LCA by Artificial Intelligence:
a new opportunity in the early design stage

In the quest of using AI in LCA, there are not many successful examples. Most of the attempts are highly inaccurate. The reason is that there is a lot of misinformation at the internet, and internet is the source of information for AI. Companies tend to exclude subsystems to get the lowest score, and NGOs tend to exaggerate (e,g, incorrect calculations on LULUC, biogenic CO2, and recycling). So, just picking data from publications at the internet, does not make sense.

However, AI is good at searching and good at performing algorithms. At this moment there is one AI tool that is really good at LCA: Claude Anthropic.
It is important here to realize that an LCA is based on the combination of a BOM (plus required energy and transport) and tables with Environmental Impact Indicators (Midpoint and Endpoint). When Anthropic is provided with the BOM and indicator tables as input, and when the system description is OK (with functional or declared unit unit), the accuracy of its LCA calculation is rather good.

The LCI tables for scope 1, 2, and 3, like Idemat,  are free available at the internet. However, the BOM is often not known in early design stages. Anthropic is able to fill this information gap, since AI is rather good in the estimation of the BOM, based on internet information of similar products.

The key to good results in Anthropic is the question: it must contain (1) the functional unit or the declared unit (2) the indicator (3) the LCI table for Scope 1, 2, 1nd 3 data (4) if available; general information on the BOM.

Case 1
“what is the carbon footprint of a cup of coffee 40 ml Nespresso made by a Krups Innisia machine, 5 year, two cups per day (cradle to grave). Take data from attached Idemat_2026RevA.xlsx for the coffee machine and the electricity, and use Agribalyse for agriculture”
As a result Anthropic will come back with complete answer that is quite detailed. The accuracy depends on the question (a better system description and a better LCI database result in better quality of the answer).

Content

Note 1: Anthropic regards Idemat as best available data for engineering & design, but when you don’t add the .xlsx it doesn’t come with the most recent data. Agribalyse is advised for food in Europe
Note 2: A Bill of Materials of Nespresso coffee machines is not available, so this information has been created by Anthropic (!!!).
Note 3: The Anthropic LCA tool highlights the most important variables of the system.

The Anthropic LCA tool Case 1

These type of LCAs have to be checked carefully: never take the outcome for granted. An issue in this tipical example is the carbon footprint of Arabica coffee: it is set to 6 kg CO2 / kg coffee, but a check in Ecoinvent reveals that this comes from a broad range of 2 – 16 kg CO2 / kg coffee (!) Here Anthropic can not be blamed: it is fact of life that LCA in agriculture normally shows problematic wide ranges.

When something is wrong, it is easy to ask for a re-calculation.

When the BOM (plus energy, and transport) are known, this type of tools in AI don’t save much time in comparison to the Excel tools that are available at this website. The reason is that the AI calculations have to be checked. But in the early design phase, it is a lot of fun to locate the hotspots in this way.

The activity flowsheet that is given below provides a good understanding what is calculated step by step:

The Anthropic LCA tool Case 2

question: “I want to make an LCA, calculating the carbon footprint of a bottle of 1 liter of glass, PET, or PE, using Idemat 2026RevA data as attached, plus the recycling content = recycling rates (RR) of the attached table_for_end-of-life RevC2 (4).xlsx for The Netherlands and Europe. Build this as an interactive HTML tool with sliders and buttons”
Note: the last sentence is needed to prevent a simple output.

The Anthropic LCA analysis Case 3

question: “what is the carbon footprint of a 1 liter bottle of glass? Give the answer in the form of a bar chart for different end-of-life scenarios: for flint, green, brown, and mixed colours. Take the data of the Idemat 2026RevA.xlsx for carbon footprint, and the recycling content = recycling rates (RR) of the table_for_end-of-life RevC2 (4).xlsx for The Netherlands and Europe”
Note: the percentages in the calculation are recycled content (RiR). The end-of-life scenario is for the remainder glass waste.