How much electricity does a heat pump actually use?

Almost every answer to this question is modelled: take a heat demand, divide by an assumed SCOP, quote the result. This one is measured. It comes from 45,602 heat pump households in smart-meter-derived data, against 1,786,768 without one — and the first thing the data does is disagree with the number everybody quotes.

The headline figure, and why not to use it

Households with a heat pump use 25.2 kWh a day. Households without one use 9.74. The difference is 15.49 kWh a day, or about 5,289 kWh a year — £1,381 at the 2026 Q3 cap.

That is roughly double what SCOP arithmetic predicts for a typical home, and if you have been quoted a figure near it as “what a heat pump costs”, it is too high. Here is how to see why.

First test: are heat pump owners just in bigger houses?

The obvious objection. Heat pump adopters skew detached, rural, off the gas grid. So compare like with like: heat pump households against households of the same property type and the same EPC band, and nothing else.

Weighted across those pairs, the difference is 15.22 kWh a day, against 15.49 unmatched. Controlling for house shape and insulation band moved it by under two per cent.

So the easy explanation fails. It is not that heat pump owners live in bigger houses — or rather, they do, and it accounts for almost none of the gap. Something else is going on, and the second test finds it.

Second test: what does July look like?

Nobody heats their house in July. If the difference between these two groups were the heat pump, it would collapse in midsummer. It does not.

In July the gap is still 9.16 kWh a day. That is far too much for hot water. Whatever it is — bigger homes off the gas grid, larger heat demand, occupancy patterns the dataset does not label — it is present in January too, and it has nothing to do with space heating.

Which gives us the honest estimate. Difference it out. Everything above the summer floor only appears when it is cold, and space heating is the only thing that behaves that way.

A sanity check on that: at a real-world SCOP of 3, 1,945 kWh of electricity delivers about 5,834 kWh of heat. A gas-heated British home’s space heating is somewhere around 6,000 to 8,000 kWh of delivered heat. The measured figure and the modelled one land in the same place, which the headline figure does not.

So is it cheaper than gas?

This data cannot tell you, and it is worth being clear about why. A smart meter records electricity in. It does not record heat out. Without heat output there is no SCOP, and without a SCOP there is no verdict — only the electricity, which is what we have measured.

What the cap does tell you is the bar. At 7.33p for gas and a 90% efficient boiler, gas heat costs 8.14p per kWh delivered. Electricity at 26.11p matches that at a SCOP of 3.21. Above it the heat pump wins on running cost; below it the boiler does. Well-commissioned systems on properly sized emitters clear it comfortably. Poorly commissioned ones on undersized radiators do not, and that is the whole ball game.

The 2026 Q4 cap is already published: electricity 26.32p, gas 7.97p, which puts the break-even SCOP at 2.97. What the October change means in full.

The finding nobody was looking for: they run at 4am

Households without a heat pump peak at 17:00, exactly like every published domestic load curve. Households with one peak at 04:00, with a second peak in the early afternoon.

The obvious explanation is smart tariffs: those are the cheap windows, and the fleet is pre-heating into them. We published that explanation, and it is wrong. Here is what corrected it.

The dataset labels how each household is billed, which allows the one comparison that settles it: heat pump households on a standard flat tariff, where electricity costs the same at 4am as at 4pm and there is no financial reason whatsoever to run overnight.

Same tariff. Same prices at every hour of the day. And the heat pump households still peak at 04:00, still putting 25% of their electricity into the small hours against 14% for their neighbours. Nobody is chasing a cheap rate. There isn’t one.

So the shape is thermal after all, and the reason is one we should have reached for first: pre-dawn is the coldest part of the night. Heat loss through the fabric is at its greatest, and a heat pump’s efficiency is at its worst, at the same hour — so the identical amount of delivered heat costs the most electricity precisely then. A weather-compensated system modulating continuously rather than firing in bursts will trace that curve exactly.

The tariff is not irrelevant, just secondary. Moving from a flat tariff to a smart one lifts the overnight share from 25.1% to 28.0% — about 3 points of deliberate shifting, sitting on top of roughly 11 points that the physics puts there whether you want it or not. The thermal effect is more than three times the size of the behavioural one.

What actually follows:

• A heat pump is least efficient exactly when you need it most. Not a scheduling quirk — the fundamental constraint. Sizing and emitter design have to work at the coldest hour, not the average one.
• Smart tariffs help less than the raw load curve suggests. Around 3 points of your consumption moves, not the bulk of it. Still worth having; not the transformation the shape implies.
• The overnight concentration is a real opportunity, though. A quarter of a heat pump household’s electricity is already being used in the cheapest hours of any time-of-use tariff, without anybody arranging it.

What this does not answer

• Your house. These are population means with a wide spread behind them.
• Whether insulation helps. It must, but this data cannot show it: heat pump households rated A–C and D–E come out within half a per cent of each other, which is far likelier to be a limitation of the model than a fact about British housing. We publish the EPC cuts and build nothing on them.
• The counterfactual. We do not know what these homes heated with before, and many are probably off the gas grid, where the comparison is oil or electric heating rather than a boiler.
• Real meters. Faraday is synthetic — a generative model trained on around a billion smart meter readings from Octopus Energy customers. Our baseline reproduces SERL’s independently published mean to within 0.6%, which is why we use it, but no single profile is a real household.

Sources

• OpenSynth / Faraday OpenSynth/cnz-faraday-5.0, Centre for Net Zero, CDLA-Permissive-2.0, sampled 1 March 2024 to 1 March 2025. Our derived aggregates, published in full.
• Ofgem price cap unit rates and standing charges, verified 2026-09-14.
• HMRC: temporary zero rate of VAT for domestic electricity in Great Britain.

Compare your own use against these cohorts with the energy use benchmark, or work out whether a heat pump makes sense for your house with the heat pump calculator.