AI for Climate and Sustainability
Most sustainability numbers are estimates presented as measurements, and the distinction is about to matter more than the number.
The hard part of a carbon footprint is not the arithmetic; it is that most of the inputs are estimated, and assurance is coming for the estimates.
AI for climate and sustainability covers emissions data engineering and calculation, supplier and value chain data extraction, activity data recovery, physical climate risk and scenario analysis, target tracking, and the methodology documentation that assurance requires.
Estimated versus measured, which is the whole question
Most organisations' largest emissions categories are calculated from spend multiplied by an emission factor, which is a reasonable starting estimate and a poor basis for a decision or a claim.
- Spend-based estimates cannot show improvement. If you buy the same thing more cheaply, your calculated footprint falls, which is a measurement artefact rather than a reduction.
- Activity data is the upgrade path. Tonnes, kilometres, kilowatt-hours and units replace currency, and the number becomes responsive to what you actually do.
- Label every figure with its method. Measured, supplier-specific, activity-based or spend-based — carried through to the reported number rather than buried in an appendix.
- Prioritise by materiality and leverage. Improving data quality on a category you cannot influence is effort spent for a more precise number rather than a better outcome.
- Assurance examines the method. Limited assurance was deferred to 1 July 2027 for those in scope, and what an assurer tests is the basis of the figure rather than the figure.
A footprint that cannot move is a report, not a management tool
The test we apply to an emissions dataset is whether a genuine operational improvement would show up in it. If switching a supplier, changing a transport mode or reducing material use would leave the calculated figure unchanged because everything is spend-based, the dataset is a reporting artefact and not a management instrument. Fixing that on the categories the business can actually influence is worth more than extending coverage into categories it cannot.
Supplier data, and the cap that changed the approach
The escalating questionnaire has stopped working
Under the revised reporting rules, undertakings with 1,000 employees or fewer may decline requests that go beyond the voluntary standard, which the Commission was to define by 19 July 2026. Strategies that assumed suppliers must answer need rebuilding.
Derive first, ask second
Transaction records, transport documents, certification data and remote sensing establish a great deal without a questionnaire, and reserving supplier requests for what genuinely cannot be derived improves response rates on the things that matter.
Extraction from what suppliers do send
Suppliers send invoices, specifications, certificates and reports as documents. Extracting structured data from those is usually a larger available gain than obtaining new disclosures.
Response rates are a data quality input
A programme with a 40% supplier response rate is producing a footprint that is 60% assumption, and that ratio belongs in the reported methodology rather than in a footnote. See data engineering.
Physical risk and scenario analysis
| Analysis | Fit | Note |
|---|---|---|
| Asset-level physical hazard exposure | Strong | Flood, heat, drought, wind and fire against located assets |
| Supply chain location exposure | Strong | Requires supplier locations, which is the usual blocker |
| Scenario analysis | Good | Directional; the uncertainty is in the scenarios, not the modelling |
| Emissions calculation and consolidation | Strong | The core data engineering task |
| Activity data recovery | Strong | Replacing spend proxies where the business can act |
| Target tracking and decomposition | Strong | Separating real reduction from divestment and accounting change |
| Supplier document extraction | Strong | Larger available gain than new disclosure requests |
| Claim and greenwashing risk checking | Good | Consistency between claims and the evidence behind them |
Decompose your reduction before you publish it
A reported emissions reduction can come from genuine operational change, from divestment or closure, from a methodology or emission factor update, from improved data quality, or from currency and volume effects. These have very different meanings and are routinely reported as one number. Decomposing the movement into those components is a short piece of analysis that makes internal decisions better and external claims defensible, and organisations that do it are notably less exposed when someone examines the trend.
How an engagement runs
Method transparency first, because it determines what the numbers can be used for.
Scope and data quality assessment
What proportion of the footprint is measured, activity-based and spend-based, by category.
Data engineering
Activity data recovered where the business can influence the outcome, with method labels carried through.
Build
Supplier document extraction, or physical risk exposure against located assets.
Trial and documentation
Methodology documented to the standard assurance will examine, with decomposition tested.
Operation
Factors and methods updated deliberately, movements decomposed each reporting cycle.
What you receive
Numbers that move when the business changes, and a method that survives assurance.
