A pale green bottle and material samples sit beside a laptop showing stylised lifecycle icons.
Illustrative workspace, not a tested product model, functioning software demonstration or environmental result.
Diagram text description

A bright synthetic scene shows a bottle, jars and dishes of pellets and fragments, wood and speckled blocks, loose samples, a notebook and a hand pointing at a laptop. Its screen contains a stylised bottle with arrows and lifecycle icons. No readable method, impact value, dataset identity or certification appears. The scene cannot establish material composition, recycled content, real supplier data or an executed assessment.

Choose product life cycle assessment software by asking it to reproduce a defined product study, then explain a deliberately controlled change. A screen full of impact categories is less useful than a model whose quantities, data sources and calculation settings can be inspected. Before comparing subscriptions, establish the study question, product unit and boundary; identify the available data; and agree what a pilot and its reviewer output must demonstrate.

The practical test in this guide separates two questions: what happens when one background dataset changes, and what happens when the impact-assessment method version changes? Neither experiment changes the product itself. Keeping those questions apart exposes hidden defaults that a feature demonstration can miss.

Use the proposed bottle study below as a pattern, not a calculated benchmark. No vendor has been tested and no environmental results are supplied. The purchase decision should follow observable model fit, traceable inputs and reproducible output—not an invented league table of tools.

Decide whether the job is full LCA or a carbon footprint

A product carbon footprint addresses climate change. A multi-impact life cycle assessment can examine additional environmental categories, provided appropriate methods and data support them. That distinction determines what the software must actually calculate. A carbon dashboard does not become a multi-impact assessment because its menu mentions water or materials.

ISO 14067's public description limits its assessment to climate change; ISO 14044's overview covers the wider LCA process, including inventory, impact assessment, interpretation and review. Those public descriptions are not substitutes for the complete standards or the rules of a particular reporting programme.

Translate the distinction into an acceptance question. If the buyer needs a climate-only study, ask for the exact climate indicator and a transparent inventory. If the decision includes possible water or resource trade-offs, ask the modeller to demonstrate those separate categories and explain their limitations. A supplier unable to demonstrate a required category may still offer useful carbon accounting, but it has not met that broader brief. The appropriate scope follows the decision, not the largest number of dashboard tiles.

Write the study brief before buying a tool

A workable brief names the product variant, the service being assessed, the quantity needed to provide it, and the recipient of the output. It also says which lifecycle stages are included. A manufacturing-only result cannot answer a question about washing, repeated use or final disposal without additional modelling.

Fictional study brief: Bottle B500, revision A, is a 500 mL uninsulated bottle with a steel body, polypropylene cap and silicone seal. For this pilot, the functional unit is provision of a container capable of holding 500 mL of drinking water for 24 hours without leakage. The assumed reference flow is one conforming bottle plus its carton. Performance is a test assumption, not a verified product claim.

The assessment boundary is cradle to factory gate: material production, incoming transport, component manufacture, assembly and packaging. Distribution, washing, repeated-use benefits and end of life are excluded. This is therefore a partial lifecycle study tied to that function, not a lifetime comparison with disposable packaging. Assembly is assumed to take place in China using 2025 foreground records. The audience is the procurement team and an independent technical reviewer; the decision is whether the software preserves the model and explains controlled changes.

A later public comparison would need an appropriately expanded and reviewed study. Do not quietly broaden this brief while judging the pilot.

Test whether the model represents the real product

The bill of materials must become connected processes, not just a spreadsheet attachment. For the fictional baseline, one finished bottle contains 240 g steel, 20 g polypropylene and 5 g silicone, with a 35 g carton. The steel-forming process takes 250 g steel and produces the 240 g body plus 10 g scrap. Assembly electricity is 0.30 kWh per bottle. These are illustrative foreground inputs, not measured manufacturing data or a complete ready-to-calculate inventory.

Connect the steel supplier record to steel production and forming; connect cap and seal records to their component processes; connect incoming transport and electricity to the relevant stages. Other process energy, transport distances and treatment routes must be completed and frozen before execution. Leaving them unrecorded is a hold, not permission to treat them as zero.

