Check, don't write

Until recently, turning a bill of materials into a carbon footprint, or reading twenty supplier EPDs, meant hours of copying numbers between PDFs and spreadsheets. An AI assistant can now do that drafting in minutes. The catch is that it can also produce a confident, tidy, wrong answer in minutes. This lesson is about the way of working that keeps the speed and removes the risk.

The shift: from writing to checking

Think about how a finance team works with a junior analyst. The analyst builds the cost model; the manager doesn’t rebuild it, but checks it: are the inputs from the right price list, do the totals add up, is anything missing, would the conclusion change if one assumption moved? The manager’s time goes into judgement, not typing.

An AI assistant with the right skill puts you in the manager’s seat. It drafts the analysis, and you check the analysis more than you write it. That only works if the draft is built to be checked. So every learnLCA skill follows the same evidence rules.

Explainer

Check, don’t write

1. Scene 1 of 6

   The old way: write it yourself

   One specialist. One spreadsheet. Every number typed by hand.

   Request→Weeks of work→One report

   More requests means a longer queue.

   The old way: a specialist writes the whole analysis by hand, one request at a time. When more requests arrive, the queue just gets longer.

2. Scene 2 of 6

   The assistant drafts

   BOM+Standard+Database=Draft analysis

   An AI assistant with learnLCA skills does the first draft.

   Methodological choices are confirmed with you, not decided silently.

   The new way: you give an AI assistant with learnLCA skills the bill of materials, the standard and a database, and it drafts the analysis. Choices such as the standard, functional unit and boundary are confirmed with you, not made silently.

3. Scene 3 of 6

   Every number cites its evidence

   Steel, 4 kg[BOM line 1]

   Steel: 7.942 kg CO₂-eq[dataset name and version]

   Paint: supplier value[EPD reg. no., page]

   Packaging weight[missing: flagged]

   In the draft, every number points to where it came from: a line of the bill of materials, a page of an EPD, or a named dataset and version. Anything missing is flagged as missing, never filled with a plausible guess.

4. Scene 4 of 6

   You check

   The checker’s list

   • Right standard and functional unit?
   • Every BOM line included?
   • Each factor matches its source?
   • Arithmetic tool-checked?
   • Flags resolved?

   The human checks rather than writes, working through a checker’s list: the right standard and functional unit, every bill of materials line included, each factor matching its cited source, the arithmetic checked, and every flag resolved.

5. Scene 5 of 6

   Scale without dropping quality

   Drafting is fast. Checking is where the judgement goes.

   Drafts for several drivers, EPDs, the Construction Products Regulation, CBAM and product carbon footprints, flow to one checker who applies the same evidence rules to each. Drafting becomes fast, and the human judgement goes into checking, so a team can take on more work without lowering quality.

6. Scene 6 of 6

   Check, don’t write

   The assistant writes the draft.

   The evidence travels with every number.

   You check it, and you sign it off.

   Summary: the assistant writes the draft, the evidence travels with every number, and a person who understands the method checks it and signs it off.

The five evidence rules

1. Every number cites its source. A BOM line, an EPD (name, programme, registration number, page or table), a dataset (name and version), or “user statement”. If a number has no source, it doesn’t belong in the output.

2. Arithmetic is done in a tool. The assistant uses its code or data analysis tool for sums, conversions and percentages, not mental arithmetic. If no tool is available, it shows every step and labels the result “not tool-checked”.

3. Missing data is flagged, never guessed. A missing emission factor is shown as missing. The assistant does not fill the gap with a “typical” value from memory, and it does not quietly top up the total.

4. Method choices are yours. The standard, the functional unit, the system boundary and the allocation approach are confirmed by you before calculating, and written at the top of the output. (LCA 101 lesson 4, goal and scope, explains each one.)

5. Every output ends with a checker’s list. The specific things a person must verify before the result is used: “check these three lines by hand”, “confirm the EPD on the programme operator’s register”, “line 7 uses a proxy”.

These rules are what make checking fast. When every number points to a page, checking it takes seconds. When the method is written at the top, you can see at a glance whether it answers your question.

Why this lets teams scale without losing quality

Without rules like these, AI help scales the wrong thing: more drafts, each needing a full re-do by an expert. With them, three things change.

• The checking is bounded. A reviewer works through a checker’s list and a few spot checks rather than rebuilding the analysis. (Lesson 6 covers how to spot-check at scale.)
• The procedure is the same every time. A skill writes down how your team does a task, so the tenth EPD is read the same way as the first, by whoever runs it.
• Gaps become visible. Because missing data is flagged rather than filled, you see exactly where to ask suppliers for more, instead of discovering it later.

What the human still owns

The assistant drafts. People remain responsible for:

• The question. What decision is this for, and who will read it? That sets everything else.
• The method. Which standard, which unit, which boundary. An assistant will follow the standard you name; if you name the wrong one, you’ll get a well-evidenced wrong answer.
• The sources. Whether a database, an EPD or a supplier figure is fit for the purpose.
• The sign-off. Anything that leaves the building (a tender answer, a customer figure, a declaration) carries a person’s name, not the assistant’s.
• The formal steps. Verification of an EPD, critical review of a comparative study, legal interpretation of CBAM or the CPR. Skills prepare for these; they don’t replace them.

A useful test

If you couldn’t defend a number in front of a verifier, a customer or an auditor by pointing at its source, don’t use it. The skills are built so that you always can.

Exercise

A colleague sends you an AI-drafted footprint for a product: “Total 42 kg CO2e. Steel is the hotspot at 70%.” No sources, no method, no list of missing data. Write down three questions you’d ask before using the number.

Answer (any three of these): Where does each emission factor come from, and which version? What is the unit and boundary (per product, cradle to gate?) and which standard was followed? Which BOM lines had no factor, and were they left out or estimated? Was the arithmetic done in a tool? Has anyone checked a few lines by hand?

What you should now be able to do

• Explain why checking an AI draft is faster and safer than rewriting it, when the draft follows evidence rules.
• Name the five evidence rules every learnLCA skill follows.
• Say which decisions stay with people: the question, the method, the sources, the sign-off and formal verification.
• Spot an AI output that can’t be checked, and ask for what’s missing.

Sources

• ISO 14044:2006, Environmental management: life cycle assessment, requirements and guidelines (for goal and scope, and reporting). https://www.iso.org/standard/38498.html
• Anthropic, Agent Skills overview (how skills load instructions on demand). https://platform.claude.com/docs/en/agents-and-tools/agent-skills/overview

Checked: 2026-09-29