In short
- Good data is data whose origin, unit, period, boundary and factor can each be named on request.
- Most bad figures are not wrong. They are answers to a different question than the one you asked.
- Rate quality per line and keep the backlog. An unrated inventory cannot be improved deliberately.
Good carbon data is data where you can name, without looking anything up, five things about every line: which document it came from, what unit it was in, which twelve months it covers, which parts of the organisation it includes, and which emission factor was applied. That is the working definition, and it applies whether you produced the number or somebody handed it to you. It matters most in the second case, because a figure from a supplier, a consultant or an acquired subsidiary arrives with none of that attached unless you ask.
What are the quality criteria that actually matter?
The GHG Protocol's Scope 3 Standard names five data quality indicators in its Table 7.6: technological, temporal and geographical representativeness, completeness, and reliability. The checklist below is our own translation of those five into seven working questions: representativeness becomes three separate fit questions, and reliability becomes a document-level traceability question. Unit integrity and consistency are our own additions; the standard does not name either, but a line that fails on unit or on cross-entity treatment is exactly the kind of line the standard's indicators are meant to catch.
| Criterion | The question to ask | What a weak answer sounds like |
|---|---|---|
| Traceability | Which document did this line come from? | "It came out of the system" |
| Unit integrity | What unit was the input, and what converted it? | "It was already in tonnes" |
| Temporal fit | Which twelve months, and how old is the factor? | "Last year" |
| Geographic fit | Which country's grid or factor set applied? | "We used the standard one" |
| Technological fit | Does the factor describe this activity or a category average? | "It was the closest match" |
| Completeness | Which entities and sites are in, and which are out? | "All of them" |
| Consistency | Was every entity treated the same way? | "Each site did their own" |
None of these are about precision. They are about whether the number can be explained. A coarse figure that can be is usable. A precise one that cannot is not.
How do you judge a number a supplier hands you?
Five questions, in this order, and you can send them in one email.
What boundary does it cover? The whole company, one legal entity, one site, one product line. Suppliers frequently send a group figure in answer to a question about one factory.
Which period? With start and end dates, not a year label.
Which scopes, and for scope 2, which method? Location-based and market-based are different answers about the same electricity, and an unlabelled scope 2 figure adds to nothing.
Which factor set, from which year? A database name and a vintage means the figure is probably solid. If nobody knows, treat it as an estimate.
Who checked it, and how? Self-reported, internally reviewed, or externally assured. All three are acceptable. Not knowing which is not.
The most common failure is not a wrong calculation. It is a substitution: a product footprint answering a corporate question, an intensity figure per unit of revenue presented as an absolute total, or one scope 3 category presented as scope 3. If you receive a lot of these, our note on what customers can reasonably ask helps in both directions.
What are the tells that a figure is not what it claims?
Six, each a minute's work.
No period on the document. The most common omission, and it makes the figure unusable in a series.
A total with no boundary statement. If it does not name the entities included, somebody decided and did not tell you.
Scope 3 as one number with no category breakdown. Usually one or two categories were calculated and the rest left out silently.
A suspiciously round total. Real inventories do not land on round hundreds. Rounding a headline is fine, but ask for the underlying figure.
A fall with no project behind it. Ask what changed physically. If the answer is a method improvement or a supplier switch, it is a restatement question, not a reduction.
A figure identical to last year's. Sometimes true. More often a copy forward.
How do you judge a number produced in-house or by a consultant?
Ask for four artefacts. Any competent producer has them, and the request is normal rather than hostile.
The factor register. Every factor used, its source database, its vintage, and which lines it applied to.
The exclusion list. What was left out and why. An inventory listing no exclusions has not been examined, because every inventory excludes something.
The estimate log. Which lines are estimated, on what basis, and what would replace them.
One worked line, end to end. Pick a category yourself and ask for the walk from source document to reported tonnes, including every conversion. That single request tells you more than reading the whole report, and where the value chain dominates it is worth doing on a scope 3 category.
Does good data have to be complete data?
No, and conflating the two is what makes people distrust perfectly serviceable inventories. Completeness is a disclosure property, not a precision property. An inventory that says which categories are measured, which are estimated and which are excluded is more useful than one presenting everything at the same apparent confidence. The second looks better and tells you less.
When judging a figure, weight your scrutiny by size. A category worth a small fraction of the total does not deserve the same interrogation as the one carrying most of the mass. Reviewing proportionally is itself a mark of competence, and a tool that cannot show you the split by category makes it impossible, which is worth testing when choosing one.
How do you record quality so it is usable next year?
Rate every line as you enter it, not afterwards. Three levels are enough: measured, calculated from a physical quantity, or estimated. Store the rating with the figure rather than in a separate note, so it travels with the data.
Then keep the column nobody remembers: what would replace this line. "Supplier questionnaire", "meter reading once the sub-meter is in", "carrier report from Q2". That column is the improvement backlog, and it turns next year's data work into a prioritised list rather than a repeat of this year's scramble.
Report the mix openly. The share of your total sitting on measured data, rising each year, is a more credible signal to a customer than a total with no quality information attached.
What does Hedgehog do about data quality, and what does it not?
The platform holds more than 20,000 spend-based and activity-based factors and lets you add organisation-specific or supplier-specific CO2 data, so a line can be upgraded as better evidence arrives rather than rebuilt. Entity management across locations and sites keeps the completeness question answerable, and roles for data owners, auditors and managers record who owns a figure alongside it. An AI guide walks through setup, upload and reporting, and human GHG experts are reachable in-app. Free account, no sales call, Pro from EUR 1,200 a year.
Two limits, both directly relevant if traceability is your reason for reading this.
Factor lookup is not fully exposed. A Mid-Market customer wrote on G2 in June 2026 that they could not easily navigate to the exact number or database of the emission factor they had chosen for an item, only the name, and noted that we are fixing it. The same review, rated 3.5 out of 5, said there is no data source management feature. If a formal factor register is a hard requirement for you, ask us where that stands before you commit.
Quality still depends on what gets loaded. A Small Business customer said on G2 in August 2026 that loading the data is the challenging part while everything works once it is in. No platform improves data it was never given, and the criteria above are judgements a person makes, not settings a tool applies.
One boundary: these criteria are written for organisational footprints. Judging a product-level figure brings in functional units, allocation rules and system boundaries, which is LCA work and something we deliver as a service.
What should you do first?
Take the largest single line in your current inventory, whoever produced it, and answer the criteria questions above for that one line. If three of them defeat you, you have found this year's real work item, and it is worth more than reviewing the whole file.
Then send the five supplier questions to your two biggest suppliers and see what comes back. You can hold the answers and the evidence against the right lines in a free account, or have the review done with you as part of footprint consulting.
Sources: GHG Protocol Corporate Standard and Corporate Value Chain (Scope 3) Standard, Hedgehog platform, Hedgehog on G2. Verified 27 August 2026.
Facts on this page were last verified on 2026-09-17.
