The matching agent

Drag and drop to get your scope 1, 2 and 3 emissions. High accuracy.

Drop the file you already have. The agent reads it, works out what every line actually is, matches it to a GHG-compliant factor and writes down why. The week that used to go into lookups is the week that disappears.

No template to fill in. No cleanup week before the work starts. Built on open emission factors.

96%of lines right as returned, on the open benchmarkread the benchmark
10,000+messy procurement lines we continuously test against
120k+procurement lines matched in early access
0-20%of a file left for you to review, and it tells you which lines

Line by line

The mapping agent in action

Procurement item

  • Description: Coffee
  • Quantity: 25
  • Unit: kg

Agent reasoning

Common food product purchased by weight. Match to coffee in its typical processed form (ground), cradle-to-shelf lifecycle.

Emissions

≈ 210 kg CO₂e

25 kg × 8.40 kg CO₂e/kg

  • Keyword: Coffee
  • Attribute: Ground, cradle to supermarket gate
  • Library: Agribalyse

What it does

Most of the work happens before the match

Five steps, in the order they run. The lines below come out of real files, and each is a different way of being hard.

  1. Your file

    Drop the file you already have

    Any export, any shape, any language, xlsx or csv.

    No template to fill in, no columns to rename, no pass to tidy it up first. Drag it onto the well and the run starts from whatever came out of your system.

    Your file

    Drop your file herexlsx, xls or csv, up to 10 MB
    • EU hosting
    • Never used to train language models
    • Confidentiality guaranteed
  2. Parsing

    No cleanup week before the work starts

    It reads the file you have, in the state it is in.

    Twelve header rows, four tabs, merged cells, subtotals sitting inside the data, three units in one column, French headers and Dutch descriptions. No template, no format. It also finds the rows that carry no emissions of their own: tax, rebates, intercompany, payroll, second-hand goods.

    Matching a tax row or a subtotal double counts it, and deleting them by hand is how a file loses a week.

    Parsing

    • Sheet1
    • Purchases 2025
    • Sheet3
    • Totaux
    1. A1:F12header block, merged cellskept for context
    2. 13BTW 21% · 262,50 EURdropped
    3. 14Subtotal Q1 · 14.300,00 EURdropped
    4. 15Second-hand desk chair · 120,00 EURalready counted when it was bought newno emissions
    5. 16Salaires mars 2025 · 48.200,00 EURwages buy no goods or servicesno emissions
    6. 17MRO-CONS-0047 · 3 PCresearched
  3. Research

    Stop asking around for clarification

    It looks things up before it matches.

    The company behind a supplier name, its sector, its brands, what a product code stands for, what the buyer actually does. Most of what used to be a round of e-mails, it finds itself, and it writes down where it found it.

    Without the lookup it becomes a generic plastic and a wrong number.

    Research

    1. ISOFIL H 40 C2 F NATunreadable
    2. Trade name, a filled polypropylene compoundersupplier site
    3. Talc-filled polypropylene, 40% fillerdatasheet
    4. Polypropylene, talc filledmatched
  4. Conversion

    Conversions, including the exotic ones

    It converts, then matches.

    Quantity, unit, currency, region, year and boundary all have to line up with the factor before the multiplication means anything. Metres to kilograms, DKK to EUR, litres of solution to grams of the metal in it, two station names to a distance.

    Bought in metres, priced in litres, booked between two station names. Miss the conversion on the gold and the line is out by three orders of magnitude. Every public library plus your own sits behind it in one vocabulary, so a comparison across them is a real comparison.

    Conversion

    1. Cable 3G2.5, 250 m

      3 cores × 2.5 mm²copper 8.96 g/cm³16.8 kg
    2. 5 L gold chloride solution, 2 g/L

      read the concentration10 g of gold
    3. Tog, Vamdrup - KBH

      resolve both stationsrail distance232 passenger-km
  5. Auditing

    You review what it flags

    A second agent checks the first.

    Four eyes on the lines that carry the emissions. It re-reads the hotspots against the GHG Protocol, disagrees where it disagrees, and raises what a reviewer would have raised. That is what turns a match into a high or a review.

    Auditing

    1. Travel invoice 04/2025 · 2,480 EUR

      one of the file's biggest lines
      first passRail transport, passengerhigh
      audit“The invoice is a return flight New Orleans - London - Brussels with one train leg on top. The spend is air, not rail.”
      resultRematched to air transporthigh
    2. Recycled steel, 3 t

      near-identical wording
      first passSteel, low alloyhigh
      audit“Close text, 2.4x lower in ADEME. The line says recycled.”
      resultSent back with both candidatesreview

The output

A workbook that defends itself

One row per line of the file you sent, with everything a reviewer will ask for already on it.

LineMatched factorConfidenceWhy
5 L SOLUTION CHLORURE D'OR 2 G/LGold, primary productionLibrary: ecoinvent 3.10, GLOhigh5 L at 2 g/L is 10 g of gold, not 5 L of it.
BAHNFAHRT HAMBURG - BERLINRail, passenger, 286 kmLibrary: DESNZ 2024, EUROPEhighNo distance on the ticket. Both stations resolved, 286 km computed.
RECYCLED STEEL, 3 TSteel, secondary routeLibrary: ADEME 23.6, FRhighRecycled, not primary. The words are nearly identical, the factor is not.
EXPRESS SHIPMENT 04/2025Air freight, long haulLibrary: GLEC 3.1, EUROPEreviewExpress means air, but weight and distance were inferred from the amount.
TRAVEL EXPENSES 04/2025Not enough in the line to matchno matchFlights, hotels, meals or mileage. Nothing in the file narrows it down.one question to ask

Download it as a workbook, or push it straight into your carbon inventory: the agent already runs inside Carbon+Alt+Delete.

Try it on your own file

Send the file you already have. We run it, and you get the matched lines back with the reasoning on every one.

Choose a filexlsx or csv, any shape. No template, no cleanup week.
  • EU hosting
  • Never used to train language models
  • Confidentiality guaranteed
What the file needs

Must have

  • One tab, one purchase per row
  • A header row, in any language
  • A column that says what was bought
  • An amount with its currency, or a quantity with its unit

Nice to have

  • Supplier name and country
  • Category, GL account or commodity code
  • Date or period of the purchase
  • Quantity and unit beside the amount