The Fine Print in EPDs

For: architects, sustainability teams and manufacturers|Category: EPDs and data quality|Read time: about 9 minutes

Two Environmental Product Declarations can report almost the same A1 to A3 carbon figure and still deserve very different levels of trust. The headline number is only as good as the assumptions holding it up, and most of those assumptions sit in methodology pages, footnotes and data quality tables that few specifiers ever read.

That matters more every year. EPDs now feed embodied carbon targets, tender requirements, green building credits and procurement rules. A number that was once a marketing extra is becoming a contractual figure, which makes the fine print behind it a commercial question as much as a technical one.

At 2050 Materials, we work with EPD data every day, collecting, structuring and comparing declarations from manufacturers so that designers and specifiers can use them in real decisions. The pattern is consistent: the products that look best on paper are not always the ones with the strongest evidence behind them.

This article walks through the assumptions that point to a high or low confidence EPD, then looks closely at three areas where reading carefully pays off most: compliance with the governing standards, how purchased electricity is modelled, and whether the EPD describes your exact product, a family of products or an industry average.

How Much Was Measured, and How Much Was Borrowed?

Confidence comes down to one question: how much of the result was measured at the factory, and how much was estimated or taken from a database? Even a well made EPD carries uncertainty. The Construction Products Association puts the minimum uncertainty in an A1 to A3 GWP total at roughly ±10%, so small gaps between two products rarely mean much on their own. As we argued in Garbage In, Garbage Out, any carbon calculation is only as good as the data going into it, and that starts with the EPD.

None of these on its own makes an EPD unreliable. What they do is tell you how much weight a number can carry. An EPD with metered plant data, a recent reference year and a clear sensitivity analysis can support a hard comparison. One built mostly on generic data is closer to an informed estimate. And while this article focuses on carbon, the same assumptions shape every other indicator in the declaration, from acidification to resource use, which we unpacked in What Else Is in an EPD Beyond Carbon, Anyway?

Compliance Is the First Filter

An EPD is only comparable with another if both follow the same rules, so compliance is the baseline for trust rather than a formality. The standards landscape is crowded, as we explored in Comparing EPD Data Standards, but five things are worth checking on the cover page and in the verification statement before reading any results.

If any of these are missing, treat the numbers as indicative only, and avoid using them in a formal comparison or a tender requirement.

The Electricity Question

The physical electricity a factory takes from the grid is the same however it is bought. An EPD can still model it as renewable by applying energy attribute certificates: REGOs in the UK, Guarantees of Origin in the EU, I‑RECs internationally and RECs in the US.

Applying them can take the electricity emission factor from roughly 0.2 to 0.5 kg CO₂e per kWh down to near zero. For electricity intensive products such as aluminium, glass, mineral wool, tiles or engineered panels, that can move the A1 to A3 result materially, sometimes more than any change to the product itself.

At under £1 per MWh, a plant can make its electricity look renewable for a fraction of a penny per kWh, without changing anything about where its power comes from.

On site generation is the strongest form of renewable claim. H.G. Matthews’ new 300kW solar PV array at The Brickworks, Bellingdon.

The contrast is clearest when a manufacturer invests in generation itself. In H.G. Matthews: Beyond the EPD, we looked at how the brickmaker’s new 300kW solar PV array at Bellingdon cuts its reliance on grid power directly, and how improvements like this can feed into future EPD updates. That is a very different claim from buying certificates at under £1 per MWh.

Programme rules are starting to respond. Practice still varies between programme operators, and some do not allow Guarantees of Origin at all. EPD Norge now requires new EPDs to report foreground electricity both ways: market based, with certificates, and location based, using the grid average. They must also state the GWP per kWh used and name the certificates and their validity period. If certificates lapse and GWP worsens by more than 10%, the EPD becomes invalid.

When you read an EPD for an electricity intensive product, look for five things: the electricity factor in kg CO₂e per kWh, whether certificates were applied and of which type, whether they are bundled or bought separately, a residual mix rather than a grid average for any uncovered electricity so the same renewable output is not counted twice, and ideally a location based result alongside the market based one.

Industry, Product Group or Product Specific?

The type of EPD tells you how closely its numbers match the product that actually arrives on site. All three types are legitimate. They simply answer different questions.

Industry EPDs are useful early on, but the best and worst producers inside the average can sit well apart. Under the International EPD System, a sector EPD must report any GWP variation above 10% between participating companies with an explanation, and its results cannot be presented as representative of any one manufacturer.

Product group EPDs can declare a volume weighted average, a representative product or a worst case result, and the cover page should say which. For ISO 21930 compliance, results must not vary by more than 10% between the included products; otherwise each indicator that varies by more than 10% must be declared, along with the GWP range. A worst case declaration is conservative. An average can flatter the heavier products in the range.

Product specific EPDs carry the most weight, but only if you know which plant your product comes from. Results averaged across several sites cover a broader spread of performance than single site results, so comparing one with the other can mislead.

Our own data shows why this matters. When we mapped EPDs for steel products across countries, the spread was wide not only between countries but within them. An industry average for any one of those markets would hide exactly the difference a specifier is trying to find.

Who Should Read the Fine Print

This is not only a job for LCA specialists.

Recap and What to Do About It

A low carbon figure is only as useful as the assumptions behind it. The strongest EPDs are compliant with current standards, independently verified, built on recent plant data, specific to the product and plant you are buying from, and open about how electricity was modelled.

Before using an EPD to choose a product or set a target, ask three questions. Does it follow the same rules as the EPDs you are comparing it with? Does it describe this product, from this plant? And if its electricity is near zero carbon, is that backed by new renewable supply or by cheap certificates?

Checking all of this by hand, one PDF at a time, does not scale across a whole project. That is why the 2050 Materials Sustainability Data API collects and structures this fine print for more than 200,000 EPDs: the governing standard, PCR, programme operator and verification, reference year and validity, and whether an EPD is industry average or product specific. Instead of comparing headline figures, you can filter EPDs by the quality of the data behind them, directly inside your own tools, dashboards and carbon calculators.

References:Construction Products Association, EPD uncertainty: a technical review (2024); EPD Norge, Guidelines on use of electricity and biogas certificates (2025); Cornwall Insight, The rise and fall of REGO prices; One Click LCA, Rules for EPDs of multiple products; One Click LCA, LEED v4.1 EPD credits.