Size a Home Battery From Your Load Profile, Not a Brochure

There is no single quantity called "battery capacity." There are four of them, they can differ by double digits on the same product, and nothing obliges a sales conversation to say which one it is quoting. The proposal on my table said 13.5 kWh. It never said which of the four that was.

Underneath, in smaller type: enough to run your home for a day.

Whose day? Twelve months of my own consumption were sitting in a CSV at the time, and the spread between my quietest twenty-four hours and my worst was a factor of four. A battery that covers the quiet day runs out at 3 a.m. on the day that matters. A battery that covers the worst day spends most of the year half empty, and you paid for the empty half.

That file is free, and your utility is already holding it.

Four numbers a brochure blends into one

Before any arithmetic, separate the terms. Mixing them is the single move that makes a battery look bigger and cheaper than it is.

Nameplate (rated) energy capacity — the energy in the cells at rated conditions. This is the number that tends to appear on regulatory equipment lists. The California Energy Commission's battery list, which I checked on 17 August 2026, carries a column headed literally Nameplate Energy Capacity Rate (kWh) across its several hundred listed models — not usable capacity, not efficiency.

Depth of discharge (DoD) — the fraction of nameplate the manufacturer permits you to cycle. Cells last longer if you never take them to empty, so the control system reserves a slice you cannot reach.

Usable capacity — nameplate × DoD. California's 2025 Energy Code puts the definition plainly: the minimum rated usable energy capacity is "the battery energy storage system capacity in kWh that a manufacturer allows to be used for charging and discharging" (CEC, Battery Energy Storage Systems requirements, read 17 August 2026). That page also sets the floors a qualifying system must clear: single charge-discharge cycle AC-to-AC round-trip efficiency greater than 80 percent, and 70% of nameplate capacity retained after 4,000 cycles under warranty, or 70% under a ten-year warranty.

Round-trip efficiency (RTE) — AC energy out divided by AC energy in over one full cycle. This one does not shrink what the battery gives you. It inflates what you must feed it. Ten kWh out of an 88% round-trip system cost you 11.4 kWh in, and those 11.4 kWh had a price.

So when a salesperson says "13.5 kilowatt-hours," the honest follow-up is three questions: is that nameplate or usable, measured at DC or at the AC terminals, and what is the round-trip figure the same datasheet states? Published specs do exist for this. Enphase's own product page for the IQ Battery 5P states 5.0 kWh usable energy, 3.84 kW continuous power, and a 15-year limited warranty up to 6,000 cycles (Enphase IQ Battery 5P, read 17 August 2026). Those are three separate constraints and you will hit all three.

Getting your own interval data

The data standard is called Green Button. It came out of a 2012 White House call to action and is now maintained by the Green Button Alliance, a North Carolina nonprofit. Two flavours exist: Download My Data, where you log into your utility account and export a file yourself, and Connect My Data, where you authorise a third party to pull it via API.

You want Download My Data. It is free, it does not hand your account to a vendor, and it is the file a good installer would have asked for anyway.

The path differs by utility. At PG&E it sits under Energy Usage Details in My Account, behind a Green Button icon, with a choice of CSV for spreadsheets or ESPI XML for machine reading (PG&E's residential download instructions). Interval granularity is a utility decision — 15 minutes and hourly are both common, and monthly-only is still the reality in plenty of service territories. If yours only gives monthly totals, you cannot size a battery from it, and you should say so out loud when a proposal claims to have modelled your usage.

Five things to check the moment the file opens, before you trust a single sum:

1. Is the column kWh or kW? Fifteen-minute data sometimes reports average demand in kW. Energy for that interval is kW × 0.25. Getting this wrong makes your day four times too big.
2. Does the year have 8,760 rows of hourly data? Daylight saving gives you one 23-hour day and one 25-hour day. Gaps beyond that mean meter outages, and a summed month with holes reads artificially low.
3. Existing solar makes the file lie about your load. The meter records net consumption, so every kWh your panels covered is invisible. Battery sizing needs gross load, which means pulling inverter production separately and adding it back in.
4. Is it a full twelve months? Sizing is decided by seasonal peaks. One month tells you nothing about the other eleven.
5. Check whose meter you are actually reading. A separate EV meter or a submetered ADU means the file describes part of the house and the sum you build from it will be short.

Worth knowing which half of this is standardised. The protocol underneath Green Button is NAESB's REQ.21 Energy Services Provider Interface, described as "a data exchange protocol for the transfer of energy usage information between a utility and a third party with customer authorization" (NAESB, ESPI standards, read 18 August 2026), and what it governs is the XML exchange. The CSV a portal hands you is a convenience export layered on top of that, carrying column headings and units the utility chose for itself. So the five checks above are yours to run rather than something the file prevents, and a spreadsheet template built around one utility's export usually needs its columns remapped before it reads another's.

Two different questions, two different sums

A battery does two jobs. They size differently, they are priced differently, and a brochure that quotes one number is quietly answering only one of them. (The two jobs also cost different amounts and fight over the same kWh once the system is running; here the question is only how big.)

Job one: carrying an outage

Decide the window first, because "backup" without a duration is not a specification. For scale: EIA reported that U.S. electricity customers averaged 11 hours of interruption in 2024, of which roughly 9 hours came from major events — Hurricanes Beryl, Helene and Milton alone accounted for about 80% of total outage time — leaving around 2 hours of routine, non-major-event outage (EIA, Today in Energy). A two-hour national average and a three-day hurricane are not the same design problem. Yours depends on where you live, and your own outage history beats any average.

