AI raises the stakes for product environmental data
For years, product environmental data has largely travelled in one direction: out of the factory. A customer asks for an embodied-carbon figure, procurement wants an environmental product declaration, or a regulator needs evidence. The information is assembled, checked, submitted and often disappears into another reporting process.
That model starts to look inadequate once AI is expected to influence engineering, sourcing, procurement and production planning. Environmental information then stops being simply something a manufacturer reports and becomes something software may use to recommend a material, compare suppliers or identify a lower-carbon design route. The question is no longer just whether the number exists, but whether a machine can understand what it means and whether the business can trust the decision made from it.
BEAMA, the trade association for electro-technical and energy systems manufacturers, is pressing for greater consistency without creating a costly new compliance regime. Its research suggests manufacturers are already doing much of the hard work; the challenge is making the information reusable enough to support more than reporting.
Reporting data is not yet decision data
Requests for environmental information are multiplying, but they do not arrive in a common language. Tasha Lyth, Sustainable Business Manager at BEAMA, sees manufacturers being asked for embodied carbon, EPDs, TM65 assessments and other environmental indicators, often in different formats.
“The request can be different depending on who is asking for it,” she says. “Your customer may ask for something in a prescribed format, or they may send you a calculator, or ask for the data in an Excel spreadsheet. That variety means the requests are not uniform in terms of how people are asking for that information.”
BEAMA’s 2026-member research found that 88 per cent of manufacturers already provide environmental data, while 80 per cent can provide it through a self-certified declaration or format. The difficulty is that two figures can appear comparable while having been produced using different boundaries, assumptions or methodologies.
Lyth says the value lies in the analytical work behind the final number. “There is so much inherent value in conducting that life-cycle assessment, learning the information about your products, where it is coming from and what optimization opportunities you have,” she adds. “The analysis can show how you could tweak product design or change material selection to support further decarbonization.”
That becomes important for AI. A sustainability specialist can interrogate a document, understand a methodology and challenge an odd result. An AI system working across thousands of product records may not know that two values were calculated on different bases unless those differences are explicitly represented in the data.
Digital product records could change the architecture
Digital Product Passports in the EU and the UK Government’s exploration of Digital Product Records could move environmental information away from repeated document requests toward persistent digital product information.
Patricia Massey, Head of Digital and Technical Standard at BEAMA, sees an opportunity to reduce duplication rather than build another reporting silo. “Ideally, information should be collected once and then be capable of being used for different regulatory, business and customer needs,” she says. “The opportunity is to move from product information being something that is repeatedly requested and reported, to information that can be maintained digitally and reused throughout the product lifecycle.”
That does not require every manufacturer to place all product information in one central database. Environmental data may already exist across product-development, assessment, supply-chain and operational systems.
“A DPP or DPR does not necessarily have to mean that manufacturers put all of their product information into one central system,” Massey adds. “The focus should be on having agreed ways of describing and exchanging information so that different systems can work together.”
For AI, this is potentially more useful than forced centralization. A model does not necessarily need one repository, but it does need common definitions and reliable relationships between data. It must know that an environmental attribute belongs to the same component referenced in engineering, procurement and production systems, even when those systems store it differently.
Massey also stresses that responsibility and verification must travel with the information. “The important thing is that any future framework is clear about what information is needed, who is responsible for it and how it should be checked. The approach also needs to be proportionate – not every piece of information will require the same level of verification.”
AI increases the cost of inconsistency
Environmental data has traditionally been forgiving of fragmentation because people have acted as the translation layer. A sustainability manager can recognize equivalent requests, while an engineer can challenge a result that looks inconsistent with the physical product.
AI removes some of that friction, but it can also remove those informal checks. If data becomes an input to automated recommendations, inconsistent terminology or assessment boundaries can be reproduced at machine speed.
Lyth describes the current problem as cumulative rather than one single failure point. “Some manufacturers may still be going through their own digitization or digital transformation journey,” she explains. “Others may have complex supply chains. Then there is the format that the information needs to be given in. It is the cumulative effect. Once you overcome one part, you may move into the next area that also has challenges coupled with it.”
BEAMA is calling for common methodologies supported by standards, product-category rules and product-specific rules. Where assumptions have to be made about factors such as use phase or end of life, alignment makes it more likely that different manufacturers are calculating on a sufficiently consistent basis.
For future AI systems, those rules could become guardrails. The model may identify patterns, compare options and surface anomalies, but it still needs to know which comparisons are legitimate.
Verification cannot become the price of participation
Greater trust normally leads to calls for greater verification. BEAMA’s concern is that a framework designed to improve environmental information could instead make it prohibitively expensive to produce.
Its survey puts the average reported cost of a self-certified assessment at around £3,200 per product. Manufacturers in the survey averaged 490 products, while only 20 per cent could currently provide data in the form of an EPD. BEAMA estimates that making product-level EPDs a market-access requirement could create an illustrative additional cost of around £1.58 million per manufacturer.
Lyth argues for a more proportionate approach. “If you have followed the standards or you have followed a clear common methodology, and you are able to evidence that in a way that is reasonable, then potentially the additional cost of third-party verification would not be necessary or certainly should not be a minimum market-entry requirement.”
That principle matters for digitalization as much as sustainability. SMEs cannot afford every new use case to require another bespoke data exercise or assurance process. Reusable information should allow the same verified evidence to support customer requests, compliance and internal decisions.
AI may eventually help reduce that burden. Once fields, methodologies and evidence requirements are clearly defined, software can help identify gaps, flag inconsistencies and reuse verified data across multiple workflows. The sequence matters: AI can accelerate a well-defined process, but it cannot compensate for an industry that has not agreed what the data represents.
Environmental intelligence belongs inside the workflow
The larger opportunity is to move product environmental data closer to the decisions that create environmental impact in the first place. Product design, material selection and sourcing determine much of a product’s footprint long before a sustainability report is prepared.
Lyth sees much of the value emerging before the reporting stage. “The knowledge you can gain from completing a life-cycle inventory and analysis to generate environmental product metrics gives manufacturers a huge amount of information,” she says. “This can not only fulfil customer requests but also support their own internal decision-making and product design to support decarbonisation.”
Once that information becomes digitally accessible, AI can help engineers and procurement teams explore more combinations than they could manually. Materials could be assessed against cost, availability, performance and embodied emissions before resources are committed to redesign.
None of this means handing environmental decisions to an algorithm. It means giving people a better way to interrogate a growing body of information at the point where a decision can still change the outcome.
Lyth’s description of success is therefore less about producing better reports than making the data actionable. “Three to five years after a good framework has been introduced, you would want confidence in the data and clear decision-making,” she continues. “Manufacturers should be able to use it to support innovation or product design, while the person procuring the product has actionable data that can inform their decision.”
That is the more important AI story. Product environmental data is becoming part of the information infrastructure on which future industrial decisions will depend. The manufacturers that benefit most will not simply collect more of it. They will make it comparable enough to trust, structured enough to exchange and useful enough to return to the people and systems deciding what gets designed, bought and manufactured next.
