Open data
UK domestic electricity consumption, by household type
150 cohorts covering property type, EPC band, month, heat pump and EV ownership — each with a mean daily total and a 24-hour load shape. Derived from 3,000,000 synthetic profiles. Free to use, including commercially, with attribution.
Why this exists
Faraday is an excellent dataset and it is essentially unusable without a few gigabytes of download, a parquet reader and an afternoon. Almost nobody who wants to know what a terraced house uses in February is going to do that.
So we did it, and published the result. These are the aggregates behind the calculators on this site. If you want to check our arithmetic, argue with our conclusions, or use the numbers for something we have not thought of, they are yours.
The cross-check
The baseline cohort comes out at 9.737 kWh/day from 1,786,768 profiles. 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, 0.6% apart.
Since the deciles landed there is a second, independent check on the same cohort. SERL publishes a median of 8.2 kWh/day as well as that mean, and ours comes out at 8.31. Two different moments of the same distribution, from two unrelated datasets, agreeing to within about a per cent each. That agreement is the reason to take any of the rest seriously.
What is in it
| Cut | Key fields | Cohorts | What it is |
|---|---|---|---|
| baseline | — | 1 | No solar, battery, EV or heat pump. The reference cohort. |
| by_property | property | 6 | The strongest single predictor in this data. |
| by_epc | epc | 4 | Published, but see the limitation below. |
| by_property_epc | property | epc | 23 | The cross-tabulation. |
| by_month | month | 12 | Seasonality for households without a heat pump. |
| by_tariff | tariff | 2 | How the household is billed. The most useful cut here — see below. |
| by_day_of_week | day_of_week | 2 | Weekday against weekend. |
| by_cluster | cluster | 29 | Faraday’s k-means grouping of the household’s area. Not a region. |
| heat_pump | heat_pump | 2 | With against without, unmatched. |
| heat_pump_by_tariff | tariff | heat_pump | 5 | The cut that showed the 04:00 peak is thermal, not financial. |
| heat_pump_matched | property | epc | heat_pump | 32 | With against without, within the same property type and EPC band. |
| heat_pump_by_month | month | heat_pump | 24 | The seasonal signal. |
| ev | ev | 2 | With against without. |
| ev_by_tariff | tariff | ev | 6 | Charging behaviour against how they are billed. |
Every cohort carries dailyKwh, samples, and hourly — 24 mean hourly values in kWh. Cohorts below 500 profiles are omitted entirely rather than published thin, which is why some combinations are absent.
Consumption by property type
| Property type | kWh/day | kWh/year | Profiles |
|---|---|---|---|
| Detached house | 10.59 | 3,866 | 319,763 |
| Semi-detached house | 10.01 | 3,652 | 636,663 |
| Terraced house | 9.41 | 3,434 | 559,328 |
| Flat or maisonette | 8.88 | 3,239 | 225,044 |
| Bungalow | 8.05 | 2,939 | 3,401 |
Bungalows come out below flats, which is not credible for British bungalows. There are only 3,401 of them here against 319,763 detached, so we read it as small-sample noise rather than a finding.
Consumption by month
| Month | kWh/day | Index vs annual mean |
|---|---|---|
| January | 10.65 | 1.093 |
| February | 10.59 | 1.088 |
| March | 10.18 | 1.045 |
| April | 9.68 | 0.994 |
| May | 9.17 | 0.941 |
| June | 8.88 | 0.912 |
| July | 8.89 | 0.913 |
| August | 8.98 | 0.922 |
| September | 9.33 | 0.959 |
| October | 9.89 | 1.015 |
| November | 10.36 | 1.064 |
| December | 10.69 | 1.098 |
A 20% winter-to-summer swing, which is what you would expect where heating is gas: lighting and a little more time indoors, not heat.
Heat pump households, month by month
| Month | With heat pump | Without | Difference | Seasonal only |
|---|---|---|---|---|
| January | 30.12 | 10.65 | +19.47 | +10.31 |
| February | 30.39 | 10.59 | +19.80 | +10.63 |
| March | 28.16 | 10.18 | +17.98 | +8.82 |
| April | 24.28 | 9.68 | +14.60 | +5.44 |
| May | 21.38 | 9.17 | +12.21 | +3.05 |
| June | 18.39 | 8.88 | +9.51 | +0.35 |
| July | 18.05 | 8.89 | +9.16 | +0.00 |
| August | 18.55 | 8.98 | +9.57 | +0.41 |
| September | 20.50 | 9.33 | +11.16 | +2.00 |
| October | 24.18 | 9.89 | +14.29 | +5.13 |
| November | 27.51 | 10.36 | +17.15 | +7.99 |
| December | 30.00 | 10.69 | +19.31 | +10.15 |
kWh/day. “Seasonal only” is the difference above the July floor of 9.16 kWh/day. The raw annual difference is 5,289 kWh; the seasonal part, which is the defensible estimate of space heating, is 1,945 kWh. Why that distinction matters.
