---
title: "Methodology"
description: "Emission factors, data sources, and calculation approach"
section: guides
order: 7
---

# Methodology

Sustifi calculates CO₂ equivalent (CO₂e) emissions using activity data multiplied by emission factors. This page documents the sources and approach.

## Emission Factors

Emission factors convert a unit of activity (e.g., 1 kWh of electricity) into a mass of CO₂ equivalent.

### Electricity

- **Finland (Helen invoice / XLSX report)**: Real Fingrid Open Data grid emission intensity (dataset 265, 3-minute resolution), averaged over the billing period. This replaces the flat annual average with an actual monthly average for that period. Fallback chain if Fingrid data is unavailable:
  1. Energy mix percentages from the Helen invoice (fossil / nuclear / renewable %), using per-source emission factors
  2. Finnish grid average 87 gCO₂/kWh (Motiva 2023)
- **Sweden, Norway, Denmark and other countries**: national grid averages from IEA / EEA (location-based)

### Transport

- **Road (car)**: UK Government (DEFRA) GHG Conversion Factors, per km by car size and fuel type (petrol, diesel, hybrid, electric)
- **Flights**: DEFRA per passenger-km factors for short, medium and long haul, with the high-altitude radiative forcing multiplier (1.9×) already included
- **Train**: DEFRA rail factors combined with Nordic statistics
- **Ferry**: DEFRA ferry factors (foot and car passenger)
- **Fuel you buy**: when a fuel receipt shows the litres, the litres are multiplied by DEFRA's kg CO₂e per litre for forecourt petrol or diesel. This counts as Scope 1 (fuel burned directly) only when the purchase is booked on an account for your own vehicles and fuel is at least 95 % of the receipt's amount; otherwise the purchase stays a Scope 3 spend-based estimate
- **Freight**: if you enter a shipment's weight and distance on a transaction, it is priced per tonne-km: GLEC Framework v3.2 factors for trucks and rail, DEFRA factors for vans and air freight (air including radiative forcing)

### Energy

- **District heating**: Finnish Energy / Energiateollisuus heat production mix
- **Natural gas**: UK DEFRA natural gas factor per kWh (net calorific value)
- **Rented premises**: if you describe the rooms you rent under **Premises** (floor area, ceiling height, heating), rent is counted as the energy the room uses instead of an estimate from the rent amount. The room's volume is multiplied by Motiva's median service-sector energy use per cubic metre (private offices for offices and studios, warehouses for storage), priced with kWh factors and reported as Scope 3 Category 8 (upstream leased assets). Each month of a lease is counted once, even when the rent comes on several invoices. Today this applies to rent rows from Lilith accounting; without premises, rent stays spend-based

### Food

- **Product-level factors**: Open Food Facts database, which includes Agribalyse LCA data (ADEME) for thousands of food products — covering production, processing, packaging, and transport

### Purchases & Services

- **Spend-based factors**: Exiobase MRIO database (EU-level supply chain emission intensities per € spent by sector)
- **Vendor's line of business**: for services, the vendor's main line of business in the Finnish trade register (PRH, TOL 2008 → NACE → EXIOBASE) narrows which spend factor applies. A product named in the transaction description wins over it, and it is not used for goods (a shop's line of business says nothing about what you bought)
- **Item-based factors**: Product-specific LCA data where available. If you know a product's carbon footprint (from the manufacturer or an EPD), you can enter it on the transaction. Footprints from the Boavizta database are only suggested, never applied without you saving them

AI suggestions only propose a category or product. Every suggestion waits for you to accept or reject it; nothing is applied automatically.

### AI Usage {#ai-usage}

AI model inference has a measurable energy cost. Sustifi tracks this for Anthropic and GitHub Copilot.

**Sources:** Lannelongue et al. 2021 "Green Algorithms"; Patterson et al. 2021 "Carbon and the Cloud"  
**Methodology:** inference energy per token × average data-centre PUE × global average grid carbon intensity  
**Overall uncertainty:** ±50%

#### Anthropic (Claude models)

Token counts are available directly from the Anthropic usage API and are used for precise calculation:

| Token type | Factor | Rationale |
|-----------|--------|-----------|
| Input tokens | 0.0000003 kg CO₂e | Baseline inference cost |
| Output tokens | 0.0000009 kg CO₂e | ~3× input — autoregressive decoding requires more passes per token |
| Cache creation | 0.00000038 kg CO₂e | Slightly above input — write overhead for KV cache |
| Cache read | 0.00000003 kg CO₂e | ~10× cheaper than input — memory read only, no forward pass |

#### GitHub Copilot (premium requests)

GitHub's usage export provides **request counts**, not token counts. Emissions are estimated using an assumed token budget per request:

- **Assumed input tokens:** 1,000 per request (system prompt + context + user query)
- **Assumed output tokens:** 500 per request (model response)
- **Factors applied:** same Anthropic per-token factors above (most Copilot premium requests use Claude models)

**Resulting factor:** 0.00075 kg CO₂e per premium request (= 1000 × 0.0000003 + 500 × 0.0000009)

**Uncertainty: ±100%.** The actual token count per request varies widely by use case — a one-line autocomplete and a multi-file refactor both count as one request. Treat GitHub CO₂e totals as order-of-magnitude estimates only.

Each figure records the factor it was calculated with (see [Updates](#updates)).

## CO₂ Equivalent

All emissions are expressed in CO₂e (carbon dioxide equivalent), combining CO₂, CH₄, N₂O, and other greenhouse gases weighted by their 100-year global warming potential (GWP100). Factors taken from a source dataset keep the GWP values that dataset publishes: UK DEFRA and EXIOBASE use IPCC AR5.

## Most Precise Figure Wins

Where several ways of calculating a transaction are available, Sustifi uses the most precise one — a measured quantity over a spend estimate — even when it gives a lower figure. Where a figure has to rely on assumptions, the assumptions are conservative, so estimates are more likely to be slightly high than too low. Each figure's method is recorded so you can see how it was derived.

## Precision Tiers {#precision-tiers}

Every emission figure shows how directly it was measured, best first:

| Tier | Meaning |
|------|---------|
| **Primary** | Measured data: a product footprint or a meter reading |
| **Activity** | A stated physical quantity (kg, km, kWh, tonne-km) × a cited factor |
| **Modelled** | A physical quantity estimated from benchmarks, e.g. rent energy from floor area |
| **Spend** | Euros × an industry-average spend factor (EXIOBASE) — the least precise tier |
| **Unknown** | No calculation method was recorded for this figure |

This follows the GHG Protocol's Scope 3 data-quality hierarchy. The tier is shown on each transaction (see [Transactions](/features/transactions)).

## Limitations

- Scope 3 upstream emissions are estimated using spend-based factors, which carry higher uncertainty than direct measurements
- Spend-based factors come from EXIOBASE 3.8.2 for Finland, reference year 2019, at basic prices (VAT excluded), adjusted to 2025 prices with Eurostat's consumer price index for Finland (HICP). Spend-based figures are the least precise tier and are marked as such on each transaction
- Electricity factors for Helen imports use a billing-period average of real Fingrid data, not hourly time-of-use factors
- Transport factors assume average vehicle/occupancy unless the user provides specific data

## Updates

- **Finnish electricity**: Fingrid data is read for each billing period
- **Other factors**: we check every source's release page twice a month and import a new edition after reviewing it. Most sources publish once a year (UK DEFRA around June)
- **Spend-based factors (EXIOBASE)**: updated when a new version is released
- Each figure uses the factor version valid on the date of the activity, and records which factor it used. Figures in a finalised VS Report are locked and do not change when factors are updated