Allocation also belongs in the model specification. For this test, assume a sole finished-product output, no credit for the 10 g scrap, and a documented cut-off treatment of that scrap. Preserve the selected background datasets' own system model and allocation settings. These are pilot assumptions requiring review, not a universal recycling prescription. Changing them during a dataset experiment would introduce another explanation for the difference.

Check dataset fit and primary evidence

A database name alone does not establish data fit. Review geography, represented year, technology, process coverage, unit and licensing conditions for the selected record. A newer dataset can still describe the wrong production route. Supplier-specific evidence is valuable only where its measurement and boundary match the input being represented.

Consider a fictional trace: supplier file S-07 revision 2 records the steel input; selection record DS-STEEL-A identifies its background production dataset; model node BODY-01 uses the 250 g input; output R-A0 identifies that node, the database release and the assessment method. Those labels are local example identifiers, not real database entries. The reviewer must be able to travel from the eventual result back through the node and selection rationale to the retained source file.

Keep the actual dataset identifier, release, system model and record checksum in the execution manifest. Preserve any unit conversion and explain whether a supplier value replaces an activity quantity, a process inventory or a whole impact result. They are not interchangeable inputs. If licensing prevents redistribution, agree a legitimate reviewer-access route before procurement. A screenshot of a total does not resolve either the provenance question or access to the underlying record.

Look beyond a single carbon number

Inspect named categories and indicators rather than a single unexplained sustainability score. The following are examples of separate characterised outputs, not calculated bottle results. Their units describe different environmental mechanisms and must not be added as though they were the same quantity.

Category exampleIndicator unit to identifyInterpretation boundary
Climate changekg CO2 equivalent per functional unitState the method, version and time horizon; not a direct measurement of every environmental effect.
Water useMethod-defined scarcity-weighted water equivalent, such as m3-world equivalent in EFNot simply litres withdrawn at the factory. Retain the exact implementation's indicator name and unit.
Resource use, minerals and metalskg antimony equivalent in the relevant EF indicatorA characterisation indicator, not kilograms of metal physically present in the bottle.

Table 1, shown one row at a time. Every field is retained.

Row 1

Category example
Climate change
Indicator unit to identify
kg CO2 equivalent per functional unit
Interpretation boundary
State the method, version and time horizon; not a direct measurement of every environmental effect.

Row 2

Category example
Water use
Indicator unit to identify
Method-defined scarcity-weighted water equivalent, such as m3-world equivalent in EF
Interpretation boundary
Not simply litres withdrawn at the factory. Retain the exact implementation's indicator name and unit.

Row 3

Category example
Resource use, minerals and metals
Indicator unit to identify
kg antimony equivalent in the relevant EF indicator
Interpretation boundary
A characterisation indicator, not kilograms of metal physically present in the bottle.

For the central test, select the climate-change category under EF 3.1 and retain its exact installed implementation and GWP100 definition. The European Commission publishes an EF 3.1 reference package; that availability does not prove a particular vendor implements it correctly. Separately verify the software's method package and elementary-flow mapping. A method update can change characterisation while the underlying product remains identical. Treat that as a methodological difference requiring explanation, not evidence that the bottle became greener.

Make revisions and review possible

Revisions should create inspectable branches rather than replace the only surviving model. Keep baseline A0, its source files, settings and eventual report together. A report without the inputs and method package cannot demonstrate that the original calculation remains reproducible after an update.

In a fictional subsidiary BOM test, revision B replaces the 20 g polypropylene cap with a 22 g cap of the same specified polymer. Retain the original cap record and record the engineering-change reason, date and author. Create model B1 and link it to A0; keep the original report R-A0 and produce a separate report only after recalculation and review. Any changed process demand or transport quantity must be recorded as part of that product revision, not concealed in a dataset substitution.

This is deliberately a different experiment from the fixed-product trials below. Do not combine B1 with a new electricity dataset and a method upgrade, then describe the resulting difference as a cap improvement. The useful software behaviour is preservation of the old model and an explicit account of the new one—not merely the presence of an undo button.