Findings, including the awkward ones
- Property type is the signal; EPC band is not. EPC band moves consumption by under 0.1% within a property type, including for heat pump households where it should dominate. We treat that as a limitation of the synthetic data rather than a finding, and build nothing on it.
- The heat pump gap survives matching. Unmatched it is 15.49 kWh/day; matched on property type and EPC band it is 15.22. Under 2% of it is house shape.
- But it does not survive July. The difference is still 9.16 kWh/day in midsummer, when nothing is being heated. These are different households in ways this data does not label.
- Heat pump households peak at 04:00, not in the evening — and it is thermal, not financial. Heat pump households on a standard flat tariff, with no cheap window to chase, peak at the same hour and put 25% against 14% of their electricity into 00:00–06:00. Pre-dawn is the coldest hour, so heat loss is greatest and heat pump efficiency worst at the same moment. A smart tariff adds a few further points on top. We published the tariff explanation first and it was wrong.
- Heat pumps add 15.49 kWh/day; EVs add 13.13. The EV figure implies a high-mileage driver charging entirely at home, which is the Octopus EV population rather than the national one.
- No heat pump cohort in a flat or a bungalow cleared the sample floor. Adoption in them is close to nil here.
- Economy 7 households use 1.49× a standard-tariff household (14.27 against 9.60 kWh/day) with no low-carbon technology involved. The dataset has no storage-heating label, and this is the closest proxy it has to one.
- Only eco and standard tariffs survive the sample floor among households with no low-carbon technology. Smart and automated tariffs are effectively absent from that group, so in this data being on one is close to a proxy for owning an EV or a heat pump.
- The LSOA cluster separates households more than property type does. Cluster means span a wider range than the gap between a detached house and a flat. Where you live — or rather what sort of area you live in — looks like the stronger predictor, which is the opposite of how most benchmarks are built.
- Weekends are barely different from weekdays. Around a per cent apart, against the much larger difference most people assume.
Licence and citation
Published under CDLA-Permissive-2.0, the same licence as the source dataset. Commercial use is unrestricted and no registration is required. Attribution is requested, not legally required, and Centre for Net Zero deserve most of it — they built the dataset, we only aggregated it.
Energycosting (2026), UK domestic electricity consumption aggregates, derived from OpenSynth/Faraday v5.0 (Centre for Net Zero), CDLA-Permissive-2.0. https://www.energycosting.co.uk/data/uk-domestic-electricity
How it was derived
scripts/derive-cohorts.py streams the Faraday parquet files in batches, parses each profile, applies the conditioning-label filters for each cut and accumulates means. It reads 3,000,000 profiles and drops any cohort under 500 samples. The script is in the repository and reproduces this file exactly.
Source dataset: OpenSynth/cnz-faraday-5.0 by Centre for Net Zero, hosted under LF Energy’s OpenSynth initiative, sampled 1 March 2024 to 1 March 2025.
Limitations, stated properly
- Synthetic profiles from a generative model, not metered readings.
- Trained on Octopus Energy customers, who over-index on smart tariffs and low-carbon technology.
- Cohorts below the minimum sample size are omitted entirely rather than published thin.
- Deciles describe variation between household-days, not between households, because the dataset carries no household identifier. That spread is wider than the spread between households, so treat it as an upper bound.
- cluster_label is a k-means cluster of the household's LSOA on socio-demographic features, not a geographic region. Cluster means are published as clusters; each cluster's country mix is in clusterProfile so geography can be joined externally.
- These are means. There are no distributions, percentiles or confidence intervals here, and nothing built on this data may report one.
- No region, no floor area, no occupancy. Two of those would probably explain more than property type does.
- Electricity only. Nothing here says anything about gas.
If you use it
We would like to know. Tell us — partly out of interest, partly because knowing what people need next is how the next cut gets chosen. Region and floor area are the two most likely additions if the source labels support them.