Then, in the spreadsheet:

E_window = sum of interval kWh across the hours you want to ride through,
           restricted to the loads you will actually back up

usable_kWh_needed = E_window ÷ (1 − reserve)

Two notes on that. The reserve is the state-of-charge floor you refuse to hit — many owners set 20% so the battery can still start a well pump the next morning. And if the datasheet's usable figure is stated at the AC terminals, do not derate it again for round-trip efficiency; the discharge-side losses are already inside that number. Derating twice is the most common way people talk themselves into an extra unit.

The energy sum is the easy half. The constraint that actually bites is power. Interval data is an average over 15 minutes or an hour, so it physically cannot show you a well pump's locked-rotor surge or a compressor start. A 3.84 kW continuous rating is a hard ceiling regardless of how many kWh sit behind it, and a two-second inrush does not care that your hourly average was 0.9 kW. This is why critical-load panels exist. Sum the nameplate draw of everything you plan to back up and compare it to the continuous kW rating, separately from the kWh math.

Job two: shifting kWh out of the expensive window

This one is pure arithmetic on your tariff, and it is the one a proposal is most likely to have flattered.

kWh_charged  = kWh_delivered ÷ RTE
value/cycle  = (kWh_delivered × p_peak) − (kWh_charged × p_charge)
annual value = value/cycle × cycles per year

p_peak is the import price during the window you are avoiding. p_charge is what the charging energy costs you — either the off-peak import rate, or, if you charge off your own array, the export credit you gave up by not sending those kWh to the grid. Both come off your rate schedule PDF, not off a national average — and on a tariff with riders or tiers, pulling those two prices off your own bill is a job of its own that has to happen before this formula runs.

Worked, with stand-in numbers purely so the shape is visible — these are not your rates and not a forecast:

• Delivered per cycle: 8 kWh · RTE: 0.88 · so charged: 8 ÷ 0.88 = 9.09 kWh
• p_peak $0.42/kWh · p_charge $0.11/kWh
• Value per cycle: (8 × 0.42) − (9.09 × 0.11) = 3.36 − 1.00 = $2.36
• At 300 usable cycles a year: $708/year

Now put the ceiling on it. A 6,000-cycle warranty at one cycle a day is 16.4 years. At 300 cycles a year it is 20. If your payback lands past the warranty's cycle budget or its 70%-retention point, the arithmetic has told you something, and it is not "buy a bigger one."

Notice what the formula also tells you: kWh_delivered is capped by how much you actually consume inside the peak window on a normal day. Not the biggest day. A battery cannot save you money on energy you were not going to buy.

The percentile step, which is where the brochure loses

Here is the move that changed my own answer.

Build one column: for each of the 365 days, the total kWh inside your target window — the outage hours, or the peak tariff hours. Then sort it and read off the median, the 90th percentile, and the maximum.

The gap between the median and the maximum is what you are being asked to pay for. Sizing to the maximum buys capacity that gets used a handful of nights a year. Sizing to the 90th percentile means about 36 days a year fall short — which is a catastrophe if the load is a medical device and a shrug if it is the second fridge and the TV.

That is a judgement call, and it is yours. But it is a judgement made against a distribution you can see, instead of against the phrase "runs your home for a day."

If your utility only publishes monthly totals, the distribution is not out of reach — it is just not yours. NREL's End-Use Load Profiles for the U.S. Building Stock puts 15-minute residential profiles by building type and climate region on the Open Energy Data Initiative platform (OEDI, End-Use Load Profiles, read 18 August 2026), which at least shows you how wide a median-to-maximum gap tends to be. Either way, put a price on the gap: multiply the extra kWh between your 90th percentile and your maximum by the installed cost per usable kWh on the quote in front of you. That product is what the last few nights a year actually cost, and it is a far easier thing to accept or refuse than a capacity recommendation.

What I got wrong the first time

I modelled an average day. Twice. The first pass used monthly kWh divided by 30, which produced a tidy number and a battery that would have been flat by 2 a.m. in January.

I also missed that the federal credit had lapsed underneath me. If a quote nets out a 30% federal credit against the battery price, check it before you believe the payback line: the IRS's Residential Clean Energy Credit page states the credit "is not available for any property placed in service after December 31, 2025" (IRS, Residential Clean Energy Credit, read 17 August 2026). Battery storage of at least 3 kWh had been eligible from 2023. Past tense. A proposal still carrying that line is either old or careless, and either way you now know which of its other numbers to re-check.

What I would do differently: download the twelve months before booking any site visits, and build the percentile column before hearing a single capacity recommendation. Once someone has said "13.5" out loud, every subsequent calculation quietly aims at it.

One structural thing helps here: the anchoring number is a product, not a result. Capacity arrives in whatever increment the manufacturer ships — the IQ Battery 5P cited above is 5.0 kWh usable per unit, and Enphase calls the design modular — so on that product line capacity moves in fives regardless of what your spreadsheet asked for. Settle your own number first, then find out which increments land near it. Reverse the order and the increment makes the decision for you.

Start with twelve months of your own data

Log into your utility account and export twelve months of interval data as CSV. Add two columns — date and hour-of-day — and produce one number: the median kWh inside the window you actually care about.

Bring that number to the conversation. Ask which of nameplate or usable the quoted capacity is, ask for the round-trip efficiency from the same datasheet, and ask what continuous kW the unit holds. If a proposal cannot answer all three from published documents, it was not sized on your house.