Run a representative product pilot

Run the pilot with the people who will own the model, not solely with a vendor's demonstration operator. The following sequence is a proposed workflow; no import, calculation or independent review has been executed for this article.

  1. Import the representative SKU and reconcile component quantities and units against the source files. Resolve rejects rather than silently dropping rows.
  2. Define the function, reference flow and included stages. Confirm the partial lifecycle boundary with the reviewer.
  3. Select background data, document allocation and freeze the manifest. Identify incomplete supplier evidence before calculation.
  4. Calculate the baseline only after the inputs and method mapping are complete; retain logs, warnings and disaggregated category output.
  5. Run the controlled experiments separately, then the subsidiary BOM revision. Preserve every predecessor.
  6. Give an independent reviewer the model, sources and reports, with legitimate access to licensed data. Record questions and the disposition of each.

For a missing-supplier case, remove the seal supplier's process evidence in a separate branch. Retain the 5 g seal quantity; do not remove the seal from the inventory. A documented generic silicone proxy may support a provisional sensitivity run, but its technology and geographic mismatch remain explicit. Hold any claim that the result is supplier-specific until suitable evidence arrives. Observe the effort needed to make that distinction and whether it survives export.

What software cannot promise by itself

Calculation, independent review and external communication are different checkpoints. A successful run establishes that the software produced an output under recorded settings. It does not establish that every input is representative, that the study satisfies a programme, or that a comparative claim is authorised.

The GHG Protocol Product Standard page was consulted on 31 October 2026 (UTC). Its guidance explains that product rules provide additional specificity for comparisons. That is a useful procurement warning: an attractive reporting template is not evidence that the proposed comparison has the same function, boundary, allocation and data basis. Obtain the relevant programme rules and qualified review for the actual intended use.

After technical review, a separate communication decision must check what the result supports, what is excluded, and how limitations will reach the audience. The bottle pilot excludes use and disposal, so it cannot demonstrate whole-life superiority or the benefit of repeated refills. Require the exported report to carry that boundary. A software licence, a vendor's assurance badge or a calculation marked complete cannot replace those decisions.

Test one variable at a time

This proposed acceptance test is unexecuted. It supplies a controlled design and decision rules, not calculated impacts, vendor scores or evidence that any tool has passed. Use the fictional B500 revision-A inventory and functional unit above. Before execution, complete the missing process and transport inputs and replace the example dataset aliases with licensed records and exact identifiers.

Freeze a baseline manifest A0: one conforming bottle and carton per defined function; cradle-to-gate boundary; the stated BOM and 0.30 kWh assembly electricity; sole-product allocation and scrap treatment; one background-library release and system model; and the installed EF 3.1 method package. Record the software build, solver settings, elementary-flow mappings, category identifiers, package checksums and export settings. An unspecified item blocks execution rather than becoming a hidden default.

TrialOnly permitted changeObservable acceptance evidenceHold condition
A0 repeatNone: reopen and rerun the frozen model.Same manifest and reproducible unrounded climate output within a reviewer-agreed numerical tolerance.An automatic update, altered mapping or unexplained output difference prevents reproduction.
D1 datasetReplace only the assembly-electricity dataset with an alternative in the same frozen library release. Keep its 0.30 kWh quantity and EF 3.1 package fixed.Dataset identifiers and geographic/technology rationale differ; foreground quantities, function, boundary and allocation do not.Another input or method changes, the alternative cannot serve the same exchange, or linked-provider changes are not disclosed.
M1 methodReturn to A0, then replace only EF 3.1 with the specified EF 3.0 implementation.Identical inventory and background records; separately named climate category, GWP100, kg CO2 equivalent and exact method-package versions.Flow mappings or category definitions cannot be reconciled, or migration changes the inventory.
B1 subsidiaryChange the cap BOM from 20 g to 22 g in a separate product branch.A0 and its report survive; B1's changed quantities and recalculation trail are explicit.The old model is overwritten or this product change is mixed into D1 or M1.
S1 subsidiaryWithhold seal-supplier process evidence in a separate data-gap branch.The 5 g seal remains; proxy identity, limitations and unresolved supplier evidence are visible.Missing evidence is treated as zero impact or the proxy is labelled supplier-specific.

