means the business case has to include the cycling regime, not just the tariff spread. Arbitrage that looks profitable per cycle can be unprofitable once you account for consuming warranty life.
The strongest commercial cases usually stack two or three uses on one asset: shave demand peaks, arbitrage the tariff spread, provide backup for a critical load. Each alone often fails the hurdle rate; together they can clear it.
The catch is that stacked applications compete for the same stored energy. A battery held at high state of charge for backup readiness cannot also be cycling for arbitrage. Decide the priority order explicitly, and make sure the control system implements it – this is a configuration question that gets deferred and then never answered.
Take your interval data, apply a simple simulated battery of the proposed size against your actual tariff, and count the value it would have produced over the past twelve months. Not a vendor's model – your data, your tariff, hour by hour.
If the answer is comfortably above the financing cost, proceed. If it is marginal, the sizing is probably wrong, or the application is. Better to find that in a spreadsheet than after installation.