Data quality assessment
What share of the footprint is measured, activity-based and spend-based, by category.
Activity data recovery
Spend proxies replaced where the business can actually influence the outcome.
Method labelling
Every figure carrying its basis through to the reported number.
Supplier document extraction
Structured data from the invoices, certificates and reports suppliers already send.
Physical risk exposure
Hazard exposure against located assets and, where possible, supplier locations.
Movement decomposition
Real reduction separated from divestment, methodology change and data improvement.
Is this the right starting point?
Worth being direct. There are situations in climate and sustainability where custom AI work is the wrong spend, and those are listed rather than buried.
Worth doing if
- Most of the footprint is spend-based and cannot respond to operational change.
- Supplier data collection relies on questionnaires with poor response rates.
- Supplier documents arrive as PDFs and are re-keyed or ignored.
- Reported reductions have never been decomposed into their causes.
- Assurance is approaching and the methodology is not documented to that standard.
Do something else if
- You want a lower reported number rather than a more defensible one.
- Activity data cannot be obtained or derived for the categories that matter.
- Asset and supplier locations are unavailable and physical risk is the objective.
- Method transparency is unwelcome because it would reveal how much is estimated.
Frequently asked questions
Marked up with FAQPage schema so these answers can surface directly in search results and inside AI assistant responses.
Why does spend-based data matter so much?
Because a footprint calculated from spend cannot show improvement. If you buy the same goods more cheaply the calculated figure falls, and if input prices rise it increases — neither of which reflects anything you did. That makes the dataset a reporting artefact rather than a management instrument. The test we apply is whether a genuine operational improvement, like switching supplier or transport mode, would show up in the number. Where it would not, replacing spend proxies with activity data on the categories the business can influence is the project.
Our suppliers will not answer our questionnaire. What do we do?
Derive what you can and reserve requests for what you genuinely cannot, because the escalating questionnaire has stopped being available. Under the revised reporting rules, undertakings with 1,000 employees or fewer may decline requests beyond the voluntary standard the Commission was to define by 19 July 2026. Transaction records, transport documents, certification data and remote sensing establish a great deal without asking. And extracting structured data from the invoices, specifications and certificates suppliers already send is usually a larger available gain than obtaining new disclosures.
What will assurance actually examine?
The basis of the figure rather than the figure. Limited assurance was deferred to 1 July 2027 for those in scope and the reasonable assurance requirement was removed, but what an assurer tests is the methodology, the data lineage and the consistency of application — which is why carrying a method label through to the reported number matters more than precision. A footprint where every category can be traced to its basis is straightforward to assure; one where estimation is buried in an appendix is not, regardless of how carefully it was calculated.
Did the Omnibus mean we can stop?
Only if you are genuinely out of scope, and that is now a specific test: CSRD applies above €450 million net turnover and 1,000 employees, with the phased waves removed, and CS3D above €1.5 billion turnover and 5,000 employees from 2029. Organisations that built for the old scope can scale back deliberately. The position we see most often is a company still in scope that paused when the simplification was announced — which is the worst outcome, because the deadline moved and the underlying data problem did not.
How should we report a reduction?
Decomposed. A reported reduction can come from operational change, divestment or closure, a methodology or emission factor update, improved data quality, or currency and volume effects, and these are routinely combined into one headline. Separating them is a short analysis that makes internal decisions much better and external claims far more defensible. Organisations that publish a decomposed movement are notably less exposed when a journalist, an investor or a regulator examines the trend.
Related verticals
Organisations in climate and sustainability usually share data, buyers or regulators with these. All fourteen are listed on the Agriculture, Food & Environment page.
AI for ESG Reporting
Data lineage, framework mapping and assurance evidence — with the post-Omnibus scope stated precisely.
Read more →AI for Carbon Markets
Remote sensing MRV, project screening and permanence analysis — with methodology chosen for certification, not for volume.
Read more →AI for Forestry
Remote sensing inventory, harvest planning and deforestation evidence — with EUDR applying to timber first.
Read more →Tell us what the problem looks like.
Thirty minutes, no charge, no deck. We will tell you whether this is an AI problem, a data problem, or a process problem — and we will say when the honest answer is to buy something rather than build it.