Table 2, shown one row at a time. Every field is retained.

Row 1

Trial
A0 repeat
Only permitted change
None: reopen and rerun the frozen model.
Observable acceptance evidence
Same manifest and reproducible unrounded climate output within a reviewer-agreed numerical tolerance.
Hold condition
An automatic update, altered mapping or unexplained output difference prevents reproduction.

Row 2

Trial
D1 dataset
Only permitted change
Replace only the assembly-electricity dataset with an alternative in the same frozen library release. Keep its 0.30 kWh quantity and EF 3.1 package fixed.
Observable acceptance evidence
Dataset identifiers and geographic/technology rationale differ; foreground quantities, function, boundary and allocation do not.
Hold condition
Another input or method changes, the alternative cannot serve the same exchange, or linked-provider changes are not disclosed.

Row 3

Trial
M1 method
Only permitted change
Return to A0, then replace only EF 3.1 with the specified EF 3.0 implementation.
Observable acceptance evidence
Identical inventory and background records; separately named climate category, GWP100, kg CO2 equivalent and exact method-package versions.
Hold condition
Flow mappings or category definitions cannot be reconciled, or migration changes the inventory.

Row 4

Trial
B1 subsidiary
Only permitted change
Change the cap BOM from 20 g to 22 g in a separate product branch.
Observable acceptance evidence
A0 and its report survive; B1's changed quantities and recalculation trail are explicit.
Hold condition
The old model is overwritten or this product change is mixed into D1 or M1.

Row 5

Trial
S1 subsidiary
Only permitted change
Withhold seal-supplier process evidence in a separate data-gap branch.
Observable acceptance evidence
The 5 g seal remains; proxy identity, limitations and unresolved supplier evidence are visible.
Hold condition
Missing evidence is treated as zero impact or the proxy is labelled supplier-specific.

Two independent experiments start from baseline A0: D1 changes only a dataset; M1 changes only the method.
Proposed, unexecuted tests: D1 and M1 each start from A0. They are not sequential changes and neither alters the product.
Diagram text description

The baseline branches separately to an electricity-dataset substitution with the assessment method fixed, and an EF method-version substitution with inventory and datasets fixed. The product functional unit, foreground quantities, boundary and allocation remain constant across both experiments. The adjacent matrix defines provenance, acceptance and hold criteria. No arrow represents a calculated improvement or a completed vendor test.

For D1, changing one selected dataset can legitimately change its linked background supply chain. Export those dependencies: the controlled variable is the chosen dataset, not a promise that every background exchange stays identical. If a whole database is upgraded, that is a different experiment and must not be called D1.

For M1, request characterised climate-change output in kg CO2 equivalent per the same functional unit, alongside each exact method/category identifier and version. Keep normalisation and weighting off in both runs. A common unit alone does not prove equivalent characterisation or comparable flow coverage. Review missing or newly mapped flows and their consequences before interpreting any difference; do not subtract incomparable category outputs to claim an environmental improvement.

Before running A0, the reviewer must agree and record the numerical tolerance and its rationale for the chosen solver and export precision. No universal tolerance is asserted here. Acceptance means the model is reproducible, provenance survives export and differences can be attributed within the declared design. An actual procurement record must contain the eventual outputs, warnings and reviewer decision. Until then every trial is pending, not a pass.

Keep the evidence handover separate from calculation

Keep product evidence preparation, model calculation and review separate when deciding what to buy. The evidence pack should contain the SKU and BOM revisions, supplier-source versions, missing-data register, study brief and the execution manifest. Preparing that pack does not itself calculate multi-impact results or certify a study.

Ask the modeller to return both the readable report and enough legitimately accessible model detail for a reviewer to reproduce it. Resolve who owns updates, who can restore a previous version, and who decides whether a changed dataset or method requires a new report. Those responsibilities matter after the demonstration ends.

Choose the tool that can carry out the required study and expose its limitations. If the controlled test cannot yet be run, record the missing evidence and defer that purchase conclusion. A transparent hold is more useful than a confident ranking built on a different product, a hidden default or an unexecuted calculation.