How do I choose the right factor database for a mixed-country portfolio?

Prioritise geographic specificity. Start with local government sources, such as ADEME for France or DEFRA for UK activity data. For electricity, use IEA grid-average factors if national data isn't available. And if you don't have activity-based Scope 3 spend data, use EXIOBASE for multi-regional input-output modelling. Whatever mix you use, match the factor year to your reporting period, check that the units line up, and document your source choices so they're ready for audit.

When is spend-based data acceptable instead of activity data?

Spend-based data works when you can’t get more detailed activity data. It’s a practical way to screen broad Scope 3 emissions, especially for purchased goods, services, and capital goods, or in cases where physical records just aren’t available. That said, figures from databases such as EXIOBASE are best treated as proxies. They come with more uncertainty than process-specific data or supplier-provided data. In plain terms, they’re a fallback, not a first choice. Use them by default only when primary data can’t be obtained.

How can I avoid year, unit and version errors in one study?

Make sure your activity data and emission factors match on four points: the exact material or activity, the unit of measurement, the energy basis, and the reporting year. Use the factor set for the year you’re reporting on, not the year the report is put together or sent in. If your study covers more than one year, use the matching factor set for each year. For every factor, record the source, version, and date. And when it comes to electricity, don’t mix location-based figures with market-based figures.

DEFRA, ADEME or ecoinvent: which factor database to use

If I need a fast answer: I match the database to the asset’s country, the data I have, and the level of detail I need. For UK activity data, I start with DESNZ/DEFRA. For France, ADEME. For US assets, EPA. For country electricity factors, IEA. For product or process detail, ecoinvent. And if I only have spend data, I use EXIOBASE as a first-pass estimate.

That choice can shift results more than people expect. For example, the 2025 DESNZ grid factor fell by about 15% versus 2024, from 0.207 to 0.177 kg CO₂e/kWh. And ADEME’s 2025 monetary factors are about 40% lower than older values. So if I pick the wrong source or mix years, the output can move a lot.

Here’s the short version:

DESNZ/DEFRA: best for UK fuel, electricity, travel and waste data
ADEME: best for French assets, including BEGES-style work
EPA: best for US grid and US supply-chain estimates
IEA: best for multi-country electricity factors
ecoinvent: best for product footprints and manufacturing routes
EXIOBASE: best when I only have general ledger or spend data

What matters most is simple:

Use national sources first where possible
Use activity data before spend data
Keep one version and one GWP basis across the study
Check units carefully: kg vs tonnes, kWh vs MWh, passenger-km vs vehicle-km
Avoid broad rows when a more specific factor exists

Carbon Factor Database Comparison: DEFRA vs ADEME vs ecoinvent vs EPA vs IEA vs EXIOBASE

Quick Comparison

Database	Best used for	Input needed	Main watch-out
DESNZ/DEFRA	UK Scope 1 and 2, travel, waste	kWh, litres, km	UK-only; check GCV vs NCV
ADEME	French assets and reporting	kWh, litres, km, sometimes spend	Version changes can shift results
EPA	US assets	kWh, fuel, spend	AR4 vs AR5 year mismatch
IEA	Country grid electricity	kWh by country	Paid access for full data
ecoinvent	Product and process analysis	Bill of materials, process data	Licence cost and model choice
EXIOBASE	Early Scope 3 screening	£ or € spend	Too coarse for final numbers

So my rule is simple: follow the data room, not habit. If the file has meter reads and fuel volumes, I use activity factors. If it has process inputs, I use ecoinvent. If it only has spend, I use EXIOBASE and label it as screening only.

How the main databases compare

These six sources do different jobs. The right choice depends on geography, the kind of data you have, and whether you need a rough screening view or numbers fit for reporting. So the question isn't which database wins overall. It's which one fits the data room.

DESNZ/DEFRA and ADEME: UK and French activity data

DESNZ

DESNZ/DEFRA is the default choice for UK activity-based reporting. It's public, free, and aligned to SECR and UK SRS S2.

France follows a similar pattern, but the reporting rules and factor library aren't the same.

ADEME Base Empreinte is France's national reference. It now brings together the former Base Carbone, Base IMPACTS and Agribalyse datasets. One point matters a lot here: ADEME's 2025 monetary factors are about 40% lower than earlier versions. If you're using an older spend-based model, it can overstate French assets.

EPA and IEA: US operations and country grid factors

EPA

EPA splits the job in two. eGRID covers US grid factors, while Supply Chain factors support spend-based Scope 3 analysis. EPA's 2023 factors use AR4; 2024 and 2025 use AR5, so don't mix years unless you re-base first [3].

US grid factors and global grid factors sound close, but they answer different needs.

IEA is the country-grid reference for multi-country portfolios and scenario work. Full dataset access is paid.

ecoinvent and EXIOBASE: product-level and spend-based analysis

ecoinvent

ecoinvent fits product-level and manufacturing due diligence. But there is a catch: it needs a commercial licence and LCA expertise.

EXIOBASE fits early-stage Scope 3 screening when spend data is all you have. That's useful at the start, but it's still a screening tool. It's too coarse for final reporting.

Use the table below as a selection filter, not a league table.

