Early-Stage LCA: How to Assess Environmental Impact Before a Process Is Designed

A standard LCA answers the question too late

A conventional Life Cycle Assessment is built for things that already exist. It wants measured inventories, a defined plant, real supplier data. By the time a new process can supply all that, the route is chosen, the equipment is specified and the footprint is locked in. The LCA becomes an accurate description of a decision nobody can reverse.

That is the paradox at the heart of eco-design for industrial processes: influence is highest when information is lowest. Early-stage LCA is the discipline of working inside that gap. Its goal is not a certified number. It is an environmental estimate good enough, early enough, to change which process you build.

This guide is written for process chemists, R&D leads and sustainability teams who need to make that call. It covers:

• the vocabulary: ex-ante, prospective and screening LCA, and when each applies
• where the data comes from when there is no plant
• how to scale lab results up without fooling yourself
• how deep an assessment should go at each development stage
• how to make decisions that hold up under uncertainty
• how AI and autonomous labs change the data available

For the bigger picture of AI across process development, see how DecaOS works from the lab to the plant.

Ex-ante, prospective, screening: what the terms mean

The literature uses several overlapping labels. They are worth separating, because each one answers a different question.

Prospective LCA is defined by Arvidsson and colleagues as follows: "An LCA is prospective when the (emerging) technology studied is in an early phase of development (e.g., small-scale production), but the technology is modeled at a future, more-developed phase (e.g., large-scale production)." The key word is modelled. You are not assessing the flask. You are assessing the plant the flask implies.

Ex-ante LCA is used in much the same sense. It is an assessment made before a technology is deployed, to guide its development. Researchers at Leiden use the term for the work of projecting an emerging technology to industrial scale.

Screening LCA is about depth rather than timing. It is a fast, simplified assessment: a few impact categories, generic background data, and a focus on finding hotspots. It is the right tool for comparing route options or solvents in a single afternoon.

In practice these are not alternatives. They are the same assessment, run at increasing depth as the process matures. The rest of this guide treats them that way.

The data problem: building an inventory when there is no plant

Every LCA rests on an inventory: the mass and energy flows in and out of each step. For an early-stage process, most of those flows have never been measured at the relevant scale. Arvidsson and colleagues call data scarcity a central challenge and encourage practitioners to combine scientific articles, patents, expert interviews, unpublished lab results and process simulations.

A review of upscaling in ex-ante LCA by Tsoy and colleagues found four families of data estimation methods across the studies it examined. Here is how each one fits a process team's work:

Three habits make the inventory far more useful:

1. Measure what the lab can measure. Masses in and out, solvent volumes, reaction time and temperature, and work-up and purification steps. These cost almost nothing to log and are the hardest data to reconstruct later.
2. Separate what you measured from what you estimated. Tag every flow with its source. When results come back, you will know which numbers to trust and which to improve first.
3. Model the auxiliaries from day one. Solvent recovery, drying, cleaning and utilities rarely appear in a lab procedure, yet they often dominate the footprint at scale.

Upscaling: from flask to plant without fooling yourself

Lab data cannot be fed into an LCA as is. A 2025 consensus paper on LCA of emerging technologies puts it bluntly: "Lab-scale setups differ fundamentally from industrial systems; thus, scaling inventory data is essential for plausible comparison across TRLs." Lab equipment is small, oversized for its batch, poorly insulated and run without heat integration. Assessed as is, an innovative route can look worse than a mature incumbent simply because it hasn't been engineered yet.

The review by Tsoy and colleagues proposes a three-step upscaling framework:

1. Define the projected technology scenario. Describe the future plant: scale, mode (batch or continuous), location and energy mix, and its year of operation.
2. Draw the projected flowchart. Lay out every unit operation the plant will need, including the ones the lab skipped: solvent recovery, purification, waste treatment.
3. Estimate projected data. Fill each flow with the best available method from the table above.

Where upscaling goes wrong

Shortcut correction factors. Applying a flat factor to lab energy use is tempting. The consensus paper warns that common correction factors "(e.g., 0.1–0.3 lab-to-industry energy use) risk oversimplifying complex systems". A heating step and a cryogenic separation do not scale the same way.

Assuming scale always helps. Bigger is not automatically greener. In a prospective LCA of graphene production at several scales, Arvidsson and Molander found the three scenarios had similar impacts, because electricity in one upstream step dominated every impact category. The only large reduction came from a process change, not from scale.

Leaving out what the lab never needed. The same consensus paper notes that early studies can also underestimate impacts when they omit overheads that only appear at full scale. In chemicals, the usual suspects are solvent losses, drying, recycle loops and waste treatment.

The lesson: upscale by engineering, not by multiplying. Model each unit operation the way it will run at scale, and test whether your conclusion survives the plausible range.

A stage-gate approach: match the depth of the LCA to the decision

The right level of assessment depends on what has to be decided next. An early-stage LCA should be just deep enough to make the next stage-gate decision, and no deeper. Spending three months on a full inventory for a route you might drop next week is as wasteful as skipping the assessment altogether.

