Embodied Carbon from IFC with AI (EN 15978 / EN 15804)

Short answer: Yes. With Archi Automate (Archigrafix), an AI client can read an IFC model and produce an embodied-carbon estimate aligned to EN 15978 / EN 15804 — broken down by material and element type, with carbon hotspots and a kgCO2e/m² intensity — all in plain language. The headline tool, ifc_carbon_estimate, returns the product-stage (A1–A3) result plus an honest coverage report telling you which elements lacked geometry and which materials fell back to a generic factor. It is an estimate with full transparency, not a certified LCA.

Embodied carbon is the greenhouse-gas burden locked into a building before anyone switches on a light: the extraction, manufacture, transport and assembly of its materials. As operational energy gets cleaner, embodied carbon becomes the larger share of whole-life impact — and the part you can only influence while the design is still on the screen. The trouble is that most embodied-carbon work happens far too late, in a spreadsheet, weeks after the geometry that drove the numbers has already been frozen.

Why estimate carbon from IFC early

If your design data already lives in IFC, you are sitting on the quantities you need: walls, slabs, columns, beams, their materials and their volumes. An openBIM model is a federation-aware snapshot of the building, and that is exactly what an A1–A3 product-stage estimate is hungry for. Pulling carbon directly from IFC means you can ask the question during concept and scheme design, when swapping a material or thinning a slab is a conversation rather than a change order.

Doing it from IFC also keeps the analysis vendor-neutral. The same model exported from Revit, Archicad or any IFC-capable tool feeds the same estimate, so multi-disciplinary teams compare like with like instead of arguing about which authoring app produced the “real” number.

Ask for an estimate in plain language

You do not write a script. You connect your AI client to the model and ask, in ordinary words, for what you want: a breakdown by material and element type, the carbon hotspots, and a per-square-metre intensity. Under the hood, Archi Automate composes governed operations at runtime — in this case ifc_carbon_estimate — reads the geometry and material assignments, and returns a structured answer.

A typical request reads like this: “Estimate the embodied carbon of this IFC to EN 15978, product stage A1–A3. Break it down by material and by element type, flag the top hotspots, and give me kgCO2e per square metre using a gross floor area of 4,200 m².” You pass the gross floor area so the tool can normalise the result into the intensity figure that benchmarks and planning targets actually use.

The reply names the worst offenders first. Concrete and rebar usually dominate the substructure and frame; facade and finishes show up further down. Seeing the hotspots ranked means your first design move is the highest-leverage one, not whichever material happened to be top of mind.

Generic factors versus live Ökobaudat EPDs

The estimate can draw its emission factors from two sources, and the choice matters for how much weight you put on the number.

Offline, the tool uses generic or sample factors. That is enough to rank hotspots and sketch the order of magnitude when you are working without a connection or just exploring options quickly. Online, you can switch the source to live Ökobaudat EPDs — Environmental Product Declarations published to EN 15804+A2 — so the factors reflect real, declared product data rather than a generic stand-in. EN 15804 is the product-level standard that governs how an EPD is calculated; EN 15978 is the building-level method that aggregates those products into a whole-building result.

You can also push that data back into the model. ifc_attach_epd resolves live Ökobaudat EPDs and writes the EPD identity plus the GWP A1–A3 figures onto the relevant materials, into Pset_EnvironmentalImpactValues. The change is previewable before it lands, so the model becomes self-documenting: anyone who opens it later can see which declared product backs each material. Beyond the product stage, you can extend the estimate to whole-life modules A4–D when you need transport, use, end-of-life and beyond-the-boundary effects in the picture.

Be honest: it is an estimate, not a certified LCA

This is the part most tools quietly skip, and the part that earns trust. ifc_carbon_estimate produces an estimate with a coverage report — it is not a certified or audited life-cycle assessment, and it does not pretend to be one. The coverage report is the honest receipt: it tells you which elements were excluded because they lacked usable geometry, and which materials could not be matched to a real factor and fell back to a generic one.

That distinction is what keeps you out of trouble. A clean-looking total over a thin coverage report is worth less than a slightly messier total that covers 95% of the model with declared data. For anything approaching EPD-grade rigour, you need the live Ökobaudat source — which means working online — and you still treat the output as a design-stage estimate, not a compliance certificate. Offline, with generic factors, the tool falls back gracefully and tells you it has done so.

Improving the accuracy of your estimate

The coverage report is also your to-do list. If an element is missing geometry, fix the model so it has volume. If a material fell back to a generic factor, assign it a proper material so it can be matched — or attach a specific EPD with ifc_attach_epd. Each fix narrows the gap between estimate and reality, and because the work happens in the IFC itself, the next estimate simply gets better without re-importing anything.

This is where carbon work joins the rest of your model-quality loop. The same governed operations that let you run a quantity takeoff from IFC feed the carbon estimate — bad quantities make bad carbon. Cleaning up materials and missing data with an AI-assisted edit and remediation pass directly improves coverage, and a consistent classification from a Uniclass or bSDD classification step makes the by-type breakdown far more meaningful.

Vendor-neutral by design

The openBIM connector is headless: it works on IFC from any authoring tool, with no CAD application or licence required, and it is federation-aware so it can reason across linked models. Guardrails apply throughout — the connector defaults to read-only, model edits such as writing EPDs run through a Preview step before you Allow changes, every operation is audited, and nothing is auto-saved. So even when you let it attach EPDs, you stay in control of what changes and when.

If you want to understand the plumbing, the broader picture of MCP for IFC and openBIM explains how an AI client talks to your models, and you can pair carbon checks with IDS validation so a model has to be both correct and low-carbon before it moves on. To get an assistant wired up to your toolchain, see how to connect Claude to your AEC tools.

Get started

Archi Automate is one Windows installer that connects your AI client to Revit 2025–2027 (Autodesk), Rhino 8 (McNeel), Archicad 29 (Graphisoft) and openBIM (IFC, IDS, BCF) via MCP. There is a 14-day trial with no key required. Estimate embodied carbon from your own IFC and read the coverage report for yourself — explore AI for AEC to begin.