For an EPBD life-cycle GWP calculation, product-specific Environmental Product Declarations (EPDs) are the expected data source once a product has been selected, while generic datasets are appropriate earlier, when it has not. Both are legitimate inputs. The distinction that matters is not which is “better” — it is whether the data you used matches the decision you were making.
Getting this wrong is expensive in a specific way: teams either spend money chasing EPDs for products nobody has specified yet, or build a baseline on generic data and discover at certification that it does not survive scrutiny.
The short version
- An EPD is a verified, product-specific declaration from a manufacturer, following EN 15804. It describes that product from that plant.
- A generic dataset gives a representative figure for a material class — “concrete, C30/37” — not a particular product.
- Early design: generic data is the honest choice, because the product genuinely is not chosen yet.
- Certification: EPDs are what a verified assessment expects where products are known.
- Generic figures are usually conservative on purpose, so a switch to real EPDs often moves the number down.
- The larger early-stage error is almost never the carbon factor. It is the quantity.
What is an EPD, exactly?
A third-party-verified document in which a manufacturer declares the environmental impact of a specific product, prepared under EN 15804 and typically covering modules A1–A3 as a minimum.
| EPD | Generic dataset | |
|---|---|---|
| Describes | One product from one plant | A material class |
| Verified | Independently, under EN 15804 | No — industry average or literature |
| Available | Only once the product exists | Always |
| Expires | Typically after 5 years | No |
| Right for | Certification, specified products | Early comparison, undecided products |
| Bias | Reflects the actual product | Usually conservative |
Two properties make EPDs authoritative. They are product-specific — this cement from this plant, not cement in general. And they are independently verified, so the manufacturer is not marking its own homework.
Two properties make them awkward early. They only exist once a product exists, and they expire, typically after five years. An EPD referenced in a 2026 assessment may need refreshing before the building completes.
What is a generic dataset, and is it inferior?
A generic dataset gives a representative value for a material category, drawn from industry averages or literature. The ICE database is the best-known example in this space.
It is not inferior. It is answering a different question.
At concept stage the question is “roughly how much carbon does a concrete frame commit us to, versus timber?” No EPD answers that, because there is no product yet. A generic factor answers it well enough to make the decision — and the decision is what matters at that point (CLT vs concrete vs steel).
Generic values also tend to be deliberately conservative. That has a useful consequence: a baseline built on generic data usually improves when real EPDs replace it. Planning against a conservative figure is the right direction to be wrong in.
So which does the EPBD require?
Delegated Regulation (EU) 2026/52 requires the calculation to follow the relevant parts of EN 15978. That sets how a building is assessed rather than mandating a single data source for every input. In practice the expectation follows the product: where a product is specified, use its EPD; where it is not, use representative data with the assumption stated.
Delegated Regulation (EU) 2026/52 exists precisely to make these choices consistent across Member States, so that two assessors do not reach different answers through different data conventions (full EPBD timeline).
What this rules out is quiet substitution — using a favourable generic figure, or a best-in-class EPD, for a product that will not be the one installed. That is not a data-quality question. It is a disclosure question, and the figure ends up on a certificate.
Does better data actually change the answer?
Less than most teams expect, and less than the quantity does.
Swap a generic concrete factor for a specific EPD and the figure typically moves by a margin that matters for certification and rarely for a design decision. Miss a slab, or model the facade at the wrong thickness, and the figure is wrong by an order more.
This is why early-stage effort belongs on quantities and boundaries rather than on precision in factors. A carefully sourced EPD applied to a wrong volume is a precise wrong answer — which is more dangerous than an approximate right one, because it looks trustworthy.
What does a sensible data strategy look like?
Concept. Generic data, full building, honest error bars. The output is a decision — frame, retention, envelope strategy — not a certificate (pre-assessment material intelligence).
Developed design. Generic data for the undecided, EPDs where products are locked. Now the mixed sources need recording, because someone will ask.
Technical design and certification. EPDs for specified products, generic only where genuinely unavoidable and flagged as such.
Throughout. Record the source and date of every factor. When an EPD expires or a product is substituted at procurement, you want to know which figures move — not rebuild the model.
Frequently asked questions
Can I use a competitor’s EPD as a proxy? For early estimating, as a stand-in for a class of product, that is reasonable if you say so. For a certified assessment of a specified product, no.
Do EPDs expire? Yes, typically five years. On a long programme, check validity before certification rather than after.
What if no EPD exists for a product? Use representative data and record it as an assumption. Non-availability is a legitimate reason; silence about it is not.
Is an EPD always lower-carbon than the generic figure? No. Generic values are often conservative, so EPDs frequently come in lower — but a below-average product will declare a higher number. That is the system working.
Does data source affect the module boundary? No, and don’t conflate them. Boundary is what you counted; data source is what you counted it with. Both must be stated. See modules A to C.
Where this leaves you
Data quality is a project-stage question, not a moral one. Generic data early is correct practice; generic data at certification, where a product is known, is not.
Elementa works from ICE-based generic data at pre-assessment and records a match confidence against every element, so it is visible which figures are indicative and which are grounded — and the baseline hands over cleanly when EPDs arrive.
Early figures are indicative and directional. They inform decisions; they do not replace a verified life-cycle assessment.
Sources: European Commission — calculation framework for new building life-cycle GWP; Delegated Regulation (EU) 2026/52 (EUR-Lex). Verified August 2026.