Each gate hands more data and less design freedom to the next. Even so, every stage produces a footprint decision, not just a footprint number.

Two practical rules keep this ladder working:

• Make the environmental criterion part of the gate. A route that fails its screening LCA should need an explicit justification to proceed, just as a route with a safety concern would. The eco-design framework applies the same rule at every stage.
• Carry the inventory forward. Each stage should refine the previous stage's model, not start a new one. The route-scouting estimate becomes the skeleton of the pilot-stage prospective LCA, and that becomes the basis for the final ISO study.

Uncertainty: making decisions you can defend

An early-stage LCA will never be precise. That is fine. The test is not whether the number is right. It is whether the decision would change if the number were wrong.

The research consensus gives a clear direction on how to handle this.

Use scenario ranges, not point estimates. Arvidsson and colleagues advise using a single predictive scenario only when one future is clearly more likely: "If no such particularly likely development exists, it is advisable to apply scenario ranges." For a process, the scenarios usually cover scale, energy mix, solvent recovery rate and yield at scale.

Worry about structure before statistics. Monte Carlo analysis is useful, but at low maturity the biggest uncertainty is rarely a parameter's spread. The consensus paper notes that structural and scenario uncertainty often outweigh parametric uncertainty, so bounding or scenario analysis may be more informative at low TRLs. Whether you will recover the solvent matters more than the third decimal of its emission factor.

Focus on what drives the result. The same paper recommends that discussion "focus on factors driving environmental impacts or hot-spot type analyses". In practice, one or two flows usually account for most of the footprint. Find them, and put your effort into narrowing those.

Compare fairly against the incumbent. Incumbents, the paper notes, "represent systems refined through years of optimization, while emerging technologies may only exist at bench or pilot scale." Compare your projected plant against the incumbent's plant, not your flask against its plant.

A practical way to report this is a decision statement, not a single number. For example:

That gives leadership something they can act on, and it shows exactly which assumption to test next.

How AI and autonomous labs close the data gap

Every problem above comes back to one constraint: early-stage processes produce too little structured data, too slowly. That is what is changing.

Every experiment becomes an inventory record. In a traditional lab, mass and energy data sit in notebooks, if they are recorded at all. An autonomous lab logs reagents, solvents, temperatures, times and outcomes in a structured format on every run. Inventory data that used to be rebuilt at the end of a project now builds up from the very first experiment.

Impact becomes an optimisation target. Once footprint can be estimated from each run, it can sit next to yield and cost in the optimiser's objectives. The system then explores conditions that improve both, rather than optimising yield first and checking the footprint afterwards. Our eco-design guide explains how multi-objective optimisation maps those trade-offs.

Models fill the gaps between experiments. Where data is missing, models trained across chemical and process data can estimate properties, yields at new conditions and the likely behaviour of a unit operation at scale. They do not replace measurement. They tell you which experiment would most reduce your uncertainty, so the lab spends its time where the LCA most needs it.

Upscaling and assessment share one data model. When the lab record, the projected flowsheet and the LCA use the same structured data, the three-step upscaling framework stops being a separate study. It becomes a view on data you already hold. You can see that loop from the lab to the plant in our 3D Industries view.

This is what DecaPlants is building: a process superintelligence platform and autonomous lab that embeds eco-design throughout the scientific method. The aim is simple. Environmental impact should be known at the moment a decision is made, not reconstructed after it.

Frequently asked questions

Assess early, decide early

Early-stage LCA asks teams to accept something uncomfortable: a number with wide error bars, produced months before anyone would normally ask for it. The alternative is worse. It is a precise number about a footprint nobody can change anymore.

The method is now well established. Model the process at its future scale. Build the inventory from whatever the lab and the literature can provide. Upscale by engineering, not by multiplying. Report decisions as ranges that hold across scenarios. What has held teams back is not method but data, and that is the part autonomous labs and process AI are changing.

To see how this fits into the wider development workflow, read our guide to eco-design for industrial processes, explore DecaOS on our Tech page, or join our Slack community of scientists and engineers working on sustainable processes.

Sources

1. R. Arvidsson et al., Environmental Assessment of Emerging Technologies: Recommendations for Prospective LCA, Journal of Industrial Ecology 22(6), 1286–1294 (2018)
2. N. Tsoy, B. Steubing, C. van der Giesen and J. Guinée, Upscaling methods used in ex ante life cycle assessment of emerging technologies: a review, International Journal of Life Cycle Assessment 25(9), 1680–1692 (2020)
3. R. Woods-Robinson et al., Controversy and consensus: common ground and best practices for life cycle assessment of emerging technologies, arXiv (2025)
4. R. Arvidsson and S. Molander, Prospective Life Cycle Assessment of Epitaxial Graphene Production at Different Manufacturing Scales and Maturity, Journal of Industrial Ecology 21(5), 1153–1164 (2017)
5. Global CO2 Initiative, Techno-Economic Assessment & Life Cycle Assessment Guidelines for CO2 Utilization, Version 2.0 (includes a chapter on assessing early-stage technologies)