Gas-heated homes use on average 16% less energy than their EPC models predict, and electrically heated homes around 31% less, according to the EPC accuracy research published by the Department for Energy Security and Net Zero (DESNZ) on 26 May 2026. The 218-page study, produced by the UCL Energy Institute with Alan Pither Limited, compared EPC-modelled energy use against smart meter and temperature data from over 1,100 homes. It is the most detailed empirical audit of SAP and RdSAP yet published, and it was commissioned explicitly to feed the EPC Action Plan and the development of the Home Energy Model.
The Headline Numbers
The gas analysis drew on 1,136 homes: 674 smart-metered homes from the Smart Energy Research Lab (SERL) Observatory and 462 homes with monitored internal temperatures from the Energy Follow-up Survey (EFUS), supported by temperature data from 200 homes in the GHG-SMETER project. The electric analysis was much smaller (43 SERL and 31 EFUS homes, excluding heat pumps), which the authors flag as a limitation.
| Gas-heated homes | Electrically heated homes | |
|---|---|---|
| Average performance gap | 16.0% less energy than modelled | 31.4% less energy than modelled |
| Worst month | Winter months | December: 47% less than modelled |
| Best month | Summer total roughly correct | July: still 20% below modelled |
| Gap after model corrections | 10.9% | Not restated (small sample) |
| Typical EPC banding effect | Mid-band | Often F and G despite lower energy use |
The 16.0% gas gap is not fixed. When the researchers rebuilt each model with the latest RdSAP version (9.94), the actual weather during monitoring, the occupants' real heating patterns, and any efficiency improvements installed since the EPC was lodged, the gap narrowed to 10.9%. More than half of that improvement came from one source: work done to the home after the certificate was issued. The single biggest “inaccuracy” in the EPC stock is not the model at all, but certificates that no longer describe the building.
Where SAP Drifts from Measured Reality
Because the study disaggregated energy use by fuel and season rather than comparing annual totals, it could identify offsetting errors that headline figures hide. Five findings stand out.
Homes are warmer than SAP assumes
Measured mean internal temperatures were higher than modelled, including in band F and G homes. EFUS data shows band G homes are on average 2.5°C colder than band A and B homes, so colder homes are real, but SAP already models a temperature reduction for inefficient homes, and the measured stock is still warmer than the calculation assumes. Underheating therefore cannot explain the performance gap. The report also found little measured support for SAP's two-zone model, which assumes the non-living area is up to 3°C cooler than the living area: in practice the two zones track each other closely in gas-heated homes.
Ventilation defaults inflate heating demand
RdSAP 2012 assumes an average heating-season air change rate of 1.02 air changes per hour for F and G rated homes, against 0.69 for A and B. Field measurements suggest real ventilation rates are often lower than modelled in both new and existing homes, and can fall below SAP's 0.5 ach minimum. The report recommends reviewing that minimum, in conjunction with indoor air quality requirements, for both SAP and HEM.
Solar gains underestimated by up to 54%
April 2021, the sunniest April on record, gave the researchers a natural experiment: a cold, bright month in which passive solar heating should show up clearly. RdSAP underpredicted the effect of solar heating by 54% against monitored data. One suspected cause is that RdSAP version 9 estimated glazing areas from property age; RdSAP 10 requires glazing to be measured, which may close part of this gap. Orientation and thermal mass effects remain open questions.
The summer fuel split is wrong
RdSAP predicts total summer delivered energy almost exactly (about 69 kWh per year per m² of floor area), but gets the composition wrong: metered summer gas use is 34% higher than modelled, while electricity use in gas-heated homes is 18.2% lower than modelled year-round. The errors cancel in a delivered-energy metric but not in a cost or primary-energy metric, which weights electricity roughly three times more heavily. Offsetting errors of this kind are why the authors caution that agreement on averages does not validate a model's assumptions.
Old homes outperform; new homes underperform
The heat-loss gap is largest in older homes with poorly insulated walls, roofs and floors, which lose less heat than modelled, suggesting the existing stock is better insulated than RdSAP's age-based defaults assume. In new homes the gap reverses: measured heat loss is higher than modelled, consistent with other empirical studies of as-built performance. Both tails matter for policy, since retrofit savings estimated from pessimistic baselines will overpromise, while new-build compliance calculations may flatter the fabric actually delivered.
The Electric Heating Problem
Electrically heated homes show a much larger gap: 31.4% on average, reaching 47% in December, meaning these homes use barely half the energy the model predicts in midwinter. The report attributes part of this to modelling difficulty (responsiveness of the heating system and zoned heating behaviour that SAP does not capture) and notes that these homes are, unusually, slightly cooler than modelled, so some underheating is likely.
