How Lodestellar Is Using AI to Improve EPD Quality in Construction

The era of greenwashing may be coming to an end, at least in the construction sector where manufacturers are embracing a low cost tool to support data-driven transparency about their environmental impacts.

“We’ve seen that manufacturers are increasingly keen to ditch vague eco slogans and self-declared green certificates, which were never really very effective,” says Lodestellar CEO Anni Oviir. “The whole topic of sustainability is moving out of marketing and into data science where it belongs.”

Oviir leads a team of Life Cycle Assessment (LCA) experts in Estonia who created Lodestellar as a tool for checking the quality of Environmental Product Declarations (EPDs).

EPDs use scientific methodology and independent verification to reveal environmental impacts in detail for manufacturers of building products who are now racing to get them published or risk  being left out of tenders.

It’s the result of looming legislation and increasing market pressures around the world, leading to stricter limits on the carbon cost of new buildings. The European Commission has just published an EU-wide framework to standardise how building life cycle impacts are calculated, making the data more comparable across competing construction products.

It means specifiers are increasingly choosing building suppliers by comparing the hard data in their EPDs – if they have them.

When a potential supplier doesn’t have an EPD for their product, specifiers have to use assumed values that overestimate the carbon cost of producing it, so merely having an EPD can be a significant competitive advantage.

Around 50 EPDs are now published every day in the construction sector, continuing an exponential rise over recent years that is overwhelming the independent verifiers whose job it is to ensure these declarations meet all the complex requirements.

EPD verifiers must be qualified, but EPD development is often carried out by non-specialists, leading to an increasingly large bottleneck.

Creating a single EPD typically costs manufacturers in excess of €10,000 and can take more than 6 months with several back-and-forth rounds of revision during verification and publishing, and then without any certainty of a successful outcome if the declaration fails to meet the standards.

Lodestellar was created to fix that by providing EPD developers with an automated quality review that provides detailed line-by-line guidance on their draft declarations, based on all the complex requirements that will be assessed at verification.

“When we catch quality issues early, even before EPDs get submitted for verification, we can eliminate a lot of cost, hassle, and delay later in this process for manufacturers,” says Oviir.

The tool is now helping some of the biggest names in the building product industry to raise their EPD quality and get them published faster.

Other sectors are watching

Construction accounts for approximately 40% of global greenhouse emissions, prompting the sector to take a leading role in carbon accountability.

The EPD system began as a voluntary industry initiative in the Nordic region where Oviir, herself an EPD verifier and LCA lecturer, has worked to support rule harmonisation and co-author her own country’s national carbon footprinting method.

Policymakers around the world now want to cement the use of EPDs, hoping it can incentivise more innovative low carbon production, and serve as a model to other sectors.

Inside the EU, the data from EPDs will soon be used in new Digital Product Passports across the continent.

“The wider availability of EPDs benefits everyone up and down the supply chain,” says Oviir. “But for this system to function and deliver maximum value at scale, we need to raise EPD quality and also EPD literacy across the entire construction sector. Results need to be reliable and understandable, including for non-specialists.”

A linter for EPDs

Estonia is renowned for its vibrant tech scene that has produced more unicorns per capita than Silicon Valley, from Skype to Wise and Bolt.

In addition, Estonia is home to building product manufacturers that are heavily export-focused on their Nordic neighbours and hence have been keen to embrace EPDs early to maintain a competitive advantage.

Back in 2022, the Estonian Economic Ministry sought to bring the two business scenes together, uniting software specialists and construction experts for a digital construction hackathon aimed at sparking “game-changing digital tools that could be adopted across the global construction sector”.

A panel of industry experts awarded first place to Lodestellar, supporting its further development, which is now led by Tanel Teinemaa, an experienced software developer who has been behind some of Estonia’s leading tech companies and is now CTO at Lodestellar.

Teinemaa points out that even the best programmers use automated quality checks like a linter on their code before it’s published, ensuring small errors are caught before adding up to big problems later.

Lodestellar is modelled on the same principle, he says.

“Like any software developer, I’m used to running automated quality checks on my code, which ultimately enables me to deliver more value,” says Teinemaa. “The people who create EPDs are developers too, of a different kind, but with just as complex and tricky work where small details really matter.”

After extensive refinement and industry validation, including its use at LCA Support, the tool was rolled out more widely in March of this year, primarily for EPD developers, although the team says there has been interest from people involved with EPDs at all levels, including as verifiers and specifiers.

