The EPC accuracy audit: homes use less energy than the model says

DESNZ has published the most detailed empirical audit of SAP and RdSAP yet — smart-meter and temperature data from over 1,100 homes set against their EPC models. Gas-heated homes use 16% less energy than predicted, electrically heated homes 31% less, and the findings are now feeding directly into HEM's development.

ΨMonkey

The tool behind this story — free to use in your browser.

Open ΨMonkey →

If you lodge EPCs for a living, there’s a government research report you should know about — because it quantifies, home by home and month by month, how far the modelled numbers drift from what the meter actually records. The EPC accuracy research, published by DESNZ on 26 May 2026 and 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 was commissioned to feed the EPC Action Plan and the development of the Home Energy Model — which makes it, in effect, the evidence base the next methodology will be judged against. It didn’t make much noise when it landed, but with HEM held back for final assurance, it’s exactly the kind of evidence being pored over right now.

The headline numbers

On average, gas-heated homes used 16.0% less energy than their EPC models predicted. Electrically heated homes showed a much bigger gap: 31.4% less on average, peaking at 47% in December — midwinter homes using barely half the energy the model says they should.

The 16% figure isn’t fixed, and the way it shrinks is telling. When the researchers rebuilt each model with the current RdSAP version, the actual weather during monitoring, the occupants’ real heating patterns and any improvements installed since the certificate was lodged, the gas gap narrowed to 10.9%. More than half of that correction came from a single source: work done to the home after the EPC was issued. The biggest “inaccuracy” in the EPC stock isn’t the calculation — it’s certificates that no longer describe the building. For homes whose boiler had been replaced since lodgement, updating the model shrank the gap from 22.1% to 9.5%.

Where the model drifts

Because the study broke energy use down by fuel and season rather than comparing annual totals, it could spot offsetting errors that headline figures hide:

The new-build sting

One finding runs against the grain of everything above, and it’s the one that matters most for our corner of the industry: older homes outperform their models, but new homes underperform theirs. In the existing stock, age-based fabric defaults are pessimistic — real walls, roofs and floors lose less heat than assumed. In new builds the gap reverses: measured heat loss is higher than modelled, consistent with every as-built performance study of the last decade. Design-stage assumptions about fabric, junctions and airtightness are flattering what’s actually delivered on site — which is precisely the gap the FHS’s tighter fabric standards, mandatory testing culture and (eventually) HEM’s half-hourly engine are meant to squeeze.

There’s a second new-build wrinkle: homes assessed as built with full SAP get a worse EPC when they’re later reassessed for sale under RdSAP, because the reduced-data procedure falls back on defaults in place of construction data that existed all along. The report’s fix is sensible — give RdSAP assessors access to the original as-built SAP calculation for any home built since 2008.

What it means for HEM and reformed EPCs

Four modelling assumptions are flagged for review in HEM as well as SAP: the two-zone heating assumption, seasonal electricity demand, the heat transfer coefficient assumptions, and the 0.5 ach ventilation minimum. The study also demonstrates that smart-meter energy signatures can validate a model version against measured reality — including half-hourly models, which is exactly what HEM produces. SAP ran for three decades without a standing accuracy audit; HEM could launch with one built in.

Two recommendations read as direct endorsements of the reform direction: drop primary energy as a headline certificate metric, and make EPCs dynamic — updated automatically when regulated work like a boiler swap is recorded, rather than frozen for ten years. That lines up with the four-metric reformed EPC due in the second half of 2027, and with the measured-performance direction set out in the SMETER Strategic Guide.

What to do with this

For assessors, the direction of travel is unmistakable: modelled assessments are heading into an era where they’ll be routinely checked against metered reality. The practical response is the same one we’d give for any compliance model — get the inputs right. Assessor error alone shifted predicted space and water heating by 6% on average in the resurveyed subsample, mostly through misrecorded dimensions, heating systems and wall insulation. And on the new-build side, the finding that as-built heat loss exceeds the model is a reminder that default junction values and optimistic PSI assumptions are exactly where paper performance leaks away.

If you’d rather your junctions were calculated than assumed, that’s our patch — modelled PSI values that stand up when someone eventually puts a meter on the result. Get in touch and we’ll tell you how we’d approach your scheme.

Sources: Energy Performance Certificate (EPC) accuracy research, DESNZ, 26 May 2026 · EPCs Overestimate Energy Use by 16%: What the Government’s Accuracy Research Found, HEM Guide

Got a project this affects?

ΨMonkey is free in your browser, so you can see where your own scheme lands before anyone else asks. Want it done for you instead? Send the drawings over.