Battery ROI calculator: methodology

The mistake this avoids

The obvious way to model a battery is gross arbitrage: capacity, times cycles, times the gap between the cheap and expensive rate, times efficiency. Almost every calculator does some version of this, and it is structurally wrong.

A household does not switch on a battery. It switches tariff. Moving to a time-of-use tariff to get the cheap overnight rate also moves every daytime unit onto that tariff’s day rate — which is normally higher than the flat rate they were paying. Octopus Go’s day rate has been running around 3p above Octopus Flexible. On a 3,600 kWh home that is £70 to £100 a year of headwind before the battery earns a penny, and gross arbitrage never sees it.

So we model the difference between two whole annual bills: what you pay now, against what you would pay on the time-of-use tariff with a battery shifting what it can. Standing charges included, because they differ too.

The calculation

daily consumption      = annual kWh / 365
peak-window demand     = daily consumption × peak share
shiftable per day      = min(usable kWh × efficiency, peak-window demand)
kWh bought to charge   = shiftable / efficiency

cost now  = annual kWh × current rate + standing charge
cost with = (peak demand − shifted) × TOU day rate
          + (off-peak demand + charge kWh) × TOU night rate
          + standby draw × TOU day rate
          + TOU standing charge

saving    = cost now − cost with

The min() is the part that matters most and is most often missing. A battery cannot discharge electricity the house will not use. Once usable capacity exceeds peak-window demand, further capacity earns exactly nothing — while still appearing on the invoice.

Where the numbers come from

The load shape

We need one figure: what share of daily consumption falls outside the cheap window. It is derived from OpenSynth / Faraday, published by Centre for Net Zero under CDLA-Permissive-2.0 — synthetic household profiles generated from around a billion smart meter readings, sampled 1 March 2024 to 1 March 2025. We use 509,082 synthetic profiles, filtered on the dataset’s own labels to households with no solar, no battery, no electric vehicle and no heat pump — the cohort this calculator addresses. Being able to exclude households that already have a battery is the reason this source beats the alternatives.

The cross-check that makes it trustworthy: Faraday gives a mean of 9.73 kWh/day. SERL Statistics Report 1 — an entirely independent source, around 13,000 real metered GB households — publishes a mean of 9.8 kWh/day. One synthetic dataset and one metered one, 0.7% apart.

One difference we previously flagged here as unresolved now is resolved, and the answer is worth keeping. Faraday puts the evening peak at 17:00 and SERL puts it at 19:00, and we did not know whether that was a timezone artefact. It is not. British clocks move twice a year and British habits do not, so if these timestamps were UTC the summer months would show the evening peak a whole hour earlier than the winter ones. Across all twelve monthly cohorts it stays in the same hour, drifting by about fifteen minutes — and the overnight trough sits at 03:00, exactly where every published domestic curve puts it. The hour index is local time and the shape is aligned to the clock. What is left of the gap against SERL is a difference of population, or of mean against median, not of alignment.

A caveat worth stating: the training population is Octopus Energy customers, who over-index on smart tariffs and low-carbon technology. Centre for Net Zero resample a non-LCT subgroup to represent broader demographics, and filtering out that technology, as we do, lands in that subgroup. It is still not a clean national average.

That gives, for the common windows:

For comparison, a 6-hour window would be 25% of the day if consumption were flat. It is nearer 15%, because households use very little at 3am — and that gap is the entire reason a battery has anything to earn on.

Efficiency, standby and degradation

• Round-trip efficiency 85%, deliberately not the datasheet figure. Tesla quote 89% for the Powerwall 3, explicitly qualified as “typical solar shifting use case, 25°C, at beginning of life”. A UK battery doing grid arbitrage in a garage in February at part load meets none of those conditions. Enphase published measurements for their own product: 90% nameplate, 85.6% over 24 hours, 84.0% over a month.
• Standby 15 W, drawn around the clock whether the battery is working or not. Small per hour, £30 to £50 a year.
• Degradation 2% a year, consistent with Tesla warranting 80% retained capacity at 10 years. The saving shrinks each year, which is why we show the schedule rather than one number times ten.

Worth knowing on cycle life: Tesla’s warranty distinguishes use cases, and “time-based control” — which is what overnight arbitrage is — carries unlimited cycles rather than a throughput cap. Daily cycling is warranted.

What actually moves the answer

This is worth stating plainly, because it is the opposite of where most attention goes. Ranked by how much each changes the payback:

1. What you pay now. The counterfactual tariff. Someone on a keen fix has a much worse case than someone on the cap.
2. Round-trip efficiency. The gap between the datasheet and reality is worth 10 to 20% of the answer.
3. Standby losses. Comparable in size, and almost never modelled.
4. The load shape. Resolving consumption seasonally rather than as one annual share changes the result by 0 to 5%, and only in a narrow band where capacity and demand are closely matched.

The tests in this repository assert that ordering, so it cannot quietly stop being true.

We can put a number on the last one, because we built this calculator twice. The first version used a load shape reconstructed by hand from SERL’s published chart anchors, before the Faraday data was in place. Swapping in half a million real filtered profiles moved the peak-window share from 90.3% to 90.2% on a four-hour window, and from 84.6% to 84.8% on a six-hour one. Two tenths of a percentage point.

We report that because it is the opposite of a flattering result. Getting the data was the right thing to do — the figure is now derived rather than interpolated, which is what this site is for — but anyone waiting on a dataset before answering this question was waiting on the wrong thing. The model was where the error lived.

Who this refuses to answer

Four household types break an assumption the model depends on, and it declines rather than guessing:

• An electric vehicle charged at home. The car charges inside the same cheap window and usually draws more than the battery holds, so it competes for the window and reshapes the whole day. This model would understate your bill and overstate the battery.
• A heat pump. Roughly doubles electricity use and makes it strongly seasonal and weather-driven. Seasonality is the one thing a single annual figure genuinely cannot handle.
• Electric storage heating. Already designed around the cheap window, so most of your consumption is off-peak and there is far less for a battery to shift.
• Solar panels. A different calculation — self-consumption and export interact with arbitrage.

What we assume, and where it could be wrong

• Perfect daily scheduling, with no forecast error and no missed cycles.
• The battery available every day — no outages, no holidays away from home.
• Rates flat in real terms. In practice the spread moves, and so does the cap.
• No export income, no grid-services or virtual power plant revenue.
• A consumption shape borrowed from a population median, not your home. If your day looks unusual — empty during the day, or a night shift — it will be off.
• The single-share approximation biases the saving upward by up to 5% where capacity and demand are closely matched. The error has a known sign, so treat a marginal result as optimistic rather than uncertain.

Sources

• Round-trip efficiency, measured rather than nameplate
• Degradation and warranty terms (Tesla Powerwall 2, EMEA)
• Domestic load shape (SERL Statistics Report 1)

Back to the battery ROI calculator, or read why home batteries mostly don’t pay for themselves.