“We’ve examined years of EPD verifier feedback and the quality issues that cause delays tend to follow a pattern,” says Oviir. “It’s missing explanations, ambiguous assumptions, inconsistent wording across different sections, omitted mandatory disclosures, and poor traceability between data, assumptions, and results.”

“As a verifier, I’m repeatedly giving feedback to EPD developers that they need to clarify the source of something, explain how something specific is done, or justify some kind of assumption. This kind of quality checking can be supported by automation so that EPD professionals find the issues faster and can focus on the bigger picture, the higher level considerations, and deliver results at a larger scale.”

For EPD developers, the stakes can be high. Any errors caught during the EPD verification process add significantly to cost and time for manufacturers, who are often under tender time constraints.

Worse though, errors that slip through the verification and publication stages could open up legal risks later for manufacturers, as EPDs become more valuable in tenders and therefore also more carefully scrutinised by competition.

“We want to raise EPD quality even beyond compliance,” says Teinemaa. “For that, we’ve had to codify and teach the tool every possible requirement in the standards, while also using the team’s LCA expertise to develop our own, more extensive checklist. It’s only in the past year that the technology has really caught up with our ambitions. The tool can occasionally be a bit too nitpicky, but it hasn’t yet missed an issue that it’s been instructed to check and it has to run many hundreds of different checks.”

Lodestellar reads an EPD, whether in draft form or already published, automatically detects which standards should apply out of EN 15804+A2 or ISO 21930, and then returns a detailed, line-by-line report with clear guidance.

Teinemaa adds they are now developing internal benchmarking methods, which will be shared publicly to transparently track how the tool performs and improves over time.

Industry feedback has been encouraging

Dr Roger Singleton, CEO of Riskoa and operator of the EmVide AI LCA platform, said independent review will become increasingly important as software and AI accelerate the production of product-level climate data.

“The future of Life Cycle Assessment is faster, more digital and increasingly automated,” says Singleton. “But speed only creates value if the results can be trusted. What I like about Lodestellar is that it adds an independent quality layer, so LCA software providers are not simply marking their own homework. As AI becomes more common in EPD workflows, that separation between creation and review will be essential for credibility.”

Professor Callum Hill is an independent LCA expert who verifies EPDs and advises on sustainable construction materials. He says Lodestellar’s value is in helping reviewers catch issues that can otherwise be buried in complex EPD files.

“From a verifier’s perspective, Lodestellar is useful because it acts like an extra pair of eyes,” says Hill. “It gives a quick overview of potential issues, shows where the relevant requirement comes from, and helps highlight things that are easy to miss in complex models and documents. That is becoming more important as EPDs are increasingly used in tenders and procurement decisions, where mistakes can later become disputes.”

“I can see strong potential for this kind of tool to be used before formal verification too, helping developers fix basic issues earlier and reducing unnecessary back-and-forth.”

The team is working on new features, and is now rolling out the ability to analyse EPD background reports too.

“As we continue to develop Lodestellar, it’s important for us to keep the tool as low cost as possible and as intuitive to use as possible,” says Oviir. “EPD developers

have told us that they are learning to raise quality through more detailed feedback then they’d get even compared to when problems are flagged at the verification stage. We want to keep that support accessible so we can raise quality at scale and make a real impact.”

“We’ve so far checked hundreds of EPDs using Lodestellar and we’ve yet to find a single one, even among those already published, that doesn’t have at least some issues that could have been spotted and resolved with an automated quality check,” says Oviir. “I’m perfectly happy to admit that includes EPDs that I’ve created myself. The standards are incredibly complex so we can all benefit from an extra quality layer earlier in the process.”

While Lodestellar is built on AI, Oviir says they want to focus on showing off the value being delivered with their tool.

“The companies at the forefront of extracting value from AI are also the companies that never use it as a buzzword in their marketing. Our team is excited about the tech, but we also know that ‘AI-powered’ can sound about as meaningful as ‘eco-friendly’. In both AI and sustainability, results must now speak for themselves.”

Oviir argues that AI will have a significant impact on sustainability in ways that go far beyond the value of any one tool.

“It’s buyers, or the specifiers in the case of construction, who drive the market,” Oviir explains. “The carbon cost is just one of many complex factors that they now have to consider, alongside other critical factors like fire-proofing or acoustics, and of course the actual financial costs. But, as AI advances, buyers are more easily able to weigh up all these complex considerations across all possible products on the market.”

”So everything is moving towards credible, verified data as the advantage. It’s not just legislation and market pressures, it’s also purchasing capabilities. So ditch any unsubstantiated eco slogans and just make sure you’ve got your data properly calculated and declared.”