Database	Geographic Focus	Method Type	Licence Model	Update Pattern	Main Limitations
DESNZ/DEFRA	United Kingdom	Activity-based / combustion	Public / free	Annual	UK-specific; limited global applicability
ADEME	France	LCA-based (ISO 14040/44)	Public / free	Ongoing	French-specific; check version and validity status
US EPA	United States	Grid and supply-chain factors	Public / free	Annual	Limited to US grid and broad sectors
IEA	International	Energy statistics	Licensed / paid	Annual	Full dataset access requires paid licence
ecoinvent	Global	Granular LCI	Licensed / paid	Periodic	Requires LCA expertise and licence management
EXIOBASE	Multi-regional	EEIO (spend-based)	Publicly available	Periodic	Screening only; lacks precision for final reporting

Choosing a database by deal situation

The comparison table above shows what each database is for. This section is about something more practical: which one to use first based on what is in the data room.

A working order of priority for private equity teams

Use the table above as your shortlist. Then use the deal situation to settle any close calls.

In most cases, the answer becomes obvious once you start with the data room. Begin with the national database for the country where the target operates. If the target is UK-based, DESNZ/DEFRA is the default starting point. If the asset is French, ADEME Base Empreinte is the reference source. Just check the version first, because ADEME’s 2025 monetary factors are about 40% lower than earlier values [4].

For multi-country electricity factors, use IEA. If emissions are driven by specific manufactured inputs, ecoinvent gives you process-level detail. If spend data is the only input you can trust, EXIOBASE works as a screening tool of last resort.

That’s why factor choice should follow the data, not the other way round.

The table below shows how that order shifts by deal type.

Database selection for services, manufacturing and early screening

When the business model is the main variable, these are the default starting points.

The matrix below turns source choice into a simple first-pass rule set.

Deal Situation	Primary Database	Refinement Layer	Notes
UK services (utilities, travel, waste)	DESNZ/DEFRA	IEA for Scope 2 if any non-UK sites	Verify Gross Calorific Value basis for gas [2]
French asset with agri-food or construction exposure	ADEME Base Empreinte	-	Use "Valide générique" factors; verify the 2025 version [4]
Global manufacturer with bill of materials complexity	ecoinvent	IEA for electricity by country	Stick to one system model throughout [7]
Pre-LOI screening with only general ledger data	EXIOBASE	ADEME monetary ratios for French spend	Flag as screening-only in the output [5]

Common library traps that distort results

Picking the right database is only half the battle. You also need to use the library properly. Even a good database can lead you off course if you pull the wrong row, mix units, lose decimal precision, or combine versions that don't belong together.

The tricky part is that the output can still look fine. That's why these mistakes slip through. The main trouble spots are row selection, units, precision, and version control.

Broad rows versus specific rows

In ecoinvent, selecting a generic "market for" entry instead of a specific activity can water down the hotspot signal and hide differences between technologies [1]. You see the same problem in DEFRA and ADEME when someone stops at a broad sector category and never digs deeper.

The fix is simple: use the most specific row that fits the activity. Broad categories smooth everything out, and that can hide what actually matters.

Rounding to zero and unit mismatches

Small factors can vanish if they are stored with too few decimal places. Once they round to zero, small but frequent activities drop out of the footprint altogether.

Unit errors are even nastier because they can blow results up by 1,000× in either direction. Common examples include:

kg versus tonnes
kWh versus MWh
passenger-km versus vehicle-km

Also check the gas basis. GCV and NCV are not interchangeable [2]. A quick sense-check on the size of every line item can save a lot of pain before those errors snowball.

Version drift across years and sources

Using the 2025 factor for 2023 activity data can understate past emissions [2]. That risk is not small. In the 2025 DESNZ release, the UK grid-average factor fell by about 15% compared with 2024, moving from 0.207 to 0.177 kg CO₂e/kWh [8].

Mixing versions creates a different kind of problem. If one part of a study uses one background database version and another part uses a later one, your comparisons stop being clean. Lock one database version for the full reporting period so the background data stays aligned [1].

These issues are easy to miss, so it's smart to treat them as a pre-calculation check rather than a last-minute fix. The table below shows the main traps and the control for each.

Trap	How It Appears	Likely Distortion	Control
Broad rows	Generic "Steel" used instead of "Steel, electric arc furnace"	Dilutes the hotspot signal; hides key technology differences	Always search for the most specific row first
Rounding to zero	Sub-unit factors stored with too few decimal places	Small, high-volume activities disappear from the footprint	Ensure factor libraries store sufficient decimal precision
Unit mismatch	Factor in kg applied to data in tonnes, or kWh vs. MWh	1,000× over- or under-statement	Normalise all units before calculation
Version drift	Mixing ecoinvent v3.8 and v3.12 in one study [1]	Inconsistent background data distorts comparisons	Lock a single database version for the reporting period [1]

Conclusion: match the source to the decision

The choice comes down to three things: where the asset operates, what data the deal team has, and how exact the output needs to be.

For most teams, the split is pretty simple. Use DESNZ/DEFRA for UK operational reporting, ADEME's Base Empreinte for French regulatory assessments (BEGES), EPA for US-based operations, and IEA for electricity grid factors across 150+ countries when local national data isn't available [4][5].

Then match the factor set to the data in hand. Use ecoinvent for process-level detail and EXIOBASE when spend data is all you have. Start with supplier-specific data where you can. If that's not on the table, move to process factors, then national averages, and then spend-based proxies for screening [6].

One last point matters more than it might seem: record the database version, system model, and GWP basis for every factor used. That simple paper trail is what makes the result defensible in committee and disclosure [4][7].