The structural issue is the rating itself. The current Energy Efficiency Rating is built on fuel cost, and standard-tariff electricity is roughly four times the price of gas per unit, so electrically heated homes cluster in bands F and G even though they use about half the energy per square metre of their gas-heated equivalents. Storage-heater tariff assumptions compound this: SAP assumes some storage heater types draw 20% of their space heating from on-peak electricity, which can cost 2.8 times the off-peak rate. A cost-based single metric was always going to sit awkwardly with electrification, and this research quantifies just how awkwardly it does.
Assessors, Defaults and Stale Data
Beyond the model itself, the research examined the EPC process. Assessor error changed predicted space and water heating by 6% on average across a forensically resurveyed subsample, with larger impacts where dimensions, heating systems or wall insulation were misclassified. The report recommends better assessor training and access to supporting data such as historic EPCs, lidar height data and improvement records.
New builds get a specific warning. Homes assessed as built with full SAP receive a worse EPC when later reassessed for sale using RdSAP, because the reduced-data procedure falls back on defaults in place of the construction data that existed all along. The recommended fix is to give RdSAP assessors access to the original as-built SAP calculation for any home built since 2008, when EPCs at construction became compulsory.
What This Means for HEM
This is the empirical evidence base against which the Home Energy Model will be judged, and the report addresses HEM development directly. Four modelling assumptions are flagged for review in HEM as well as SAP: the two-zone heating assumption, seasonal electricity demand for lights and appliances, the assumptions behind the building heat transfer coefficient, and the 0.5 ach minimum ventilation rate. An appendix sets out additional input data HEM would need. None of these guarantee HEM inherits the fixes, but they are now on the record as known weaknesses with measured evidence behind them.
The validation point may matter most. The study demonstrates that energy signature methods using smart meter data can test a model version against measured reality, and notes that diurnal signatures could validate half-hourly models, which is exactly what HEM produces. The authors argue model validation should be continuous, with an annual comparison of modelled against metered energy use now feasible. SAP ran for three decades with nothing of the kind; HEM could launch with a standing accuracy audit built in. Given the delayed HEM launch was attributed to assurance and robustness, evidence of this sort is presumably in heavy demand inside DESNZ right now.
What This Means for EPC Reform
Two of the report's recommendations read as direct endorsements of the reform direction already announced. First, it argues primary energy should not be the headline metric on a certificate, since it is not comparable to metered data and distorts the picture in gas-heated homes; the reformed EPC planned for the second half of 2027 replaces the single cost-based rating with separate fabric, heating system, smart readiness and cost metrics. Second, it makes the case for dynamic EPCs that update automatically when regulated work such as a boiler replacement or insulation installation is recorded, rather than remaining frozen for ten years. The research shows this is technically feasible today using EPC register input data and NEED improvement records, subject to data-sharing permissions.
What to Watch
- HEM launch: whether the accuracy findings visibly shape the assumptions in the released version, particularly on ventilation minima and zone temperatures
- The HEM: EPC consultation response, still outstanding, which will determine how existing homes are assessed under the reformed certificate
- Dynamic EPC provisions: whether the EPC Action Plan picks up the NEED data-linking recommendation
- Follow-up research on electric heating: the authors call for targeted recruitment of under-represented home types, including heat pump homes, new builds and the least efficient stock
Milestones land on our Timeline & Status page as they are confirmed.
Frequently Asked Questions
What is the EPC performance gap?
It is the difference between the energy use an EPC model predicts and what the home actually uses at the meter. The research measured an average gap of 16.0% for gas-heated homes and around 31% for electrically heated homes, with the electric gap peaking at 47% in December. In both cases real homes used less than the model predicted.
Does this mean my EPC is wrong?
Not necessarily. EPCs were designed to compare homes under standard occupancy, not to forecast a specific household's bills. The research shows the underlying model tends to overestimate energy use, especially in older and electrically heated homes, and that many certificates are simply out of date: over half of the measurable gap reduction came from improvements made after the EPC was issued. See what the changes mean for your home.
Why do electrically heated homes get such poor EPC ratings?
Because the current rating is cost-based and standard-tariff electricity costs roughly four times as much as gas per unit. The research found electrically heated homes use about half the energy per square metre of gas-heated homes, yet often sit in bands F and G because their heating costs can be twice as high. This is a core motivation for the reformed multi-metric EPC.
Will the Home Energy Model fix the performance gap?
Partly. Several recommendations target HEM directly: reviewing the two-zone heating assumption, the minimum ventilation rate, and heat transfer coefficient assumptions. HEM's half-hourly calculation can also be validated against smart meter data in ways SAP never was. But no model fixes assessor error or out-of-date certificates, which is why the report also recommends dynamic EPC updates linked to improvement records.
Related Pages
SAP vs HEM
How the Home Energy Model differs from the SAP methodology it replaces.
The New EPC Metrics
The four-metric certificate replacing the single cost-based rating.
Why SAP Is Being Replaced
The long-standing criticisms of SAP that led to HEM's development.
Timeline & Status
Live tracker of HEM, FHS, EPC reform, and ECaaS milestones.