Estimates in GRI Reporting

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Few first-year sustainability reports are built entirely from measured, system-generated data. Scope 3 emissions may rely on spend or supplier proxies; workforce totals may require estimates where local systems do not align; supply-chain coverage may be extrapolated from samples; and biodiversity assessments often use secondary or modelled datasets before site-specific ecological surveys are complete. Treating all such information as prohibited would make many disclosures impossible. Treating every approximate number as acceptable would make the report unreliable.

The real reporting decision is therefore not 'estimate or no estimate'. It is whether the proposed estimate is fit for the decision and disclosure, whether the remaining uncertainty could change the reader's conclusion, and whether the organisation can explain and reproduce the method. A transparent estimate is a controlled reporting method. An unexplained number is only a data gap disguised as precision.

GRI 1 does not establish a universal numerical error threshold or a single hierarchy of acceptable estimates. Instead, it sets reporting principles that control the judgement. Under the Accuracy principle, the organisation reports information that is correct and sufficiently detailed for an assessment of its impacts. The accompanying guidance asks the organisation to identify estimated data and explain the underlying assumptions, techniques and limitations. Under Verifiability, the organisation should retain original sources, evidence supporting assumptions and calculations, internal controls, and clear explanations of uncertainty.

Comparability adds a time dimension. Methods and assumptions should be applied consistently. Where a new method, definition, acquisition, disposal or correction materially changes earlier information, GRI 2-4 requires the organisation to report the restatement, explain why it was made and describe the effect. Completeness prevents an estimate from being used to hide an omitted entity, location or impact that is necessary to understand the organisation's performance.

An estimate should be accepted because it can support the intended information need, not simply because it produces a number. The reporting team should first define the exact requirement, population, boundary, period and decision. It can then test the estimate against the criteria below. A weak result does not always mean omission: the method may be improved, the disclosure may be disaggregated, or a range may be more honest than a point estimate.

Figure 1. Decision logic for measured data, a usable estimate or a permitted reason for omission. This is an LRA implementation aid; the relevant disclosure requirement and GRI omission permissions must still be checked.

The method is unrelated to the disclosure or impact and would mislead readers about what has been measured.

The uncovered population could contain significant impacts, but its size or characteristics are unknown.

Reasonable alternative assumptions produce results that would change a material conclusion and the uncertainty cannot be communicated meaningfully.

The calculation cannot be reconstructed because source data, factors, transformations or approvals have not been retained.

A legal, contractual or scientific judgement is being replaced by a generic numerical proxy outside the reporting team's competence.

The estimate is used to support a claim of achievement, reduction or effectiveness that the evidence cannot demonstrate.

A reported estimate should be managed as a controlled data point, not as a footnote added at the end of drafting. The record should allow a reviewer to trace the published value to its source population, method, assumptions, uncertainty, controls and future improvement. The same record also supports consistent treatment in the next reporting cycle.

Figure 2. Evidence stack behind a reported estimate. Each layer should be visible in the internal record even when only the material methodology and limitation are published.

The public explanation should be proportionate to the estimate's significance and uncertainty. A routine low-risk estimate may need one clear methodology sentence. A major Scope 3, workforce, supply-chain or biodiversity estimate may require the population, coverage, factor or model source, assumptions, known limitations, sensitivity and improvement plan. The aim is not to publish every spreadsheet detail. It is to give information users enough context to understand what the value represents and what it does not represent.

Uncertainty should be managed in proportion to decision sensitivity. Not every estimate needs a formal confidence interval. However, the reporting team should understand how the result moves when the most judgemental inputs change. Sensitivity analysis can reveal that a seemingly minor assumption drives the result, that a ranking of impacts is unstable, or that a published reduction is within the range of methodological uncertainty.

Improving an estimate is not merely a data-project objective. It can change comparability. Before replacing a proxy or model, the team should assess whether historical information can and should be recalculated on the new basis. If a change materially alters earlier reported information, the report should explain the reason, the effect and the treatment of comparatives under GRI 2-4. A lower emissions or incident figure after a methodology change must not be presented as operational improvement unless the evidence supports that conclusion.

Define the target state: measured, supplier-specific, site-specific, more representative secondary data, or a better model.

Prioritise estimates that are large, decision-sensitive, linked to targets or public claims, or likely to enter assurance scope.

Assign an owner, milestone and data-quality criterion rather than a generic promise to 'improve data'.

Test the new method in parallel with the old method where possible and quantify the effect.

Apply a documented restatement criterion and obtain approval before changing the published trend.

Update the methodology note, source register, calculation file, Content Index reference and AI-ready record together.

The reporting team first defines the three separate populations rather than applying one blanket 'estimated data' label. For purchased-goods emissions, it uses spend-based factors for the acquisitions and supplier-specific physical data for strategic suppliers. For non-employee workers, it applies the estimate permission and rounding guidance in GRI 2-8, supported by contractor rosters and invoice records. For biodiversity screening, it uses a named geospatial dataset to locate potential high-risk sourcing areas and explicitly avoids describing the model as a site assessment.

The team calculates the estimated share of each total, performs sensitivity analysis on the largest Scope 3 factor, reconciles worker estimates to contractor spend and access records, and records the model resolution for biodiversity. It discloses the limitations next to the affected metrics and assigns separate improvement milestones: supplier activity data within twelve months, a harmonised contractor register by the next year end, and field assessment for two high-risk sourcing regions. The estimate records make clear that the three methods answer different information needs and carry different uncertainty.

Why this works: it identifies the affected figure, population, method, reason, factor basis, uncertainty, control, improvement action and comparative treatment. It does not imply that the estimate is exact, externally assured or evidence of emissions reduction.

Modelled biodiversity data can be used when it is fit for the reporting purpose and is identified transparently rather than presented as measured site fact. Explain the model, source, resolution, assumptions, coverage and limitations, retain reproducible evidence and review controls, and improve or withhold conclusions where uncertainty could materially change the reader’s understanding.